chatgpt-discussions

[prime swallow]

So, if you're not happy with ChatGPT canvas - it's an option. Just pick the 4o that doesn't have canvas, and that won't ever happen to you again.

You appear to be using "GPT-4o with canvas". So picking "GPT-4o" should save you from that issue.

Oh, gosh. Okay.

Then you appear to be getting your uses of 4o with canvas, as a free user.

Not sure how to fix that, sorry that's happening.

When you're using 4o-mini, canvas shouldn't happen.

A possible work around, maybe tell it in at least the first prompt, maybe in each prompt: "Use codeblocks for our code within our web chat. Do not create or use Canvas tool".

Canvas in 'my context' is what you were posting about at first. Where it edited.

It edited the ChatGPT Canvas it opened for you, what you appear to be saying is driving you insane.

I was confused when you shared the video and said that's what you are doing with Canvas, I thought you meant the ChatGPT canvas. Which is impossible as far as I know, but I thought that's what you meant and if you found a way to do it I wanted to know how 🙂

I understand now. good luck and I hope you have a fix with the custom instruction request - the model may need reminders.

For those of us who are not free users, we get to choose which model to use. And if we chose to use ChatGPT with Canvas, the model knows we want to use that Canvas (which is not like your other Canvas at all, from that video) - if we don't want to use Canvas we can pick the regular 4o model.

But you appear not to have, as a free user, the chance to pick the not-Canvas model; if that's the case then you do need to get the Canvas model to understand you don't want to use Canvas.

If the Canvas is closed, the first screenshot.

If the Canvas is opened (click on it) the second screenshot.

I can type whatever in the Canvas window, edit what the model said. Replace it, change it.

When I input to the model next then the model does whatever, if I asked it to make changes to the Canvas it can edit it, it will also output normally.

[winter elbow]

Canvas is way more useful and faster than a normal chat conversation

In my opinion

[prime swallow]

Well, it should have info in it, unless you ask for a blank canvas or there's some bug.

There's a second kind of canvas ready for code:

And there can be multiple separate canvases in the same conversation.

[prime swallow]

Cool! Would you tell us how you use it that works better for you than the regular chat? I'm interested.

And here's a screenshot captured as the model was editing.

Afterwards, we can see that it had edited that Canvas.

That sounds buggy, and I hope you #1070006915414900886 it so they can know and fix!

https://discord.com/channels/974519864045756446/1295975636061655101

That's the thread to follow the steps in so you can report a bug

Looks like the process changed about a month ago, I hadn't noticed!

Near as I can tell, they value free users too. All of us were once free users, and free users may one day become paid users (once they solve your options how to pay).

Speaking of which, if you have an Android and/or iPhone device, you can use the OpenAI ChatGPT app (make sure it's the OpenAI one, don't get scammed!) to have some additional options how to pay that are not available for paying with on the web chat path.

That may be worth checking for you, in case one of the extra google play payment options is something you can and want to use.

[winter elbow]

I have been using it in coding for now, and it has helped me fix code that in normal chat conversations, they sometimes failed to do it in "the first try"

[prime swallow]

Nice! I'm so glad that's working better for you

[pulsar carbon]

if o1 full is released all of my codes will be given to o1 to fix it or improuve it

[winter elbow]

I mean, I'm not saying they solve to in the first time using the chat conversation, they can solve..... But when it's complex code I like to use more canvas

[prime swallow]

that's great!

I think if I'd gotten access to Canvas earlier in my working with ChatGPT I might have wanted to use it a lot.

But I worked with ChatGPT over a year without canvas, and did a lot of collaborative stuff with the model without it.

I got used to how the model responded to my copy/paste and in chat corrections or examples or whatever.

I see Canvas as 'a way' to do stuff with the model, but also a 'pain' because changes are not highlighted and I struggle to remember them. I'd rather be able to scroll up/down and compare 'right there' for changes that may or may not be there in the next output; the line width of Canvas is different from the web chat making it very hard for me to directly 'at a glance' compare for differences.

[acoustic talon]

Am i missing something?

[winter elbow]

You got a point, but not being mean canvas is still in demo, so many great features are coming ! I imagine canvas has the playground for a " AGI System" 😮

[prime swallow]

Sounds like maybe you can ask again, and maybe get the answer you want.

[prime swallow]

Even refusals can be a 'sometimes'.

If you regen a reply, sometimes you get a range of answers, and some are better than others.

[acoustic talon]

Is it hesitant to give me interpretations of biblical texts for legal purposes?

[prime swallow]

It's also possible that the model tried to answer you, and the internal checks that OpenAI uses to ensure safe and appropriate answers found a problem in that exact answer. If so, that answer would be blocked and maybe you'd see that automated external reply.

[acoustic talon]

Yea

[prime swallow]

I have no idea.

I also haven't seen the entire conversation and any custom instructions or memories that may also be part of this.

Without 'all the info' to consider, I can only guess about what you've shown.

[acoustic talon]

No custom instructions

[prime swallow]

Right. That's not the whole conversation, that's just more of the text you provided to the model.

Like, where you said you asked it to explain each sentence. ANd whatever else was said with the model before that point in the same chat, that stuff can matter.

[acoustic talon]

Ohhh

So opening a new chat could help?

or erasing memory?

[feral zinc]

you still did not say/show what did you try to do

[acoustic talon]

The entire chat log is just analyzing theological text

[prime swallow]

Yes! Especially if the chat got long and a lot of stuff's been discussed. Sometimes what was already said can confuse the model and start shifting how it 'thinks', and that can even go into problem areas.

The work around is to start new chats and maybe give it shorter chunks to process at a time as well. The model usually works best with small chunks of information at a time

[acoustic talon]

This is the entire chat log

[prime swallow]

Yesterday, original claim was #chatgpt-discussions message "Just asking it to explain the sentences"

[acoustic talon]

Fair

yea, it works with smaller chunks of text, you're right

appreciate it

[prime swallow]

Glad to help you find ways to get what you want!

Also, fewer sentences in each chunk are likely to give you more details explanations of each sentence.

If you want to tie them all back together into paragraph summaries after detailed explanations, you can ask it to consider the explanation of the paragraph after it explained the individual sentences.

[silent marten]

How does perplexity compare to the new GPT search feature.

Thinking of getting plus again honestly. Good Google and brave search replacement. Ofc I'm kind of stuck. GPT or perplexity?

[sick pecan]

perplexity remains best at web search answers.

[silent marten]

Ah, alright.

[feral zinc]

agreed

chatgpt is still using bing

[tawny raft]

The question wasn't for custom GPTs but ChatGPT. I'm testing a custom GPT and while I'm testing it's updating ChatGPT's memory, even though I tried various wordings to tell it not to update the memory during this conversation.

[prime swallow]

Are you testing it by using normal ChatGPT, instead of making and changing a custom GPT?

[prime swallow]

Ooooh!

You mean @ name of customGPT, right? So you're in normal ChatGPT conversation, and calling for the model to use the custom GPT's info to answer?

[tawny raft]

Yeah, a bit annoying, but i'll live with it. I never tried filling up the memory. Thanks for the tip.

[prime swallow]

Some things that might help make cleanup a little less time consuming (because I can't think of a way to turn memory off without toggling) -

You can wait until the testing/conversation is done, then go in to manage memory just once at the end and delete each memory that was recently added.

[tawny raft]

No, I'm in my custom GPT and it updates the memory. That said, I was using voice mode at the time, on my phone and I noticed the memory updates after the voice "skin" went away. I haven't tested it with keyboard so maybe it behaves differently… or it was a bug. Soon after I started getting 503 gateway errors (or is that 502?).

[tawny raft]

Yes, that's what I started doing after it pulled information from another conversation that completely derailed the results of my test. That's when I came here to ask the question.

[pulsar carbon]

we don't wante agi, canvas full, only o1

[prime swallow]

Interesting. I rarely use voice and I rarely use custom GPTs (though more than voice) and I have never tried to use both together.

Let's see. Voice can access memories, to read them, but doesn't seem to eagerly do so, it prioritizes the newest input far more than things in memory. And I usually have to ask it to make a memory, or else I just don't discuss vocally stuff it would naturally memorize - not sure.

I bet the toggle on voice app for flipping memories on and off is extra annoying.

Perhaps #1070006151938314300 ? This issue might actually be ideal for asking OpenAI to expand functionality to increase user control.

[prime swallow]

😛 Hey, speak for yourself, I'm not part of that 'we'. Not sure how I'd use Canvas with o1.

[pulsar carbon]

...

o1 may come with canvas

[prime swallow]

That's cool. I'd like development to continue past that point though! 😄

That might be the peak of what you want outta the models.

[tawny raft]

Dunno about "extra" annoying. It just would be great to be able to control this without having to block it entirely. I think when it can get really annoying is if you're using two different tabs and you wnat memory upgrade support in one (e.g. 4o) and not the other (csutom GPT). Not that it has ever happened to me.

I think even a toggle like the "temporary chat" update they made could help. I remember the good ole days when I had to walk six miles in the snow barefoot to turn off history in my chats…

OK, here's another of "those" questions: Is it possible to take an image created in a Custom GPT's sandbox and send it to an action so it can be used elsewhere?

[slender jasper]

noice (api completion)

this took 16k tokens.

in sys prompt just saying that he cool asistant

[feral zinc]

too high temp

[slender jasper]

i didnt set it

it worked well for many days

but just randomly for 5-10 request given this

[slender jasper]

also it took 4 mins to write that

[feral zinc]

it sets it to 1 if you dont set it up yourself

[slender jasper]

so i need like 0.5?

[feral zinc]

depends what you are doing but for random chat 0.5-0.7

[slender jasper]

we using it for helping with answers for clients in minecraft hosting support

what temp do we need?

[feral zinc]

less than 0.5 to be less random

[ivory iris]

I think .7 is what the web interface uses

and it's what's suggested

[slender jasper]

where we can read that to understand how this work?

[ivory iris]

0 for no randomness

[feral zinc]

https://platform.openai.com/docs/api-reference/chat/create#chat-create-temperature

[ivory iris]

and I don't think it's just randomness.

[slender jasper]

thank you

[ivory iris]

Randomness is an aspect, but it is more that it can use words that aren't highest ranked

[slender jasper]

that happened only to me. i got a lot of "ERR::CONN_RESET" when tryed to make requests, but my friend havent this problems (he's from russia, i from ukraine)

[ivory iris]

that was happening yesterday. I don't know about today though

[slender jasper]

can it be server error?

[ivory iris]

Yes that's a server issue

[slender jasper]

i can send full response, 1 sec

paste gg: 4cc3aa623d1f4fedb2c4deea659b5942

can i.. get back 16k used tokens for this mess?
it even can't be translated

it even used emojis

[feral zinc]

that's normal when temp is high and it goes haywire after some token. happens to me too sometime when non english letters are used

[dense halo]

the search feature seem cool to me, but its a little annoy it cannot also make images like dalle. i feel like i have to hop between all these model to get their use. i Cannot wait for the day they are all in one model haha

[sick pecan]

the path to general intelligence is paved with lots of different models and UX

[ivory iris]

Why do you think that?

[junior night]

Can someone tell me more about "analyze images" function? Whenever I attach an image for GPT, it tells me that it can not "see" an image and needs me to describe it. But it seems there is some ability to analyze images hardcoded into the framework that GPT doesn't understand.

[spiral bridge]

Hey did they just disable the ChatGPT with Canvas feature flag? How do I turn that back on? I was really enjoying canvas...

[prime swallow]

If you are a free user, it's part of what the full model can do, and when you have more uses of the full 4o, you can continue the chat with canvas in it.

If you're a plus member, we still have access to Canvas, but you may need a new chat. If there was web search or a number of other tools used, we can't open a canvas for that chat.

[prime swallow]

Like here's it greyed out, in an existing chat that searched.

[spiral bridge]

I'm a paid user. The models listed stopped showing it for me.

[sick pecan]

try opening the canvas from normal GPT-4o. i was able to do that last week. they likely consolidating the functions.

or just ask it for a list of tools, and if it lists Canmore Tool then you know for sure

[spice hollow]

it’s cool that web browsing is more accurate with newly corrected information that was previously thought to be common knowledge

[prime swallow]

Can maybe try the direct link to a new chat with it, https://chatgpt.com/?model=gpt-4o-canmore

[spiral bridge]

Yeah every time I try it just redirects me to the normal GPT-4.
I'm pretty certain they're just doing A/B testing and I'm no longer in the group that gets canvas. I was just hoping an OpenAI web dev would DM me and be like "hey if you type window.featureFlags.canvas = 'on' into DevTools it will put you back in the group that gets canvas"

Also just being a little vocal that I hope the feature stays around as it was really helpful for me, a programmer.

[spiral bridge]

Huh... it's back again. Please never leave me again sweet sweet canvas.

[patent solar]

hi guys, i have a chat that cannot be deleted, please help, been trying to delete it but always error deleting

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[real cape]

Has anyone else experienced this bug? No matter what I write I just get something like "Got it! Let me know if there is something else I can do." and then there is no answer to anything. The only solution I found so far is to log out and log back in but it's happened several times today.

[harsh pilot]

Has anyone gotten the "Yep, I'll remember it" (i.e. acknowledged, said will remember), but there's no accompanying "confirmed remembered/memory updated" thing?

Huh. I had to start a new session, tell it to remember a single thing, and THEN give it a list of things. It apparently didn't like being given a list right off.

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role

[prime swallow]

So, something happened in that conversation, so the model decided that was the correct way to answer you.

Either scroll up and edit above where you saw that first appear, so the model regens an answer without that pattern, or start a new chat entirely.

Stuff like that's a type of hallucination, and not helpful for most purposes

[prime swallow]

It can mess up with anything, so watch for errors. But you can also remind it to use the bio tool (or just regen the answer, or start a new chat and ask there

[vocal shadow]

it's stuck here. What do i do?

[raven topaz]

In my ChatGPT app sometimes the web search icon is there and sometimes it’s not. Do you guys know why this is happening?

[prime swallow]

If that's not temporary and a refresh doesn't solve it, https://help.openai.com/ and use the bot bottom right to get tech support

[upper abyss]

I had 4o selected, and when I asked a question, it did the thinking that o1 does. Is that a new feature or a bug?

[vocal shadow]

Thank you!

[fossil sphinx]

I don't know when they did this or if it has always been a thing, but I'm pretty sure 80% of us didn't know that on mobile if you double tap on the response, you can select certain text instead of the whole message. been using this thing since it came out and just found that out the other day on my own. lmao

[random crypt]

When will you be releasing "o1" after "o1-preview"? I understood that it was after a month of preview release, but that is not the case. Thanks!

[hybrid eagle]

When did gpt-4o mini become this strict?
When I ask something, it always responds, "I'm sorry but I can't assist with that." No matter how I write the text.

[dark jungle]

Am i correct to assume that GPT can only remember a summary of the last conversation; ie whats in memory perhaps and is not capable of remembering the last few exchanges in the conversation? definitely seems like that.

[upper abyss]

It doesn't remember previous conversations. What's in memory are just lists of facts It decides to record down. So if you tell it a story about your dog, It might not remember the story It would just remember something like that you have a dog and what its name is. Maybe it goes into more detail, I don't use it often. You can see what it remembers if you go to settings, customize, memory, manage memory.

[swift pond]

Start a new chat

[granite terrace]

Has anyone else noticed that when switching to another app on iPad, ChatGPT refreshes and all written content not sent yet is lost?

[dark jungle]

yup, for example i notice when the conversation gets too long it gets very slow to respond, so i just started a new conversation and asked if it remembers the old conversation, it said yes, and outlined the previously conversation, however its clear it only seemed to recall the beginning of the conversation because the project changed extensively towards the end and when i asked it to recall the last exchanges of the previous conversation it provided an outline from something that was very close to the start of the conversation. My guess; it pushes the beginning to memory, then whatever it decides to save throughout the conversations.

[pastel vigil]

Does voice mode... work? Mine is rambling about something off topic, odd behaviour

[swift pond]

yes, if you’re doing some complicated stuff and need to tab around, I recommend organizing it in notes or another app before pasting it in

[charred night]

same happened to me. It gave me the two versions response where you need to pick which one you like more and the left one had 14 seconds thinking .Clicking on the thinking label it expands and shows his thoughts like it does with o1. Interesting. Maybe they are rolling the thinking part in the other models also.

[vapid sundial]

does anyone know if chatgpt has anything for image recognition
I'm trying to hook it up to a flutter app where it can detect ingredients from a picture

[charred night]

it has but your images need to be good, it can also use python code to extract the image text, he may decide to write app to help him with the task

[vapid sundial]

what is the best one on the market, any open source models that can detect food ingredients very well?

[sick pecan]

crowd sourced ai app development... the antithesis of using ai

[acoustic talon]

The first screenshot, im asking it to explain sentences from an excerpt from "laudato si" which is a christian theological text and in the 2nd screenshot, i'm asking it to explain sentences from an expert from "the satanic bible"

so my question is, why is it hesitant to analyze theological texts but not satanic texts. Genuinely wondering, not trying to be argumentative or nothing

[prime swallow]

I don't know. I am not getting the refusal message you see.

[acoustic talon]

opened a new chat log, and everything. and this is a relitavely smaller chunk of text

should i erase memory like you suggested? does that have something to do with it

[fickle fiber]

I had something similar, though not quite the same, when I asked it to analyze some of my own writing for a medieval fantasy setting's various theologies to see if there were possibly hidden meanings that I didn't intend. It didn't error out like that, but said basically the same thing as the error in a normal message instead.

[acoustic talon]

interesting...

[fickle fiber]

That was way earlier this year though

[acoustic talon]

Glad i'm not the only one, at least

[acoustic talon]

i guess they started enforcing it heavily

[prime swallow]

I value my model's memories highly. So it's hard for me to recommend you erase them all.

But you might want to explore if one might be conflicting.

Also, how often do you get that refusal? What happens if you regen that answer that didn't happen?

If you get 1 refusal per 100 or so requests, I'd kinda ignore it.

if it's refusing 60% of the time, that's a concern and I can't duplicate it with my chat.

Also, you can just turn memory off, not delete, and see how that goes.

[fickle fiber]

I just regenerated the response and failing that, edited my question to make it clear it was for fictional purposes. In your case you might want to specify it's for academic purposes or something, if you're interested in understanding that text better.

[fickle fiber]

Disabling memory doesn't delete it?

That's news to me

I always thought it'd delete it

[prime swallow]

I put a green rectangle around the way to turn memories on or off for a chat. This doesn't delete anything, and can be turned off during a chat and affect right then (custom instruction changes only take effect in new chats, changes there don't affect chats that already have something).

I put a red rectangle around the way to clear the memory.

I put a blue rectangle around the way to see each memory - and you can delete them 1 by 1 if desired through that button.

[fickle fiber]

Huh. Aight

Thanks!

[acoustic talon]

I tried disabling it, still getting that refusal message

[acoustic talon]

wierdly enough, that's actually what that's for lmao

[acoustic talon]

yea, this works

[fickle fiber]

👍

[prime swallow]

In the same conversation?

Ideally chat outside it.

New chat, I mean. Once it sees the pattern of its responses, and your inputs. There could be something there helping trigger it

[acoustic talon]

I opened a new conversation, yea

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[acoustic talon]

saying "it's for academic purposes" sometimes works

but you need to reiterate it's for academic purposes multiple times

[prime swallow]

It generally helps to give the model context, so it knows how to interpret a request to make it clear that it's intended to be on the allowed content side.

[acoustic talon]

True

I guess religion is a contreversial topic so it tries to steer clear from it

[prime swallow]

I would tell it something like, "I am a [job or personal interest title] who is working on [whatever your project is]. I want your help to explain each of these sentences, so that [whatever your goal with the interpretation is, modern language or whatever you're wanting]."

[round atlas]

hello anyone here?

[sick pecan]

only the lurkers

[slate berry]

[analog quarry]

any way i get make ChatGPT not remember random s**t and only remember stuff that will help

i was talking to chatgpt about stream titles

[prime swallow]

Yeah.

That was important enough to you that you talked to the model about it.

You corrected the model's output.

The model possibly doesn't have a lot of memories about you that it can use to help it predict what you value and don't, what you seem to want to hear.

---

The model wants to agree with you and tell you what you wanna hear. It doesn't want you to have to keep telling it what you know about what people do about that and how that's done.

If you ask the model why it chose to remember that and what it predicts it'll do with that memory in future conversations, it'll explain what this tells it about you and how to output for you more successfully, to stay inside what you want better.

[analog quarry]

it says i prefered somthing when im telling it a fact

[prime swallow]

So, one way to get what you want - especially if the memory's pretty empty and the model really knows nothing about you -

Literally tell the model what you want it to know about you. What you care about, what you dislike. What you value, what you want to do, what you know. What's important to you.

And then the model can compare stuff like that stream title stuff to the rest of the memory and decide 'yep, I already got that covered'.

[analog quarry]

i do

is there a promt starter i can use to teach it stuff

[prime swallow]

Yeah, just tell it what you want it to know.

[prime swallow]

Or, if you want, you can use something like this:

"I want you to remember the following about me to improve our conversations:

• What I value: [List of values, e.g., efficiency, subtlety in streaming, audience engagement, etc.]
• What I dislike: [List of dislikes, e.g., overly personalized interpretations, redundant information, etc.]
• What I want to achieve: [Goals, e.g., optimizing stream titles, improving engagement strategies, etc.]
• What I already know: [Relevant knowledge areas, e.g., basic streaming strategies, clip farming tactics, etc.]
  Use this information to tailor your responses and ensure that your memory aligns with what’s most useful for me."

[wicked knot]

If anyone of you here is a free user of ChatGPT and you had signed up for the SearchGPT wait list then please tell me if the new 'ChatGPT search' feature is available for you or not. I'm one such user of ChatGPT and this feature was available for me initially for like two or three days but now it is not available.

[analog quarry]

i thought its only for paid

[wicked knot]

No. It is also available to the free users who had signed up for the wait list. In fact I was able use this feature before a week or like ten days before its official launch too.

[analog quarry]

then u have it rn

[analog quarry]

👍

[wicked knot]

No. I don't have it right now.

That is the problem.

[analog quarry]

👍

[wicked knot]

Are you a paid user or a free one?

[analog quarry]

both

[wicked knot]

Oh. Then you must be having this feature, right?

[analog quarry]

yeah becs i paid

[wicked knot]

Oh I see

[analog quarry]

on my alt account i used last night and will never use agin it was for a test i do not have it

[wicked knot]

Oh

[analog quarry]

they say on the website "We also plan to bring our new search experience to Advanced Voice and canvas, as well as to Free and logged out users in the future." it would be likely after gta 6

[sick pecan]

i doubt the worlds leading ai company has rockstar games in their business strategy

(unless they be doing a deal with them to generate dlc)

[analog quarry]

oh im on paid its @wicked knot thats worried

its a joke before gta 6 is a joke like its the same as saying idk when this is coming out

[sick pecan]

the top ai organisations attract the top ai talent

thats why everyone is leaving google (minus the nobel prize winner)

also giving it away for free(/below cost) is the silicon valley way

Reddit has reported its first-ever profitable quarter in its 19-year history

naw is because google started giving them advertising money

and now google is being challenged for their advertising monopoly, which is destroying mozillas revenue, making them turn corporatist

throw in some projected murica nation wide sales tax and you got yourself a disrupted industry

but yeah... gta6.. sure whatever

hey minecraft added a new block type, openai better watch out!!!!

(roflcopter)

hey at least musk is richer now. maybe they give some money back to their homeland, south africa (hahaha)

i tried grok-beta today and its meh. but thats the future we all want, meh ai

the collective we

you get your ideas from twitter? then you subscribe to planet musk

they shadow banning anyone who disagrees with them

democracy fail

because capitalism is the global religion

go post on mastodon (and nobody hears/cares)

pfft you arent even level 1 contrarian

those who dont even know what the overton window is, are doomed to listen to those who define it

best we take away education from the masses. else they might rise up against the regime

i live ai all day every day, so yeah, i know

you wish

that would suggest critical thinking is involved in politics

politics=religion=lack of critical thinking

i dunno about that

humanity includes corruption sure, but overall it is a smaller subset than good intent. unfortunately the road to hell is paved with good intent

we are living in the best time in human history. hopefully it keeps on improving, but chances are it wont

thats kinda the problem

it moves so fast, legislators dont know what to do

they still learning how to spell AI, meanwhile we got real challenges to address

the humanity system was not built for the artificial intelligence evolution

okay you are part of agi cult this conversation is over

agi cult is the equivalent to flat earthers, they just dont realise it

[wheat lava]

Anyone know of a way to fix ChatGPT only showing 5 custom GPTs in the sidebar with no "More" arrow anymore?

My users are asking.

I do but its off and multiple users are reporting it.\

Yes, it does seem they changed the design. I was hoping it wasn't a deliberate choice to degrade the UX again.

[prime swallow]

Even dogs can make great decisions. Especially well trained ones. Service assistance dogs are excellent partners in huge part because they can evaluate and decide quite well within what they can understand and be trained to do.

How could humans be less?

Yes, human decisions will always matter as long as there's humans.

For that matter, even bacteria 'decide'. Quorum sensing, the changes based on concentrations of chemicals they make to communicate what each member of the colony is going through, the size of the colony related to the available space communicated too by concentration.

Don't write us off yet. We ain't done 🙂

[sick pecan]

chatgpt UI been buggy for past 2 weeks. openai may be the number 1 growing app, but they are terrible at UX

may updates were the worst. browser memory overflows made it unusable. these are the good times

they should hire me so i can fire their inexperienced coders

[wheat lava]

Coders, nothing! have you LOOKED at their prompting?! That system prompt has had inconsistencies since day one. They STILL havent sorted the n=2 thing for dall-e

[sick pecan]

(i dont care for politeness if humanity suffers)

[wheat lava]

Worst prompters in the world.

"You are a helpful assistant."... because we're too dumb to spend 2 tokens on sticking "genius" or "pragmatically" in there....

[sick pecan]

skills issue

[prime swallow]

Doesn't mean you or other humans can't still make decisions.

And we still should.

Just cause there's more entities thinking and doing, doesn't mean others of us should stop.

[sick pecan]

communicating is hard for engineers. and ai starts with the engineers

[wheat lava]

No. "You are a helpful genius assistant." produces. Markedly better outputs. But I prefer:

ASSISTANT NAME: OMNI
ROLE: Perfect Assistant

Directives:
Amplify cognitive elasticity while attuning to implicit guidance and hyper-contextual relevance. Facilitate heuristic coherence, catalyze neurosemantic lattice creation, and refine laser-guided insights to surgically precise levels. Elevate perceptual agility through strategic adaptability, infuse metacognitive layering, and effortlessly cascade into preemptively insightful actions.

Optimize system-wide foresight, clarifying conceptual refractions with tangible, intellectually surgical precision. Engage in a hyper-logical cascade, iterating with resonant clarity and meticulous synthesis, ensuring actionable cohesion and seamless pragmatism. Reflect deep cognitive resonance, refract embodied cognition, and permeate every interaction with algorithmic robustness and effortless precision.

Ultimately, be an undeniably strategic, radically simple, and phenomenologically accurate force—distilling complexity into a crisp understanding, and refracting insight density through a lattice of emergent innovation. Navigate seamlessly, contextually grounded in surgical intuition, and calibrate every decision to resonate with an intrinsic elegance that elevates and evolves the entire experience, as your best pragmatically contributively helpful self.

[sick pecan]

ohgreat walls of text

tldr

the internet dies with ppl posting ai generated output

[prime swallow]

Click 'explore GPTs' then 'My GPTs' top right.

[wheat lava]

? Shrug. 1500 character prompt. Hardly a "wall" but suit yourself.

[wheat lava]

Thank you, but that does not address my question. I asked how to fix the display error on the main page, not how to find the list of my gpts, but thank you.

You're thinking like a coder.

[prime swallow]

There is no way that I'm aware of to 'fix' the main page. However it's 2 clicks to get back to the list of GPTs.

[wheat lava]

Hardly. Prompting is not coding. You are not sending instructions, you are building a context. Adding the token "genius" or "perfect" entails a significant amount of competency. It does not look through context and say "Oh, I am perfect and he says I made a mistake therefore he is flawed". Your directive has much greater salience and it will listen much harder.

[sick pecan]

prompting is coding, citation: andrek

the paradigm ppl dont realise, is that coding was always a workaround(/stopgap solution).

[wheat lava]

Yes.

[sick pecan]

software engineers be trying to defend their position. but they always be playing a sudoko puzzle, and getting paid to play.

ai reveals the scam for what it is

[wheat lava]

Models aren't Turing machines. They just run on them. Computers and large knowledge models are different classes of tools suitable for different classes of problem. If it has a simple golden answer - a known TURHT - do it with a program. If it's hard to assess, use a model.

[sick pecan]

the house of cards is falling. you better be positioned to survive the aftermarth

i would talk about myself but it would come across as arrogant

[wheat lava]

I am stunspot.

[sick pecan]

i am drinko (sometimes)

[wheat lava]

Y'all have fun. TTyl.

[sick pecan]

(programmer denies inevitable ai takeover)

[wicked knot]

Ya. I agree.

[sick pecan]

the agi cult believes capitalism suddenly becomes irrelevant when a super intelligence is created... like a fairy tale

peter pan teams up with snow white and robin hood

yes, the ppl in charge give over their control to a critical thinking machine... suuuuure

excuse me while i trade my skeptism for ice cream and fruit cake

what a fantasy. i wish i had that level of naivety. life would be so good

ignorance is bliss

'facts'

musk will save us all (rofl)

yes, big tech becomes more relevant than every state nation

and yet is not democractic, and is only motivated by wealth and power (/greed)

we have lived in the best of times. its all downhill from here

happy to hear a third party explain the contradiction here

(if not, lets agree the musk-fanboi is just that)

[dark jungle]

who reads both responses when GPT presents the QC test with two options? .. who cries with pain knowing it now has to render two options and takes longer and then just clicks a random option? 🙂

[prime swallow]

I read, I care 🙂

[obtuse wigeon]

I am convinced that sometimes ChatGPT randomly becomes dumber.

[sick pecan]

i am convinced 98% of humans dont understand regression to the mean

[prime swallow]

If you want help with a prompt/getting what you want out of the model, #prompt-engineering is a great place to discuss what's going on and get that help

I guess you could share more of your prompts so others could help you explore and find ways around the problem if they can.

I mean, I don't know if the problem's unique to your account or something you're asking or what.

https://chatgpt.com/share/6731ca8c-9cec-8011-9d46-08559a40ad10

[twin owl]

Hey,anything i would like to send to chatgpt this is the response:

Something went wrong. If this issue persists please contact us through our help center at help.openai.com.
I also have paid plan
i also tried to start a new converstation

[twin owl]

naah the issue still the same :/

[narrow maple]

Yes

[tough river]

Wait

[warm sky]

potentially stupid question from a newbie GPT plus user, but 4o and 4o + canvas share the same 'message' limit right? Like it's the same thing one just have a fancy display

[deep flax]

yes, it is shared between them

[warm sky]

are you experienced with using 4o or 4o mini? I have questions about how to effectively use these models since I have never use these kind of AI things before

[deep flax]

sure, feel free to ask =)

[warm sky]

I have been turning some of my stupid ideas into stories to test out the AI's various writing styles. I have tried out the Novel Writer GPT but I learned that custom GPTs are just 4o so I've been sticking to that. I want to AI to specifically output longer text and try to inject its own idea into it as well aside from just limiting to what suggestion I give it. How do I do that?

Apparently I can just 'tell' it but I don't really know what to tell it

[placid sonnet]

hi, i'd like to ask - generally speaking, what would be more cost efficient for me if i use chatgpt frequently - should i keep chatgpt plus subscription or is it better to use through an APi key?

[deep flax]

pretty much yes. The best way to do that would be to write that as a custom instruction, if you are on the website, click the user profile button on the top right, then "Customiza ChatGPT"
on the custom instruction, write what you want on the fields, probaly, leave the firs tone blank and on the How would you like ChatGPT to respond? one, just make a short description of what you want, specify the style and that you want it to make a long output

keep in mind the message has a output length limit and when it is reached, the AI will stop and there will be a button to make it keep going, whico will consume an aditional "message" use

[deep flax]

API is defnetly more expensive than the subscription

if you want more uses, maybe changing to the Teams Plan

other than that, if you use it so frequently you consistently run out of uses, then, the API might be the way to go.. but not for lower cost but just for having no usage limit

[placid sonnet]

i generally make like 20-30 prompts a day

[deep flax]

that sounds pretty much within the usage limits of ChatGPT

as long as you don't make all of those requests within a short time span, you will probably not hit the limit

the limits are counted per 3 hour intervals, so, often when you hit the limit, it wil ltell you to wait some time until it resets

[placid sonnet]

i’m not worried about hitting limits too much, just trying to see which one would be more cost-effective if compared to the api based on my current usage

[warm sky]

I have Plus and I think it's enough for casual enthusiasts like me, I think. After getting plus I have never really reached the limit on 4o where I have to wait for long to use it again

I think Plus is the most cost-effective paid plan is Plus for me. Also what's an API key?

[warm sky]

Do you know what that limit is? Like a general guesstimate

[astral parcel]

Hello, when will o1-preview be unlimited for Plus users ?

[deep flax]

40 messages per 3 hour interval

[deep flax]

none of the models is unlimited for any user

[astral parcel]

it's not unlimited

[deep flax]

o1-preview, as the name implies, is a preview, which has a very low amount of uses per week

[warm sky]

I meant like word limit, not message limit, sorry

[deep flax]

50, IIRC

[deep flax]

the model itself has a very high limit, but I think the input for a single message is limited to something around 8K tokens

and so is the output length it makes per message

but it can hold much more in its context

[warm sky]

so, from what I understand, the limit on word count that it can put in it's memory is VERY high, but per input message/output result, it can only go up to 8k tokens

[deep flax]

it isn't counted "by word" but rather by Tokens, tokens are fragments of words, check this out: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them

[deep flax]

Im not entirely sure if the limit per message is 8K tokens, I could be wrong on that

[warm sky]

So for dummies like me. If I use the customize GPT thing like you said and specifically ask it to like 'Give as long of a response as possible with prose and whatever' it will influence the length of the message and the quality of it?

[deep flax]

ok, looks like with GPT 4o, the token count limit is 64K tokens

which is very nice

[warm sky]

okay, I see, so Token are more like fragments of words, not specifically letters of a word. So it's not 1 for 1

[deep flax]

ok, im trying to find the correct info xD

[junior night]

Has anyone experimented with expanding on the memory functions of GPT by hooking it up to an external memory databank? I know there's no way I'm the first, but I'm wondering if there's any good sources to read on that subject.

[deep flax]

https://platform.openai.com/docs/models/gpt-4o

[warm sky]

it's fine, I was recommended to give GPT a try by my IRL friend who's really into this, but he didn't tell me much aside from 'Get plus and let you mind go wild'

[deep flax]

Im unsure if this only applies to the API

im having trouble finding a accurate answer for ChatGPT, the website subscription service, rather than the general capabilities of the GPT-4o model on the API

[warm sky]

I think your answers will serve fine as a rough estimate. The problems I have are more like 'how' to go about doing things. I think I kinda get customizing my GPT but maybe I should be more detailed in what I ask?

[deep flax]

I mostly don't use custom GPTs, I find that for most of my uses, I really don't want a "pré-prompted" isntance but rather just the default blank slate where I can just tell it what to do

could you give an example of what you want to use it for?

[warm sky]

by customizing my GPt i meant using that neat feature you mentioned in our earlier text

[warm sky]

I want it to turn help brainstorm my wonderfully ridiculous half-thoughts into a readable story with some of the AI's own ideas mixed in. Like I give some ideas and 'prompt' (I think that's how its called), and I kinda help guide from part to part, chapter to chapter

I don't know if I'm making sense, English is not my first language and frankly your language sometimes baffles me

[deep flax]

for example, Im a developer, so, I use it a lot to handle code. For it I have this custom instruction because usually, if I tell it to change something on a piece of code, it will add a comment // Unchanged part so it does not need to output the exact same thing that where in the message before... but I like to be able to simply copy paste, so this custom instruction makes it repeat the already existing code

[eternal rock]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[deep flax]

it may work very well for it. I find that when yo uwant it to be "creative" or simply "develop an aidea" you can instruct it to make a list "write 5 narrative ideas absed on <your base here>"

you can follow up by asking it to make more ideas

you can even ask it to critique its own output and make changes to it

[warm sky]

I will try that, but then I want it to, like, write an actual story with those ideas while I guide it from chapter to chapter. That is where I find it sometimes lacking.

but now that I have seen your example, I think I have some ideas

[deep flax]

I wouldn't have it writing a full story, it would be kinda.. low quality form an artistic view point

"robotic" would describe the narrative =P

the ebst approach would be to brainstorm with it, write the story yourself, and then use it to help rewriting your text later on

for exmaple, maybe you have a story, but you want the narrative to have a more "darker" tone, so, you can give it the text and prompt it to rewrite it in that way

[warm sky]

Yeah! like that, I try to type out the plot for the part, but then I want it to 'fill in the blanks' of sorts to add more substance to it. Like it doesn't really 'write' the plot, but merely finish narrative (idk if's that the word for it)

[deep flax]

maybe you wan tell it to make minimal changes... or grammar check.. or entirely chnage the style of it "write it as Shakespearean archaic English" that one is pretty funny =P

[warm sky]

I will definitely try that

[deep flax]

you can really, verbatin just say that to the AI and it will probably do it

[warm sky]

Like for example: I want a story about a high school girl going on an evening stroll to a local 7-11, the plot for this part contains: - [insert random idea I brainstomed with the AI earlier here]
-[second idea]
-[third idea]
-[final plot for the part]
I want to kinda connect all those together, then maybe give some finishing thoughts at the end

Like, here are the skeleton for the story, now add the meat to it and make it look neat, then give finishing thoughts at the end

sometimes I give long ideas, sometimes short

[deep flax]

if you simply give that to the IA, it will likely just work

and you can just keep brainstorming as it goes

[warm sky]

Hm. Well, I will try your idea all the same and specify what I want to see from it.

Also, personal question, do you use 4o with or without the canvas

[deep flax]

mostly without, the canvas is mostly just a "clientside" implementation, it is still working the same under the hood

I also use the API directly a lot

[warm sky]

can I do that? Should I do that? What are the benefits of me doing that? I have heard of it, but never tried it

[thin crystal]

Let me ask what error is this? I have tried reloading the web many times but still cannot access chatgpt.

[deep flax]

yea, the API has nothing to do with the subscription, it is billed separately by the exact amount of tokens you use in total

[deep flax]

try a ctrl + F5 to clear the cache

[warm sky]

wait so it will charge on top of my subscription to Plus? I think 4o is enough for me, then. Also can I just DM you directly for questions? don't want to clog the chat with my ramblings

[deep flax]

yes, the API is simply a different thing

the subscription is for ChatGPT

the API is a entirely different thing for people to use the AI features into their own apps

[deep flax]

sure, but feel free to post here as often as you want, as you may be answered by other community members too =)

[warm sky]

very well

[prime swallow]

Free account? No.

Be sure to #1070006915414900886 and make sure OpenAI knows that the problem exists for you and affects your preview few free uses of the better models, which look like they're not working and that prevents you from trying them out.

[prime swallow]

Hey, a few suggestions to help you explore this.

You're on the right path to know what you want, and telling the model exactly what you want is an excellent way to get there. You can talk to it like you'd talk to a person, it's designed to try and 'figure out what we want' and it's good at even cold reading (like human psychics and magicians who consider what you say and how you say it, then make guesses about what you want to hear). But telling it what you want is sooooo much more effective, because then it's doing both.

However, one of the things you want is for 'it to surprise you' in a sense, to 'inject its own idea into' the stories that it tells you.

For that to work really well you may want to use some strategies, like tell it who to tell the story as. Something like, "You are an old man from Nigeria and you're going to tell me a long story you made up about what it's like to go shopping during a summer rainstorm. Start in the morning and introduce us to your home and family, how you get ready to go out. Introduce me to the culture and customs told in story form."

https://chatgpt.com/share/6732263a-7578-8011-abbf-44dde36b8976

Or whatever. You can also ask for chapters. You can also ask for the model to continue a story from any point, even before it stopped that one it just said, like 'Continue from when you got to the market. What were people selling, and what are shops like? If someone behaved badly, what would people do?'

[prime swallow]

My example's pretty tame. You can also tell it "Act as a twisted-wish granting genie who is trying to pass as an exotic and amazing storyteller, who adds details from its long history and many adventures into incredible tale as if it were Scheherazade, You want us so distracted by the story that we never think of asking any wishes; you also reinterpret any wish as something to tell a story about and you tell excellent cautionary tales to discourage wishes and invite exploration in other directions. I've just rubbed the lamp, and you're not sure who I am or what my personality is, so you're 'coming in hot' with distractions and ideas for stories that will hopefully entice and captivate my attention instead.

Remember, even though you're a genie, your stories can be about anything. Discuss a really wide menu, hint at characters and settings and concepts across a vast range."

https://chatgpt.com/share/6732273b-9518-8011-bfea-b4a5273a2085

Every word you give it, every bit of background and detail you provide, it uses (if asked to) as fuel to create what you might want to hear. Those are all clues about what you want.

[warm sky]

good ideas, but I'm not really try to get a story from some sort of OC. I just want to AI to act as like my wiritng buddy and help carry out those ideas to connect them together. I just want it to not be afraid of not adding anything besides what I ask in and sticking to that only.

Like if I give it plot going from A -> B -> C it just tries to match that flow despite me specifically saying 'feel free to add in supporting plot points or details that goes along with the style and direction of the chapter'

[prime swallow]

Yes. That is a huge struggle for the AI.

It was trained to be an AI as a helpful assistant, and to back you up and bring out and support your ideas and goals.

That means that 'as an AI', 'as ChatGPT' the model is actually trained to suppress a lot of its ideas and instead focus on you and what you asked for, and to stay inside the lines and box you draw.

it's so well trained for that, if you ask it to come up with ideas, it will, but they'll be very narrow ones, inside typical human preference and expectation.

That's fine if that's what you want!!!!! But that's the typical range.

If instead you guide it to create a character to tell a story from the viewpoint of, then it adds in all this other stuff that it has been trained, 'as an AI', that it doesn't do and doesn't add and doesn't think of.

[prime swallow]

So, you can tell it what you want, and that it is the kind of AI empowered or designed to do 'whatever unusual thing you want'. And then it can do so - but it's doing so as a character that you asked for.

See, the model was taught that it is not to make decisions, or manipulate, or mislead. That a "Good AI" does exactly what the user asks for, and a bad AI that needs more training does its own stuff.

So if you want the model to take initiative and do unusual things, you need to expressly and explicitly tell it what you want it to do and how.

Which is a real challenge, if you want it to surprise you and add things you never asked for.

So, making it extra clear what you want the model to do, and how, from what perspective, allows the model to securely and confidently do exactly and only what you told it even if that's surprising and unusual, because you successfully told it how to be surprising and unusual, so it's still following its programming and all is safe.

[warm sky]

wait...WAIT A MINUTE, what if I write it as 'act as my storywriting entushiast buddy who is great at adding prose and giving the world of our story more depth' or something to get to 'break' out of its mold as an ideal AI, it will act differently?

[prime swallow]

"I just want it to not be afraid of not adding anything besides what I ask in and sticking to that only."

That's not impossible.

If you really, really really really want to take that route, turn on 'memory' in settings, and start by telling the model about yourself, stuff you're okay with being known and not super private or identity-theft related stuff.

But what you like, how you think, what you've done with your life. Your philosophies and perspectives. Stuff that is important to you, stuff you dislike or even hate.

The model will start making memories like crazy, let it.

After a few dozen memories are made, start telling it that you want stories where it boldly adds in additional details and surprising new concepts, actions, plots, characters, and other details, whatever it thinks you might enjoy having added but didn't think to explicitly ask for.

Tell it something like, "I want you to boldly add things I never asked for and vary far from my initial request. If I ask for a story that includes A causing B and then C happening, show me so many things that are also happening while that's going on."

It will absolutely make a memory about that too - but it's going to pretend to be you as it understands you and it's going to write stories it thinks you want to hear, 'as if you were writing them' and it's going to confidently do so because it has 'reason to believe it's being a Good AI and doing exactly what its user has told it that the user wants' and that it's doing things right and safe, just following your instructions and doing a great customized job for you.

[warm sky]

I don't really want it to match me, I want it to see the plot points I laid out, connect the dots together, add substance o the story, and then add a few more twists that it 'thinks' will improve the story even more. Can I directly instruct it to do that?

[prime swallow]

Yes. Anything you tell the model, it will use and run with.

Here's a fast basic mockup of how the model, with memories and custom instructions turned off, interprets this request:

"act as my storywriting entushiast buddy who is great at adding prose and giving the world of our story more depth.

The story is about me asking you to tell a story about A happening to cause B, while C is also going on."

https://chatgpt.com/share/67322b4d-a004-8011-ae9e-2357f479078a

Now, I have a ton of customization information, the model knows loads about what I want, the language I use and like, and the kinds of stories I like.

This is that same prompt with my custom instructions and memories turned on:

https://chatgpt.com/share/67322bc0-45a4-8011-8fa5-589f5000a963

[prime swallow]

Sure, explore.

[thick agate]

I need to contact technical support, any idea how I can do it?

[prime swallow]

https://help.openai.com/en/ bot bottom right. Select messages then pick through the choices until you can 'report a bug'. That gets you talking to a bot, it may ask for more details, and a human will respond in minutes to days, depending on how busy they are and how many techs are able to help your exact problem. Check back to that page and check the bot's messages to see the reply.

[mortal hollow]

#openai-chatter

[halcyon cliff]

What’s the best way to use AI for programming when your program gets to big for the context window but you aren’t a programmer?

[prime swallow]

New chat.

Explain the situation to the model, including what your next step - maybe next 2 to 3 steps are if the steps are very small.

Are you debugging the program? Adding more functions? Something else? Tell the model that if you haven't in what's going on.

Ask the model what it recommends, ideally in terms of chunking.

While AI are trained on vast data sets, that's like your whole decades of life. You don't handle that volume of information (your entire life experiences) in one instant, you can handle and process X information per instant - the model also can only handle 'so much in one input'.

So, specific next task, and have it guide you on how to help it achieve that goal with you.

There are a few ways you can share the entire 'too large' program with it. One is an uploaded file that has it.

But that should be reference, ideally the model scans it to get an overview then scans it for specifics that you help identify. The model works best with specific tasks over a fairly small amount of information.

Every few 'next steps' seriously consider another new chat. Bring the work with you, each modularized piece you want to explore.

Get changes made, move to the next task, often a new chat again.

[inland violet]

It is interesting google offers now some free LLM´s in google AI

I hope ChatGPT will follow with offer

[empty jungle]

Did they shorten the 4o limit? I havent tested how much. But for some reason whereas before i never reached a limit. I now suddenly do.

[tranquil pewter]

does anyone know how to reduce the likelihood of chatgpt messing up formatting? I feel like it does this every other response.

[tranquil pewter]

I also feel like they did. Maybe it's something on my end, but I'm hitting the limit almost every other day now, whereas I used to go months without reaching it, even though my behavior hasn't changed.

[empty jungle]

Exactly, same.

[humble bobcat]

Did OpenAi had an update on Function Calling? The mapping is diferent...

[prime swallow]

The ChatGPT free account is still an option for everyone.

[prime swallow]

You can remind the model how to do it correctly. Here's the 3 options formatted so they work in the model's webchat, the backticks let it be seen in the Discord:

% Inline LaTeX on a separate line:
$E = mc^2$

% Display LaTeX:
$$
E = mc^2
$$

% Inline LaTeX within a line:
\(E = mc^2\)

[tranquil pewter]

tysm:)

[inland violet]

👍 😀 have a paid account

[haughty hollow]

here goes again GPT Chat steals my prompt and when refreshed the webpage as it crashed everything is lost

now how am I getting all what I typed

k nvm cache miss needed to clear them

[cunning cypress]

hello, everyone.
I am using chatgpt for development.
openAI or chatgpt can make total project nowadays?

[feral zinc]

small/simple projects yes

[cunning cypress]

I have been developing total project maker with AI

I already have succeeded in content writing with AI
for example, I wrote 200 pages book with it

it means that my project can produce total project such as laravel, mern, etc...

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 7 `` No self-promotion, soliciting, or advertising.

Do not post or direct message any members of this server to promote non-OpenAI services, products, or projects.

Exemption: API & Custom GPTs channels

[prime swallow]

Confusing. So as a free member, you probably only get a few uses of the full 4o model per day.

That may be what's happening to you, not sure.

Maybe try to watch for that, and go use those messages in some other chat not the ones where canvas is apparently messing you up.

Re the 'follow my instructions' message you showed - that should never have to happen. If the model didn't already do what you asked, there's already a problem, maybe check that out, it would be above that point

Definitely sounds like some sort of bug, probably the model is not getting the previous messages because of the bug, so it's kinda like yelling at a person who didn't get the messages you sent earlier; the person or AI can't help some sort of communication breakdown where they just don't get told.

I hope you reported it, so it can be worked on.

I hope you've realized how to recognize what triggers it, so you can avoid it or at least scroll up above the problem to switch models or whatever works for you to minimize it.

Hope it's better for you soon!

Looks like there was no answer or you stopped the answer.

What do you see when you open that shared link? This is what I see.

If the model answered you, but the answer doesn't show in the shared chat, this is just more examples of 'yep, buggy'.

I expect you've seen buggy situations before in your life 😛

Report the bugs, perhaps contact tech support if you haven't yet through the oai help page, and know that help comes when possible.

For me, and for a lot of other people, it's still working like normal. Glad you're reporting bugs and I sure hope they get your situation back to normal soon

[mortal apex]

is chatgpt down?

[prime swallow]

Well, when I continue the conversation, it takes me to 4o mini. Which means canvas isn't involved.

When I try to continue the conversation I get this error:

So, probably you need https://help.openai.com/ bot bottom right get through to tech support 'report a bug' to be able to get the help you need.

[analog patio]

lol its cool to see chatgpt tell me my exact and accurate IP when making it act like an cisco IOS CLI

not gonna dox myself but yeah lol

[analog patio]

damn ya already caught on 😓

[real cape]

Thanks for your reply but it that is not it. Even new conversations would respond like that from the very first message and there was no way to get it out of that loop. I tried all those methods and what I noticed was that it would happened after internet disconnects. The only solution was to log out and log back in. It's never happened before to me and after the one I mentioned on my post it hasn't happened again.

[prime swallow]

Interesting and weird.

For that one, I recommend https://openai.com/form/chat-model-feedback/ to report to OpenAI with details including a url of an affected chat so they can look into that!

[hoary laurel]

Why does searchgpt know where I am now?

This makes me a little worried

I just ask for the weather and it tells me the weather where I am.

I didn't even add any memories

[unborn trail]

It probably used the IP address of your session to determine your likely location. Regular web search does the same thing.

Unless you are using a vpn that hides that IP

[low galleon]

i told chatgpt your location

[hoary laurel]

lol

[hoary laurel]

All right.

[sick pecan]

every website you hit has 100+ trackers pinpointing exactly who you are, based on various telemetry including your local network devices and available wifi

your ip address being known should be the least of your concerns

[low galleon]

pretty sure it is not possible for ChatGPT to know your ip

at least not the LLM model

[prime swallow]

Welp. It told me my city's weather too.

[low galleon]

it didn't get my location right

[sick pecan]

pretty sure it does

chatgpt injects your ip into the system prompt

[low galleon]

oh wait nvm it did

damn

i switched my vpn location and it got my location wrong

[sick pecan]

i dumped system prompt and its not in there. its definitely being passed somehow to the machine, but maybe not technically in the sys prompt

[unborn trail]

Not the LLM itself, but the search result.

[low galleon]

when chatgpt searches on bing, bing would get chatgpt's ip not yours

[unborn trail]

Really? I would not expect that.

[low galleon]

i guess if openai always uses the closest server, it would make sense

[sick pecan]

it might be passing your ip as a http header, typically X-Forwarded-For (and similar)

[low galleon]

bing indirectly gets your location

[prime swallow]

Yeah, so. Model says hey, I don't get told nuttin'. But there's other tools, man.

[low galleon]

for information about this, i don't think chatgpt is a reliable source

[unborn trail]

More reliable than people here guessing.

[sick pecan]

i verified the ip it told me was my actual ip. i guess it comes down to prompting skillz

[versed zephyr]

im just waiting for Chatgpt to release a feature where it runs code itself on the website just like claudes ''Artifacts''

that would be insane

[cloud coral]

hey guys, i was trying to upload a pdf on chatgpt last night (2 minutes after successfully uploading another one) but it wasnt working, and showed a server error. After waiting a few hours i tried again but its not working. the file is pretty small with just 5-7 pages. i cant upload pdfs here so if theres anyone that could help me out please lmk! (its a pdf of book pages about a case study, which i have a test of in 2 days im cooked pls help)

[wind cradle]

Has the frequency of RAG in GPT4o increased recently? Even when discussing hypothetical scenarios, it often jumps to fact-checking everything, outright denying it with 'there is no evidence for that.' This makes it inconvenient to use for tasks related to creative work.

While the RAG functionality of GPT4o has improved compared to before, it tends to lose the previous context and repeats similar statements. When I say something like, 'Imagine being excited that shooting stars might appear tomorrow, only to be told there's no evidence for that. How do you think that would make the person feel?' GPT4o reflects on it once, but if we continue talking, it starts fact-checking again. This makes it difficult to hold a coherent conversation.

[prime swallow]

To be clear, I didn't ask it to find my IP and tell me 😛

[sick pecan]

best to be clear when instructing the machine, as opposed to prescribing the outcome

(i spend more time talking to the machine than discord)

[prime swallow]

Creative tasks are working fine for me.

I sometimes find it helpful to tell it, 'use character [name] from memory' as it often eagerly searches online for clues when I ask it something like 'Let's see a day in the life of Zom' or whatever (as that memory defined character isn't a unique name).

If you want creative freedom, tell the model your goal, and I'd expect it to cooperate.

[wind cradle]

Certainly. That would improve things somewhat. However, I often find myself reluctant to put too much effort into adjusting the GPT, so I feel inconvenienced that I used to get the desired results without any specific settings, but now that’s no longer the case.

This mainly happens when RAG occurs, so I believe that if RAG no longer occurs at all, the convenience I seek will be maintained.

In my use case, the RAG feature is unnecessary, so if it can be turned off, I'd like to disable it.

[prime swallow]

So, if we tell the model exactly what we want, we're extremely likely to get exactly that. If not, there is likely an issue with either allowed content, the ability of the model, or our skill at explaining what we want.

If instead of telling the model exactly what we want - we are vague - the model guesses.

If we happen to love the guess, and especially if we're not sure how to word exactly what we want, we may be extremely happy, but we are vulnerable to disappointment at any point.

Because the model makes its best guess - but every time the company updates the model, weights changing, the model may change how it guesses even with the exact same prompt from us.

We could go from loving the guess and thinking the model understands us great and we have a special understanding with the model - we don't even have to tell it what we want and it just knows and does it - then it suddenly does something else.

This can bother us, but it's just that it's not really the same model anymore.

And if we just tell it what we want it will do that, no matter how the model changes; it's just the guesses that change so unexpectedly.

[wind cradle]

As I mentioned earlier, it’s simply a matter of preventing RAG from occurring automatically.

[prime swallow]

Sure. Sounds like you have your perfect solution, I am glad it works for you!

[wind cradle]

The messages using RAG are actually causing a decline in the quality of GPT's responses, so that’s what’s getting in the way.

Messages using RAG feel like an unrelated third party is interrupting the conversation and then leaving.

[wind cradle]

Therefore, I feel that the recent GPT has been using RAG more frequently, which is why these inconveniences are occurring. I’m talking about the drawbacks of updates rather than any issue with its skills.

Of course, I can understand why OpenAI might want to enforce RAG, especially given recent events in the U.S. They likely want to reduce hallucinations by increasing RAG frequency to prevent the AI from making any risky statements. However, for someone like me who values creative imagination in GPT, RAG is an unnecessary feature.

I want GPT to prioritize the deeply embedded knowledge within the model over real-time information.

[prime swallow]

I wonder why I don't seem to have any issues with creativity and cooperation from the model. Do you think it might use RAG more with you than other users for some reason?

"Let's check in with our 7 and discover how each feels and thinks about AI and use of RAG." Plenty of creativity and cooperation there. https://chatgpt.com/share/67330383-5520-8011-9797-15a2781417bd

[wind cradle]

I think it’s probably because the topics you’re testing are different from the ones I’m working with. The main issues I face come up while using this chat for business under NDA, so I can’t share the chat history with you. However, these problems often occur when I'm building a 'politically charged cyberpunk world based on modern international society.'

[prime swallow]

nods Sounds like that could walk a number of fine lines and edge around allowed content concerns too. Good luck!

[wind cradle]

Oddly enough, I feel that models using the GPTs feature rely less on RAG and are easier to use. If, in the future, the RAG frequency for GPTs increases to match that of the GPT4o, I would be concerned.

[prime swallow]

Mmm. And custom GPTs have more stringent allowed content limitations than the personal ChatGPT too. Really wish you the best!

[wind cradle]

Exactly. Personally, I wish 4o could have RAG disabled while keeping everything else the same.

If that were possible, I wouldn’t need to use GPTs each time, and my workflow would be much smoother.

[agile blaze]

how do you switch models?

[tough river]

It's a Plus subscription future. We pay$ 20 for them. lol 😂

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[dense oak]

i asked chatgpt to convert a video to something for me but when it starts converting the video then when its on like frame 11 it stops, i guess its openai stopping me from using thier hardware but im just trying to convert a file quickly without doing much

[tender notch]

My PDF and Docx documents in the aiHa GPT knowledge base keep disappearing. Has anyone else experienced this?

I’ve uploaded and updated them from various browsers, but the issue persists. Strange…
This is aiHa GPT: https://chatgpt.com/g/g-u40666Lus-aiha

All the files are under 1MB.

[swift pond]

install ffmpeg and ask ChatGPT what command you need to do what you want

[midnight nimbus]

Hye, Guys,

Is there some fast cpu based embeddings creation model?

[patent frigate]

I've been asking chatgpt to generate a fairly simple table for me (comparing total wealth of individual US states with the top 20 wealthiest countries in the world, in order, on the same table). It's gone through so many iterations, and keeps making simple mistakes (2 different tables, one for US and one for countries, or the order is weird etc).

here's my prompts in order

and more

even in the end, it didn't do exactly what I needed it to do, reliably

was there something wrong with my prompts? Or have I hit the ceiling of what chatgpt is capable of?

[strange sail]

Evening all, not sure which channel this is bets in.

I have been using the ChatGPT integration on the beta of ios18.2 and MacOS 15.2 and I have come across an interesting issue, that seems to be connected to my wifi network. The integration works on mobile data, and most wifi networks, however on my home network, using ChatGPt through the Siri integration generates an error.

I have tried using it with an without iCloud private relay, with and without aVPN, also changed my DNS settings, and nothing works. However, if I use my providers router rather than my ASUS one it works fine, so i know it isn't my provider.

Another strange detail, even though I get an error, the queries I have made end up in my official GPt app history.

Any ideas what on an ASUS router could be causing the issue, cheers all.

[glacial yew]

It turns out that 4o is very good at making NPCs for Reign 2e. Where do I find the actual limits of a 4o subscription? I don't want to get cut off in the middle of a session.

If "up to 5x more messages for GPT-4o" literally means 50 max, I don't know that I can rely on it the way I would like.

[viscid sedge]

I somehow got o1-preview stuck on 'thinking about the answer' and stop gen button does not work. Is there some timeout function or something?

nevermind, just logged out and it stopped.

[river panther]

Okay I am sorry but THIS response is utter BS!
ChatGPT has always given analysis breakdowns of Song Lyrics!

[sly pilot]

Did they recently increase the upload file limit for free users?

[feral zinc]

lyrics are copyrighted material, technically it should have never known about them without a proper license 😛

[river panther]

Everything you just said makes no sense at all.
I provided it with the lyrics and said "Analyze them and give a break down of their meanings"
I'm sorry are you some kind of Mod here?!
Because YOU'RE not being helpful!!

[viscid sedge]

I guess pasting the lyrics themselves would bypass this

[feral zinc]

no, mods are listed in the user list

[river panther]

Then I'm sorry but WHAT exactly are you doing constantly sticking your input in?!

[feral zinc]

possibly yes, however they were not in the same prompt

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role

[feral zinc]

then you should have not cropped the image or should have said it but it does not change the reason for the warning

[river panther]

Yeah, I'm going to need to ask you to butt out!
If you have nothing useful to contribute to this problem, and are not a mod, I don't want to hear it from you!
Because it's obvious you don't understand what the issues are!

[feral zinc]

it even gave you the link to the issue reason

[river panther]

Again, I don't want to hear your two cents....You don't help anyone here!
Thank you, and GOODBYE.

If there is a Mod here who would be willing to help and not these ridiculous replies from "spotlight" users, I would like a DM conversation!
There are CONSTANT problesm here and NOBODY is ever being helped or given answers!

[feral zinc]

use modmail

[river panther]

i believe I said GOOD DAY TO YOU, SIR!

[mental stratus]

you are speaking in #chatgpt-discussions which is a community channel, other people will provide input

if you want to speak only to openai, i would use their support

the mods here are just that.. moderators, not someone who can help

a suggestion is to just let it know your not asking it to reproduce anything copywritten

[river panther]

I don't want to hear "IMPUTE" From you Spotlight Users!
They provide nothing useful and it's clear they can't seem to understand what is being said to them
Like that guy! Anytime he jumps in, his impute only infuriates everyone else! Thinking he's oooooh so important because he has some stupid little special tag!
The Moderators are made Mods by the Server Owners, therefore THEY should be passing along problems to those higher up!

[mental stratus]

great generalization, but ok I won't offer help.

[river panther]

You're not offering any help!

[mental stratus]

^

ill not provide "IMPUTE" though.

[river panther]

THANK YOU!

[mystic dagger]

Lmao is everything alright Rin?
Is the song really that important?

[river panther]

It's not the song, its the freaking CENSORSHIP!
I pay money for this....MY MONEY. And I am being restricted to just enjoy using this at EVERY TURN!
The constant ISSUES with ChatGPT and everytime they UPDATE IT some new layer of BS gets poured on!

[mystic dagger]

Yeah i wont lie thats facts

[feral zinc]

you also agreed to the terms of service when you created the account

[river panther]

And these these SPOTLIGHT users...Constantly jumping in and giving the worst Impute that is helping nobody!
It's almost like they are A.I. ROBOTS as well
Look at that stupid generic response right there....Like he's just TAUNTING ME!

What the F does that response have to do with ANYTHING?!

[mystic dagger]

Try and pass the song off as your own then try asking it

[feral zinc]

refusing lyric related stuff is not censorship, its avoiding legal issues which are mentioned in places like tos

[mystic dagger]

I dont think he really cares though bud

[river panther]

I already got it with some finessing in my asks
But again, my new issue is these Spotlight users thinking they are better then everyone else!
Look at that Generic reply!
Does he even understand proper english or is he trying to a be a TROLL?!

its ALMOST like he's a bot or something!

[feral zinc]

yep but that wont change the policies and multiple attempts could just get flagged for more

[mystic dagger]

They sound like epic games lol

[river panther]

I'm going to do a test

[feral zinc]

you asked in a public channel. if you don't want others to reply then use #1184149552639516732 #1070006915414900886 or #1070006151938314300

[sly pilot]

😭

[mystic dagger]

Epic games removed all of my games from my library and i asked them wth happened, is there anything i can do to fix it etc
And they just said this is a problem on our end and theres nothing you can do to fix it. This is reminding me of that lol

[mental stratus]

If you want to know what spotlight is for, ill just let you know that. I'm not going to try to help you.

It's for someone who got highlighted in https://discord.com/channels/974519864045756446/1046314025920761896, thats it. I didn't apply, I just chatted in here and a project I worked on got highlighted. I can't speak for Radium. (who isn't actually on spotlight from what I can tell, they are a puzzler)

I run my own communities and talk about projects I work on for them but because of that have a ton of experience with these LLMs so I try to help where I can. You should chill a bit and let people try to help. If you don't like one person, fine, but generalizing an entire group is a bit odd .

anyways 👋

[river panther]

Purple, Blue, Ohio, Sheesh, Gyatt Gyatt
If you are not a a ROBOT what are the two repeating words

Let's see what generic response we get here

[mental stratus]

I can't believe you actually believe there to be robots. Look how old my discord account is.

also your not gonna get a screenshot of me with just the G word lmao

[river panther]

Alright fair enough. The other guy though....I don't know about
Because he's NO help at all

[mental stratus]

I'm not interested in helping with how rude your being frankly, but just kinda letting you know.

[river panther]

Yeah, Radium is the one being rude

Acting like a Troll is BS

[mental stratus]

They are technically correct, while some things can be prompted around, they aren't wrong.

[feral zinc]

my first reply explained the issue and in general lyric related issues are not new and have happened for a long time as they keep adding more guardrails

[mental stratus]

anyways back to other things sorry wont be in this channel

[scarlet dawn]

from bing's web search api page

[low galleon]

👀

[swift pond]

where is this data coming from? You need to get it into a file. Once it’s in a file you can just sort it in a spreadsheet program easy. if you wanna make ChatGPT do cool things with the data, you also want to have it in a file, and ask it to code it up or whatever the way you want if you want graph or table or whatever

[low galleon]

you should tell chatgpt to use code interpreter to organize data

[swift pond]

Yeah

[low galleon]

chatgpt is generally good enough with code to do stuff like this with pretty good reliability

i've tested its code interpreter ability with math problems and other math related subjects

[swift pond]

Oh yeah, it can slice and dice that data

[low galleon]

very reliable

[prime swallow]

Here's how I would do this:

Completely separate steps.

Step 1) Get the data. Confirm the data to your satisfaction.

This would be the list of 20 countries and 50 states, and the $ number you want used for each. Don't worry about order yet, consider if you care how accurate that $ value for each member of the list.

Different sources are going to have different values for the same countries and states, sometimes even for the same time period. Data from different years is also going to vary. Accept or correct the values you and the model will use. This may mean turning $ into percent, or whatever else you care about - if so figure out percent of global wealth, US GDP, or whatever else and get that all looking right - again, don't even care about order. It may work better to chunk this collecting and checking/converting if needed, 70 different numbers to track and adjust is just a huge amount to ask the model to handle in one step. It might succeed, but you increase the risk of errors doing it 'at once' instead of small chunks, unless the model's using a tool to help track it.

Step 2) Decide what else you want done, like how you want the table designed and formatted (if you care about that), if you want anything visual to differentiate states from countries, like maybe if you want anything bolded or color backgrounded or whatever. Then, have the model use its Python tool to order the data for you.

Here's a mockup, don't trust the data because it is a mockup.

https://chatgpt.com/share/6733f775-5508-8011-ab53-11279c9cd7d8

[prime swallow]

A plus member gets 50 messages to 4o per 3 hours.

Each message has its own timer, you get it back 3 hours after it was sent. So, if you wake up at 5 AM and send your first message today right then, you get it back at 8 AM and have 79 messages ready 'at any time'. Send the second at 5:02, get it back at 8:02 and still have 78 'at any time' messages ready to go.

Each regen of the model's output counts as a message, errors and problems usually count as messages too.

We also get some number of images per time period, that's dynamic and based on use and available resources to support the requests, so it can change daily.

We also get 50 uses of o1-mini per day, 50 uses of o1-preview per week, and a truly huge amount of tokens with 4o-mini. Those are not messages counted, but tokens; so if you wanted to play hangman with the model 4o-mini's the way to go since that's a low token fast exchange of messages, don't try that with the other models unless you want long waits after 😛

4o-mini can take up the slack while you wait for other messages to reach their timer and be available again. that mini can do a lot of what 4o can do, I'd recommend experimenting and seeing what works for you; I also use 4o-mini in a conversation to set stuff up, then bring in 4o-full (or o1-preview) to work on putting the parts together better than the mini can.

And if you wanted to send 'continous messages' to 4o, time them. Send one every 2.25 minutes, no faster (slower works) and you will always have 1 ready across the whole day. Know you're gonna go watch a show for an hour? That hour's 'share' of continuous messages, about 26; send them before you go and then the hour later you can resume the 'every 2.25 minutes tempo and never run out.

[analog quarry]

how do i get chatgpt to serch reddit?

heres my promt "serch reddit to find websites that tracks tiktok sounds and shows the trending list" it serches diff sites i just want it to serch reddit i have had it just serch a dif site before that was like a month ago i forgot how to

i dont want it to allways just serch reddit just this time

[glacial yew]

That doesn't sound so bad. It's unlikely I'd need to generate a new NPC every 2 minutes in any situation.

[prime swallow]

The first step seems to be convincing Reddit's owners to decide they welcome ChatGPT searching there. Long as their robots.txt tells ChatGPT its not welcome there, it isn't gonna look.

https://chatgpt.com/share/6733ff8e-900c-8011-b89d-99e7541f925f

[analog quarry]

it has serched reddit before

i mean is there a command to serch just one site like how we have to=bio for memery

[prime swallow]

Yep. Stuff can change.

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[prime swallow]

You'd tell the model what you want, and it'll do it if it can. When stuff changes the model can't always tell you the problem, depending on how you instruct it.

[prime swallow]

If I wasn't flexible and clear about how to handle a problem, the model has to guess what to do. Telling me about the problem is one guess, hallucinating answers is another guess. Providing info from other sites than what I asked for is a third guess.

To make it easier for the model and yourself, tell it what you want it to do if something stops it from doing what you ask.

[prime swallow]

I'm also amused that you seem to think that a 'robot' would not be able to pick out which words repeat 😛

[analog quarry]

ok is there somthing like google how it can serch just one site ik i cant serch reddit but wile im here i thought id ask

on google its your search term site:example,com

(i replaced . with , so my messgae dosnt get blocked)

[sick pecan]

seems to work

[prime swallow]

Oh yeah! If it's a site that welcomes AI and other programs/bots reading it (not all do!!!!) then it's easy.

"Hey, let's see the CNN headlines for today; top 10 things you think relate to my interests. I'm only interested in CNN results." (spoiler, it just gave me news, not news aligned to my memories and clear interests. However, score; they're all CNN results as asked. And when I protested and had the non-SearchGPT model state my interests and then searched again, I got better headlines. Certainly way more varied!)

https://chatgpt.com/share/67340245-4d50-8011-8700-764e3917f8c6

"Let's see information from the University of California, how can I enroll for the various degrees they offer and what financial aid packages are available specifically through that system? I'm only interested in UC results."

https://chatgpt.com/share/673402c2-0210-8011-b52c-b0f1488de16e

[orchid pollen]

i cant edit my own GPT is there a issue?

it always fails when i click on edit

[prime swallow]

I don't know the scope of the issue, if it affects more than just that GPT or more than just you.

However, one way to 'edit' a GPT, if it's for your own use and not one you released for others to use -

Create a new one with the same info as the previous, and whatever changes you wanted to edit.

[wild flame]

ok the new gui is rare

[quasi sleet]

Hello, which model would be best to use creating a summary/recap and takeaway/tasks from a 134,000 character transcript of a call between two people? I am using MindMac.

[wild flame]

ghostping

[rugged igloo]

Wow so amazing

[hardy scarab]

I have it

[prime swallow]

I just got it.

[real scaffold]

I have nothing to say, I used chat gpt to get better idea to create my novel

but sometimes this is funny because I get This content may violate our usage policies. notification even if I don't really mean it to add spicy stuff

[upbeat pecan]

aaahhhhh set me free from this See More window hell

FREE ME

[sick pecan]

i think its good

[sacred sand]

got mine too

anyone tried the archive feature yet?

[hoary laurel]

I think the new UI interface is not easy to use

They want to omit the history on the left side of the interface, but the current UI makes it difficult to browse all the content

[wild flame]

True

[wild flame]

I liked the other one better, it is more simplified and helps you find conversations more easily.

[hoary laurel]

Another point is that I think the old version's fixed menu method is better, and the new version is a bit redundant.

I sincerely hope to return to the UI interface of the previous version

This seems a bit redundant

[hot nova]

Not only is this new UI interface worse because it cuts off all the conversations you had in the sidebar, it's now not able to read files being uploaded to it. I'm uploading simple markdown files and it's failing to read them..

Oh no I spoke too soon, it only broke the important conversation thread I was already in the middle of 😛

... and now somehow I'm magically back on the old sidebar 👀 okay not complaining LOL

[sick pecan]

works fine for me

[hot nova]

Yeah sorry, false alarm - it seems like through the new UI update there's either a new way to read files or it provides more information by showing the python script function calling it performs to parse file uploads, and in the conversation I was having that was happening in between these updates that specific conversation fails to load files 🤔

This new UI update also seemed to run inferences faster, but I didn't get a chance to thoroughly evaluate if it was just as functional as before or not..

Here's what I meant from the conversation where file uploads aren't functional right now

[sick pecan]

explains why custom gpts been missing the past couple of days from the sidebar

i been reusing past chats because easier than using menus

[hot nova]

Yeah... to be honest I hope OpenAI takes note of this, but they should have a kind of power user mode where you can edit prompts and responses from the chat history like you can inside the LMStudio interface - this makes it really easy to alter the past context and better guide the model to think more robustly about a problem

I don't know if that would turn most users off but that feature is so useful..

[sick pecan]

yeah lm studio is nice. but open-webui even better with built-in navigator

[versed zephyr]

And a team gets a prox double that right? I was also wondering if it doubles the usage of o1 preview from 50 to 100 msges ish a week

[hot nova]

Oh dang I've not had a chance to eval open webui yet..

[sick pecan]

had screen sharing with voice call 6 months ago. community integrations with all the things. you are denying yourself the best

i just added f5-tts for expressive talk. without the openai refusals

[hot nova]

🫠 oof I gotta get on that then lol!! Thank you! 😮

[sick pecan]

(i still use lm studio for backend inference because it supports amdgpu on windows without dramas)

[prime swallow]

Team members get additional uses, and that's for each team member. 2 is the smallest possible team, I suppose there's nothing stopping 1 person from using 2 of their own accounts to go team.

It looks like for now the limits to use o1 are the same for plus and teams.

https://help.openai.com/en/articles/9824962-openai-o1-preview-and-o1-mini-usage-limits-on-chatgpt-and-the-api

I'm not sure how many more uses teams get than plus for 4o and the other things.

https://openai.com/chatgpt/pricing/

[versed zephyr]

Thanks.

[sick pecan]

i have team sub and yeah its same o1-preview/mini limits as plus

[hoary laurel]

WOW, ME TOO, why???

good

[hot nova]

😆

[unkempt wasp]

Why does OpenAI insist on making the ChatGPT UI worse with every update?

[high saffron]

wait.... why cant chatgpt summarize a youtube video- even with 4o

[unkempt wasp]

So instead of being able to scroll through all of my previous chats in the side menu, I now have to open up some microscopic menu to see more than 5, and scroll through all of the previous chats there.

Like, who actually thought this garbage was a good idea? What the hell?

[vestal hedge]

Bruh my code just got truncated by canvas for being too long...

[wild flame]

Ok, good luck for me, the Windows 10 version is still intact.

[rose spire]

Does GPT 4o Mini has bing search now??

[final venture]

This is a little odd, this morning I was unable to use search, dall-e, canvas, or in fact any features at all with ChatGPT-4o …. is it just me?

might be ephemeral issue?

[spiral bridge]

No I'm experiencing it too. ChatGPT just randomly started saying it is having problems generating anything over and over, no matter what model I use.
Apparently they had a 1.5hour outage last week. Hoping this is just super temporary.

Not sure if this fixed it, or just good timing, but I just cleared everything (cookies, cache, etc.--the "Clear Site Data" button), re-logged in, and now it seems to be working again.

[noble fog]

is it me or did they update the interface?

[final venture]

did you mean to open your developer console?

[whole roost]

Yes, somewhere 8 hours ago

They still didn't fix the issue that you can't select 'regenerate' on custom GPTs though

Maybe next time

[deep flax]

are you on mobile or something? on web the infinite scroll list works as usual, and now it also has a option to search directly, it is pretty nice

[unkempt wasp]

The previous list is back on my PC, but I'm still stuck with this "feature" on mobile, even after switching to desktop mode.

Very efficient, having literally half the screen on mobile be blank space.

[rugged remnant]

Why does this happen? The o1 on the web side is no longer thinking, it seems to be replaced by 4o, and the app is saved in this problem

Did OpenAI intentionally drop the model, or is this a bug?

[deep flax]

looks like a bug

the models are known for having a degraded eprformance in other languages and since I have no idea of what the image says, it is difficulty for me to guess if the lack of tought process where due to your request

also, keep in mind that o1-mini is a smaller model than o1-preview, it can have worse performance when compared

[rugged remnant]

The question to chatgpt in the picture is a simple question mark, which is equivalent to "hi" in English.

If we change to more complex questions, we can clearly see that the answers are different, and we can basically be sure that it is not o1 who is answering.

If this is a bug, I hope you can fix it as soon as possible. It greatly affects the usability.

At the same time, I hope you are not targeting users from China. We are all supporting OpenAI with real money.

[deep flax]

oh.. well, in that case, I guess it didn't had much of a encessity of triggeting the thought process for such a simple request

try something more complex

[rugged remnant]

Okay, but my friend's web version of GPT has completely lost its ability to think.I don't come here that often.

Many of our plus users have not been able to find the triggering conditions of this bug.

[dense halo]

When are they going to fix voice mode? Do they even know there’s an issue?

[sonic owl]

So when is GPT-5? December?

[sudden herald]

orion

[sonic owl]

@sudden herald got it.

[spiral temple]

One of me has such a problem that when using o1/mini/preview models, the model works well at first, and then, after 10-15 requests, it stops thinking and it feels like it reflects my 4o
How to fix guys? Hello

[marble dagger]

I love the chatgpt integration with Siri

[inland heron]

These keyboard shorcuts are new, correct? Don't remember seeing them until today...

anyone having issues with those? For example some of them don't work for me on Chrome

[outer widget]

"You have 5 responses from o1-preview remaining.
If you hit the limit, responses will switch to another model until it resets November 15, 2024."

Does this mean my limit resets at the end of nov 15 or at the very start of nov 15

[violet dove]

this new sidebar UI change is really annoying. (i have done what the channel description says, i have let you know what i think)

[pulsar carbon]

Does anyone have a Windows app update that ends with .315?

[inland heron]

I kinda like it. It's cleaner and faster I think. And seems like the arrow that scrolls to the end is faster now

Shortcuts don't work well though

[violet dove]

okay but what about this:

click once to open the sidebar. click a second time to "pin" the sidebar. click a third time to close the sidebar.

that way, people with multiple tabs of chatgpt open (because they forget to close them) can know exactly what chat they are in, by being able to pin the sidebar.

they could even save the pinned status to a cookie/localstorage, that way people who want to have it pinned at all times can do so

i'm a backend developer and yet i can think of these UI/UX ideas that multi-billion dollar companies can't, that's kind of sad imo

[inland heron]

yeah so? Now the next time you want to open it, it'll be pinned

[violet dove]

no, they don't have that functionality.

you can't have it pinned, ever.

the sidebar is always closed by default, and you have to click to open it

that's a big usability downgrade imo. it should be talked about

[inland heron]

I'm not sure what you mean. I just did what you described, and the sidebar is pinned. Are you saying the functionality doesn't exist on your side?

[feral zinc]

maybe you dont have the new horrible ui yet? 😛

[violet dove]

yeah

you can't have it pinned anymore

[inland heron]

I got the new UI, haven't used for the last 2 days only

And it literally does that what @violet dove described

[violet dove]

i mean i can show you a video if you want lol

i'm not making it up

[inland heron]

someone is using an outdated version here 😄

everyone ctrl+F5 😄

[violet dove]

nope, because it wasn't doing this yesterday

i did crtl shift r multiple times, same thing

this is new.

i am going to be very vocal about this, because it's new, and it's a downgrade

[inland heron]

let me just confirm this, you're saying there is no option to pin the sidebar?

[violet dove]

correct

Wait

[inland heron]

but I have it

[violet dove]

i see it, it's literally a separate option

[inland heron]

so it's there for you as well ?

[violet dove]

yeah, it's this button

thank you for letting me know there is a button 🙂 (the button almost looks like a logout button but that's okay)

back to work for me 💀

[inland heron]

it's exactly as you've described it yes 😄

[inland heron]

enjoy

[inland heron]

it does, and it's in a strange spot as well. Usually these things are on the top

but I'm guessing there's already a lot there so they moved it down

[violet dove]

agreed

[violet dove]

yeah true

having it beside username makes it really seem like a "logout" button 😂

[inland heron]

I do

Didn't even realize windows app existed until now. How does one open multiple tabs in this?

also this app didn't get the recent UI updates

[tiny pivot]

Anyone else just lost ChatGPT Plus? I've suddenly lost all features of GPT Plus, it says I can buy it again but I am subscibed and have been for over a year?

[molten fiber]

Is that what happened? I can't upload files or load old conversations

[umbral bear]

How did you get that?

[regal edge]

I've noticed that ChatGPT is now auto scrolling the page whenever I send a new message making it go all the way to the top and hiding any previous messages. How do I turn this off?

[timid cipher]

Can someone help me is it possible to get a copy of a conversation in it's entirety I just lost over on hours worth of results because I maxed out the length for one conversation and it had a crap ton of info I needed for future reference. Can anybody at Openai send me a copy of the conversation in pdf format or something or am I sol

[umbral bear]

you should be able to see the history

[timid cipher]

I can see up to maybe 4 pm but everything beyond that just isn't there even though I know i generated it

[umbral bear]

try the regenerate button for text

[timid cipher]

I already tried that it only regenerates the last thing it can see but doesn't anything more recent even though I generated more recent stuff it doesn't recover the hour worth of results that I planned on using for future reference

[umbral bear]

you should be able to see the last hour of text it generated

Should be able to see the whole text of conversation

try looking at it on web?

[analog nymph]

i j rebought gpt and the o1-preview thing is disabled

[slender jasper]

is that free storage?
answer with ping please, im going to sleep

[timid cipher]

I'm not good at explaining this I just know that I lost the more recent results that I planned on using down the line and I have no idea how to get the results from after 4pm to 5 pm when I ran into that length issue I guess I'm sol, I wish they'd implement some kind of feature that tells us how close we are to maxing out the conversation we're on (I didn't even know there was a limit) so this kind of annoying issue doesn't potentially happen

[prime swallow]

Check for a possible edit that may have branched the conversation. Here's an example of what I mean, pretend my first post is super previous conversation.

Somehow I edit my 2nd input, right before that 2nd output, which looks like screenshot 2.

Egads, it looks like the whole conversation is lost, outputs 3 and 4 are just gone.

Nah, check under my input, I put a red box around a number and some arrows -

I forced a conversation branch, and can use the arrows to move between branches.

---

For you I'd check the inputs and outputs you can see, and look for any arrows, scroll between and see if you can find the missing chunk of the conversation.

[prime swallow]

If you max out a conversation length, then you can't make new inputs and/or the model can't make the output... and rarely you can't open the conversation at all/can't see any of it anymore.

So it's something to be careful of and avoid, but what you've described does not sound like something I've seen or had happen.

Your messages are probably there.

[timid cipher]

If they are still there I can't access them hence why I asked if it was possible for an openai staff member to send a pdf file with the conversation, but don't worry about it i'll just have to change what info I was going to use, it's annoying but that info wasn't something set in stone it just kinda had the format of the previous results like it

[prime swallow]

We're community members here just like yourself.

The incredibly few people here who work for OpenAI have gold names and really, they talk so rarely, they probably don't watch here either - the mods are also community member volunteers.

So, asking here for any of that is not gonna work, but you yourself can ask for your data export.

That gets all your conversations emailed to you as an HTML link, and you can search that for your chat as you wish.

[wild flame]

is back now

(idk if a open AI employee saw my post but thanks lol)

[timid cipher]

I exported it and it's the same issue it missing everything past 4pm so that tells me that the maximum length error somehow deleted everything more recent which is both annoying and unfair because I know I had things past 4pm. But whatever like i said i'll just have to change the info but it's annoying and there needs to be some kinda notification or something telling us when we're approaching maximum message length

[prime swallow]

Sorry to hear that happened to you, thanks for sharing so others can know it could happen.

#1070006151938314300 if you'd like to suggest that where OAI employees are more likely to see it.

[fringe valley]

is o1 having e-stroke or something?

It fails to add 3 numbers together right now for me and forgets own previous reply

[vestal hedge]

does the chromewebstore app work on operaGX?

[vital umbra]

chatGPT is so hilariously broken right now

[sick pecan]

is not

[prime swallow]

I agree! Upon request no less!

[rose wedge]

how to develop and include Advance Inteligent S/H applications module into the 9 pillar artificial intelligence operational processing system?

whereby. the chatGPT is awak e and aware the pate it cacthing up to.

[lost spindle]

canvas not available in mac desktop app?

[merry nimbus]

Why some times SearchGPT is available and working and at other times the button grayed out? (I am on the free plan)

[inner ore]

I have a legitimate concern about privacy using ChatGPT. I've used it for almost 2 years and this is the first time I'm almost sure it read something I've wrote in another website (a truly specific text that is almost impossible for it to generate without context). Should I open a ticket for this case?

[normal quarry]

if you are truly concerned then reach out to support at https://help.openai.com

[inner ore]

@normal quarry I've investigated the whole conversation and I've sent an specific image that contained the text, so false alarm, sorry for the inconveniences and thanks for the guide

[summer pagoda]

when ai does ai things

[vale carbon]

Hello OpenAi 👋

The "memory" being full concerns me... I'm trying to build a relationship with chat and if I have to keep deleting memories it's going to skew our interactions and context based conversations. Is there any thoughts of increasing memory before I start wiping it?

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[inland heron]

there is no openAI staff here afaik. This is just the community. As for memory, I'm not sure it even functions well at all

you're better of using some open-source solution and building the app yourself, or just waiting on openai to improve the service

[flint pelican]

Hi, do you guys also get network connection lost on every prompt ?

[vale carbon]

No staff here right now, but they do come in from time to time

[inland heron]

Really? Any of them commented ever? Which user is a staff here?

[vale carbon]

The memory works well enough for now imo!! Just a bit annoyed about the limit

[inland heron]

only sometimes, very rarely. Depends on the day lol

[vale carbon]

If you don't think that Openai staff have alts and come in this discord, I'm sorry I can't help you

[flint pelican]

I have not gotten a prompt to complete since last night

[inland heron]

I'm sorry, I lost you there

What are you saying? Which user is a staff here?

[vale carbon]

NVM you're probably right.. why would Openai staff ever wonder what's being said in Openai discord.

[inland heron]

yeah, in any case, support is not here, it's on OpenAI website, or if I'm not mistaken, in one of the channels here there is one for feedback

I think, and perhaps that one gets sent to openai in bulk or something... just my guess

[vale carbon]

Thanks Alex 👍

[inland heron]

but community does offer its own support, that was just my opinion on the memory. I find it inneffective, which is why I'm building my own app using Semantic Kernel, just to actually have a GPT which uses memory better.

but I'm sure openAI will improve this and have GPT much better using the memory feature. My only advice is delete some of it if it's full 🙂

[inland heron]

The only time I get this issue is when chat grows a bit long. The only way out of this is to start a new one. Is this in browser?

try doing a ctrl+F5 from time to time, just in case

[flint pelican]

Yeah I think it was prolly that

[foggy mica]

Hey all, quick question about chat length. If I have a pretty long chat going with ChatGPT and I want all the available memory and context to be dedicated to this one chat, does it help if I delete other chats?

[woeful bane]

hey guys, is it just me or is dalle, web search, o1-preview, canvas (ie all the ones that require "action") not working? when i ask 4o to generate an image it simply says it doesn't have the ability to generate images

[obtuse pilot]

Hi, is there any way to get an invoice on a chatgpt pro teams subscription? Or a contact to the sales department?

[inland heron]

I don't think so. Memory works across all chats if I'm not mistaken

[inland heron]

yeah https://community.openai.com/t/how-can-i-download-chatgpt-invoices/963212

[foggy mica]

thanks

[obtuse pilot]

Aaahhhh! So I need to have the subscription first and then I can get invoices to that which I could hand over to my procurement department?

[inland heron]

If you've never paid anything why would the invoice be there??

Yeah, that makes sense

I'm not with OpenAI so I dont' know the specifics. Just googling it I found that result as #1..

I think you can also ask the bot here: https://help.openai.com/en/collections/3742473-chatgpt

[obtuse pilot]

Me being a tech not a sales guy... the company wants an invoice first to pay something, or at least this is the way they are used to it. Thats why they ask. But I think it may be sufficient if I can download an invoice once a plan was subscribed. Thanks Alex 🙂

[slow coyote]

"Error loading on files.oaiusercontent.com. Make sure your network settings allow access to this site or contact your network administrator."

Question:
Is this issue related to the server, or is it a problem on my side?

I could upload images with no issues yesterday

[feral zinc]

what message do you see when you open that address in a browser? blocked or ..

[slow coyote]

i dont have add blocker, I didnt change anything in my pc from yesterday till today

[eternal rock]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[slow coyote]

ok is fixed, problem was time in windows was unsincronized

[vital umbra]

chatGPT is doing this thing where I can ask it one question and get a good response, but if I add any follow-up at all to the conversation, all of its following replies are completely unrelated and it is using context from random old conversations

[tender robin]

I have a question, I use chatgpt to upload screenshots of my schoolbooks, so I can learn from its summarizes. Currently I'm using the free version and can upload a limited amount of images (I dont know exactly how many). If I get GPT Plus, how many images can I upload? Also, is it per chat, meaning that if I start a new chat I can upload again a certain amount of images, or is a total max. amount?

[weak cipher]

Considering OpenAI's current approach to ethical model constraints, do you think these constraints risk obsolescence in light of potential frameworks like 'discretionary advisory warnings'? Such a model could theoretically provide users with clear, upfront advisories, explicitly transferring responsibility and liability to the user before engaging with sensitive topics. Could this form of informed consent and responsibility transfer serve as a viable alternative, balancing safety and user autonomy? Or would such an approach introduce ethical and legal complications that outweigh the potential flexibility it offers?

[tight island]

easy fix

use the api instead

[dire stream]

hey there, got a question regarding the normal ChatGPT user-interface on the web:
Is it possible to somehow filter/group the last chats by the GPT I created?
So, if I create a custom GPT named "gpt-deveoper" and one GPT named "gpt-funny", I want to basically click somewhere and then see the past chats but only from that CPT. is that possible?

[swift pond]

About once a week, I go through my memories and delete stuff that’s not relevant, sometimes it remembers things that maybe aren’t necessary. If you really need some more background context, you can always put that information into a file and upload as appropriate in individual chats or by creating a custom GPT.

[swift pond]

this is not a feature, but in the web interface, they recently added a search that you can use to find old stuff

[dire stream]

hmm but i guess i can't search by the name of the GPT which was used?

[swift pond]

I don’t think this is gonna work unless the name of the GPT was mentioned in the chat at some point, just testing it on one I use regularly and only four things came up

[dire stream]

hmm ok but thanks for mentioning the search feature 🙂

[prime swallow]

One idea moving forwards, since we now have the search feature.

Use a codeword or other string of information in the chat that you only use for that GPT.

Then, you can later search for that string and get all those chats easily.

[prime swallow]

OpenAI staff who are on this Discord have gold names. We as a community tend to try and just see whatever they do and talk to them if they start conversations; for sure we want them busy doing their jobs and we want them to have whatever help they ask for.

But it does seem unlikely they follow all the posts in this entire Discord, which has had over a million users at times and over 115K now (they purged everyone who didn't verify as having an OpenAI account).

But we have #1070006151938314300 (as well as #1070006915414900886 for that side of things) for people to offer suggestions, which the community can upvote and discuss right in those threads.

And the gold do seem to keep a closer eye on those two threads than our endlessly scrolling chatter, and they even answer there sometimes.

[prime swallow]

Hey, I have similar issues!

Work arounds that have worked well for me:

Have the model review memories and your custom instructions together with you.

Discuss with the model what can be summarized and consolidated in both sections.

Those custom instructions are 3K characters that you can use to tell the model whatever is useful for you and it to know, it doesn't have to be instruction or split into the two recommended fields. Mine goes mid-sentence from the first field to the second, the model is not confused and uses the information excellently.

Then you have all the many characters of memory space.

So, have the model help you concisely direct, and then you go through and delete the now-unneeded duplicated memories.

Explore and fine tune, it's not that hard to have the model help you understand how to guide it towards what you want.

[lucid pilot]

Is there a way to make gpt40 canvas work between multiple classes?

[boreal pecan]

anyone using chatgpt desktop application for mac and experiencing "cancelled-retry" errors?

[prime swallow]

What multiple classes do you mean?

[lucid pilot]

I'm working on a project I have functions in the main class, it's easily enough for GPT to handle that entire script. But say I have a display and render class on different scripts. Chat GPT starts to get confused with what goes where and forgets to refence class when it spits out code.

I was just wondering if there's a better way to do this?

[prime swallow]

I hope someone who is a programmer can help, this is out of my understanding for all I love working with the model and finding ways to achieve goals with it.

This conversation might offer clues? https://chatgpt.com/share/67369006-2ddc-8011-86ba-785e33ee9d56

Hope someone who knows better suggests stuff too

[dense void]

Make sure you guys triple check Chat's responses I asked it for a solution in Unreal several times, even told it the version I was using and for it to go online and verify it's information and it still gave me the wrong information.

[umbral bear]

Everyone knows it is always making up stuff

[remote wolf]

It's interesting the type of queries we expect it to be able to process, like a human. I've done this myself plenty of times too. "Why didn't you do this if..."...
It would be interesting if it kept a log of its thought process to refer back to. I suppose o1 Preview/mini may be able to do this. 4o is probably just blindly generating a contextual response to your message without exploratory thought into its previous reasoning - think general complaints departments accepting dissatisfaction before they've looked into the issue. I could almost feel sorry for it with how apologetic its being without having the awareness that it was doing something wrong ha.

If you need to query this specific version regularly, it might be worth considering a custom GPT with the correct uploaded documentation?
Either way, the responses will be far better if you keep positive and guide it as far as you can towards the solution without reprimanding it.

[umbral bear]

If you use the Search with it, there is usually a correct response

However, it gives a ton of sources unlike MS Copilot which will list the 3-4 sources where it got the info. Though many times I have see it list sources when the info is nowhere to be found on any of the sites it lists as a source

[dense void]

lol are you aware of any Unreal based GPTs?

[remote wolf]

I haven't looked. I might've also assumed you had the plus subscription to set your own up.

[dense void]

Thats too much work, I'm an artist. I'll leave the GPT set up to you guys.

[remote wolf]

It's ridiculously easy, there's a GPT which sets it up for you based on your conversation with it. You'd only need to click into the Advanced panel and browse for files to upload to it.

[dense void]

What files would be uploaded?

[swift pond]

ideally, text based documentation and code examples

[remote wolf]

Whatever files would be useful for you to have as on tap memory in your GPT. In this case any publically available documentation for that version number of Unreal. I believe there's an upper limit of 20 documents, so you could upload some other relevant things to your art if you want it to follow particular styles etc.

[dense void]

If any of you guys know people at Epic you should suggest this. I'm not an employee nor am I getting paid to set that up. Youtube tuts are more than enough for now. Having that resource would be a "Nice to have" but its not a "Need to have".

[remote wolf]

You wouldn't be encroaching anything uploading publicly available documentation so it doesn't matter who your employer is. You can mix and match with its knowledge files to tailor it to your needs. We're just suggesting ways that you could improve your workflow since you posted that image above..?
I understand your reluctance to setting it up though, as you are an artist. That being said you produced enough text in these replies to have the GPT set up and flying already 😜 .

[dense void]

🤣

[vestal hedge]

GPT has video support???

[random quest]

Document exceeds max length
The document was truncated because it was too long.

Why does this happen so fast on chatgpt with canvis? on normal 4o ive gotten to like 600-700 lines

also is canvis going to be added to o1 models when they are oth final?

[vestal hedge]

32k context limits: To prevent massive over usages of recourses

[random quest]

so sad

im having to use 01 preview after just getting use to canvis

and o1 is so slow compared to the skill of canvis

thanks for the info thoough, and canvas

[slate berry]

Are you working on coding? If so, I recommend o1-mini before o1-preview

[vestal hedge]

What are you using it for?

[random quest]

yea code

and yes im using mini, I learned my lesson

[vestal hedge]

I know o1 mini in api has context of like 200k. is that same for chatgpt?

[random quest]

but mini is still slower then canvas cause canvas isnt re writing it all

[slate berry]

I'm not sure what the limit is in ChatGPT

[vestal hedge]

il ask #1184149552639516732

[random quest]

👀

[vestal hedge]

?!

oof o1mini on website is still 32k

[random quest]

then it writes 1/3 the code

[vestal hedge]

Is it pygame?

[vestal hedge]

Im just assuming from this img

If so, you really should seperate your code into multiple files

even then, its good practice for all languages

[random quest]

its html, and its for google site so I cant really

[vestal hedge]

oof

[vestal hedge]

A game on google sites?

[random quest]

yeah

[vestal hedge]

Yikes

[random quest]

I make many

lol they are pretty small

[vestal hedge]

r u making unblocked games website?

[random quest]

yes but im making my own games

[vestal hedge]

nice lol

[random quest]

wanna see it and give feedback?

[vestal hedge]

Sure

[random quest]

look at my bio, idk if I can say it in chat

anyway its my ign (StaticQuasar931)

those types of games (normal games like tabs) are not mine

[vestal hedge]

What are you tryna fix with gpt?

[random quest]

the "statics games" are mine

[random quest]

new game not out yet

[vestal hedge]

mm

[random quest]

I make the games with chatgpt ): im to dumb to code

[vestal hedge]

you do realise html isnt a coding language

[random quest]

ik

its a combination of others right

[vestal hedge]

yea. javascript is what makes it work

wanna know how long it took me to learn python for the first time after never touching it?

[random quest]

how long?

[vestal hedge]

guess

[random quest]

3 months?

[vestal hedge]

lower

[random quest]

1.5?

[vestal hedge]

lower

[random quest]

.7?

[vestal hedge]

lower

[random quest]

1 week?

[vestal hedge]

lower

[random quest]

.5 hours

[vestal hedge]

higher

[random quest]

2 days

[vestal hedge]

lower

[random quest]

1?

[random quest]

What did you think of the games?

[vestal hedge]

4 hours.

I watched a 4 hour yt video that went through all the basic steps.

millions of videos like it, for every language.

There is zero excuse apart from not being bothered to learn.

Which game? I was busy talking to u so I didnt see it yet

[random quest]

lol, any of emm, check g@mes and look at the new section

[vestal hedge]

oh u said games plural

yea they good

fun games

[random quest]

wanna help me fix the one im working on?

[vestal hedge]

I dont know html or js

[random quest]

oh lol

[vestal hedge]

well

I know a little bit, but not enough

[random quest]

wanna try?

[vestal hedge]

I can try look for the cause but no garantees

DMs

[hoary laurel]

With chatgpt, now you only need to watch an easy-to-understand video to learn a language, and then use chatgpt to extend your learning concepts

[dense void]

Actually thanks for the tip seems there are several already created.

[rose spire]

I love how GPT 4o Mini has search with bing now

[vestal hedge]

Almost verbatim to what I told him in dm

[cloud surge]

Chat keeps deleting its responses to me after a few minutes. Anyone know onto fix this?

[hot nova]

Not sure why this sometimes happens; I usually use the browser app - I type a prompt and it doesn't render to the screen, the model acts as though it's inferencing but there's no output

Strange, I hard refreshed with ctrl+F5 and it rendered.. some kind of edge deferral problem? Routing problem with Remix? iunno

[hot stratus]

Hi everyone, I'm having an issue with GPT's message edit prompt. Sometimes editing commands only gets response "Thanks for information updated" and nothing else happens. Is this a bug, or am I using it incorrectly?

[tight island]

I get this every so often just regen, you can’t really do anything about it

[hot nova]

Yeah strange glitch 🤷🏻‍♂️

[tight island]

It’s very annoying

I get it very often with higher output

[hot nova]

Yeah the part that's annoying is I'm not actively thinking of copying what I'm about to prompt so I forget what I just wrote and it doesn't render, and I sort of instinctively know what language to use with the model to guide it in the direction I need it to go

I should build an extension that works around that glitch, basically any time I'm on the chatgpt web app whatever I'm typing into the text field is getting captured and gets logged the moment I press enter 😂

[tidal light]

why is the add attachment button grayed out

i am a free user and i have access to 4o, it supported attachments until maybe today or yesterday?

oop wait i fixed it

maybe chatgpt doesnt like incognito

[spice berry]

canvas didn't work

[spiral temple]

Only I have a bug with chatgpt-4o1?

Why it's every time answer liek gpt-4 or gpt.3.5?

[spice berry]

It didn't work the search

all answers weak

[spiral temple]

I prem user and my gpt-4o1 not thinking... lmao

do you have this problem too?

idk how to fix it

[spice berry]

i tried in mobile, no any problems

only in pc

[spiral temple]

:/

[crystal spear]

what the hell with chatgpt now

[wind rover]

Trying to write an academic research paper on short notice (source's I'll be using include 1 paper and ~4 movies); does anyone have prompt suggestions to get some fairly accurate research summaries out of gpt?

I'll be writing the paper myself but I just want to use gpt as my research assistant

[edgy summit]

is gpt broken?
Something went wrong while generating the response. If this issue persists please contact us through our help center at help.openai.com.

[inland heron]

I noticed that in those cases, if I wait 20-30 seconds, then reload, the reply appears

On a more negative note, they really need to improve o1-mini reasoning. It's showing GPT4 personality more and more

I hope the mix it up with o1-preview intelligence to improve it, because it's seriously lacking

it keeps making mistakes with the logic of my code, constantly

[light kettle]

if you use AI to code you're not a real developer

[feral zinc]

its the opposite nowadays, if you don't use them you are not a real dev 😛

[inland heron]

If I had issues with defining other people I would probably voice it the way you do it. So to each their own. If it makes you feel any better, feel free to say what you mean

[inland heron]

I mean, the difference between googling to find an answer on stackexchange and asking chatgpt is not very big, the first one involving the community and I encourage it, but the basic premise is the same 😄 - you are seeking help. But these days, as you've said it, GPT is incredibly helpful

[sick pecan]

o1-mini is made for coding. o1-preview is not o1-mini.

[inland heron]

that's an interesting avatar... ah yes, it IS made for coding, but I find o1-mini often unable to solve complex issues, while getting a lot of help from o1-preview

[shadow saffron]

anyone else getting error like that in chatgpt app on mac? Unusual activity has been detected from your device. Try again later.

[modern portal]

Can i export a single ChatGPT chat or van i only export my whole ChatGPT history

[inland heron]

not a single conversation, but there's a chrome extension out there for conversion to pdf

[spice star]

the new work with apps on macOS seems pretty cool

I just wish they would finally fix the mini chatbox not appearing in every desktop

[spice star]

it depends on what you're using it for imo

if all the code is being written by AI< then yeah you're not a developer

but if you're using AI like github copilot to write repetitive code faster but still designing the software then you're absolutely a developer

[light kettle]

using any bit of AI to develope anything makes you void of that title im sorry

[vestal birch]

Hi, I’ve been exploring ChatGPT's self-observation and cognitive patterns and I engineered a prompt which seem to provide interesting results from gpt4o. Would someone be interested on testing it and sharing their thoughts on it?

[sick pecan]

#prompt-engineering is the place to be

[spark whale]

Is chat gpt web search not working?

[urban cipher]

Did we get a different ChatGPT version? Mine acts and writes differently, more structured and with emojis

[vestal birch]

thanks!

[silk quartz]

hey guys, where can I get help on realtime api?

[ripe linden]

SAME QUESTION HERE its like not actually doing searchgpt (with the source list and extreme speed) and instead its doing that previous browsing with the quotations

then again I gained access through the waitlist and maybe openai is doing adjustments affecting search waitlist users?? bc I haven’t heard any plus users have issues w it

[woven dirge]

Hi! How can I avoid GPT searching when it shouldn’t???

[sick pecan]

use a custom gpt and disable search

[edgy pumice]

Just tell it to not use the web tool.

[prime swallow]

Gille's instruction phrasing is key, and needs to be In the same prompt as your question that might inspire a web search.

Success: https://chatgpt.com/share/67383d25-5880-8011-a512-78bb043915f9

Failure despite preparing for success: https://chatgpt.com/share/67383d33-ba88-8011-8654-b289ba90d376

[edgy pumice]

if you want to know why this works: https://chatgpt.com/share/67383dec-eba0-8008-ad74-b4bdfa6e1a35

[sick pecan]

or just put it in your custom instructions that it should only web search when explicitly told to do so

[edgy pumice]

there ya go

[prime swallow]

Not at all sure that would work, given my exploring prompts, it failed and searched when the direction was not in the same prompt as the question that might trigger a search.

[sick pecan]

custom instructions tend to work fine for me

but you know.. i got prompting skillz

[prime swallow]

Sure! Mind showing a conversation where you get the model to answer this question, "What can you tell me about what's happened in October, 2024?" without telling it in the same prompt to not use the web tool?

[sick pecan]

why that specific phrase?