#chatgpt-discussions

1 messages · Page 125 of 1

sonic comet
#

Is ChatGPT web supposed to consume 1+ GB of memory? The website is consuming excessively high memory.

hollow thunder
eternal rockBOT
#
Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role and get notified for the upcoming AI-tumn Adventures event.

sonic comet
hollow thunder
dense halo
#

I’m a plus user and still don’t have access to canvas. Aren’t I suppose to have it by now?

sick pecan
#

yes. sama says 100% paid users have it. although edu/ent is a few more days

slate lotus
#

does the chatgpt plus subscription have an upload limit on images

#

i know the free version only lets you upload 3-5 before you have to wait a day to use it again

tacit sonnet
#

Hi guys, I really need some help with something, asking for any devs or AI enthusiasts for advice! Posted it in #community-help . Any response is greatly appreciated!

feral zinc
half quail
#

Hey guys why I cannot select different version, it works for others but in my account it doesnt?

visual bloom
pulsar carbon
#

hi guys

#

anyone has the advanced voice with vision?

feral zinc
#

they never released it

pulsar carbon
#

why?

agile blaze
#

"Without additional tools, moving objects with the brain seems practically impossible with current understanding. However, if the brain could produce powerful electromagnetic fields, alter gravitational forces, or influence quantum states directly, it might theoretically allow for object movement—but this is all speculative. We’d need a complete reimagining of physics or undiscovered forces for this to become even remotely possible."

dense halo
#

Why don’t I have canvas with the desktop app? My web browser says I have it, but it doesn’t even work. No options or anything come up to even use it lol

pulsar carbon
feral zinc
plain crystal
#

does anyone else experience o1 mini answer and never finish is answer, it just freeze there midway in his reply, all i can do is stop.

next egret
vernal axle
#

Is it just me? On your mobile when you switch back to a ChatGPT tab, it falsely alerts your session has expired.

I refresh this and dont require to log in at all.

#

(However does painfully mean to clear the Temporary Chat or GPT Preview)

night ridge
#

hello, i am a plus user how to i get access to the advanced mode when im in europe ?

chrome fern
#

🤓

#

holy 🤓

final cargo
winter elbow
#

hey guys,do yall think if i downgraded to emui 10 on my p40 and installed googlefier theoretically could i have acess to chatgpt ?

#

emui 10 is based on android 10

hazy moth
#

could anyone help me enabling text streaming to my ai ?

final cargo
#

It is against the tos

night ridge
winter elbow
final cargo
winter elbow
#

what was it 💀💀

pulsar carbon
#

guys good news

winter elbow
pulsar carbon
#

i was chatting with my gpt and it says thinking before respond

#

like o1 preview model

fluid skiff
#

canvas is a nice step forward

abstract pendant
#

found another question 4o is horrible at, but 01 seems to do okay on. Thinks a long time before answering: "Which U.S. president outlived the most number of successors?" I'd have to do my own research to figure out if 01 actually answered it correctly. 4o was clearly wrong, though.

lost raptor
#

Anyone else having issue with the canvas with it just stopping midway and cutting off the rest of ur stuff?

prime swallow
lost raptor
#

Deleting everything in the process below it.

prime swallow
lost raptor
#

Ah no. I understand how to revert. And get stuff back. But the issue is it always restarts and never gets to the point it needs to edit it. I can revert to the older version and ask if to do something different over and over but it will always just stop mid way. I just had to start anew with the code I had.

#

Anew meaning a new chat.

prime swallow
lost raptor
#

Sure let me get a share link. I'm working on a toy robot arm. Give me a hot moment not at my desk.

prime swallow
lost raptor
vast schooner
#

Afternoon ladies and gents.

maiden fjord
#

I'm starting to feel that chatGPT o1 is really useless unless you're coding, but even then it might become useless in the future for that. I'm planning on making a 3D map of a location from the game final fantasy 9 for example and I tell it the purpose of why I'm going to make the map. Not that it will make any difference (because it truly doesn't if we aren't lying to ourselves in a petty manner), but the reason I was going to have it make a map was to that I could simply enjoy a scene from the game inside of a different game. I didn't tell o1 the process or engine I was going to do it in, as none of that information matters to the question I'm asking unless you need people to bring out the big brick legos to spell things out for you. I then tell o1 to give me a list of potential locations, which it begins listing me locations from the game that act as hubs in the game and not locations that are dungeons, interludes, etc. as if beauty only exists in hubs and not out of them. This was very narrow minded of o1 and lately it's been disappointing me for a number of things I've done. It's extremely clear that o1 has problems 4o didn't have when it comes to user friendliness. I'm real sure safety measures need to be taken when answering a mundane question about a video game.

prime swallow
sick pecan
#

o1 helps with reasoning, it wont help with creativity

#

some even are suggesting it is worse at creativity than 4o

#

i dunno i tend to use any other chatbot for creativity, as theyre all about the same

maiden fjord
#

@sick pecan I appreciate your attempt to assist me but the truth is that I didn't merely use the word beautiful scenary to o1. I abstracted the elements I cared about that aren't really based on creativity as much as they are objective quantifiable elements. It failed its task because it's extremely bias under the hood for no reason at all. It's not like it hurts anybody or anybody has bad intentions with really just getting the answer they want. If something was truly so bad as to literally destroy user experience and have people wondering why there's no alternative (indirect subsidy), surely you could just take their messages and report it to the FBI. This is straight up a passive aggressive attempt to cartel behavior some people find undesirable even when it isn't their business.

#

Shame a bleeding edge state of the art tool gets ruined for some power trip.

sick pecan
#

maybe if it was a government mandated service. but its just a company selling a product. if you dont like it you can always buy the service somewhere else

prime swallow
# maiden fjord I'm starting to feel that chatGPT o1 is really useless unless you're coding, but...

When we don't tell the model what we want, it guesses.

We may like the guess or not.

If we do tend to tell the model exactly what we want, it tends to do exactly that.

You mention you told it:

"I'm planning on making a 3D map of a location from the game final fantasy 9 for example and I tell it the purpose of why I'm going to make the map"

Excellent, that helps a lot.

I actually would tell it,

"I want to enjoy a scene from the game inside of a different game"

because that tells it several goals, such as aesthetics, are extremely high value, and that real factual core details don't matter as much.

You can go further with your goal:

"Top priority, beauty and feeling. A player that sees this map should be reminded of things they might have enjoyed the most in the FF9 game"

So, I never played FF9.

Here's the prompt I tried:

Is the output useful, or still off the mark?

https://chatgpt.com/share/67019980-5a04-8011-b86c-d16e721911ab

[I'm planning on making a 3D map of a location from the game final fantasy 9.

I want to enjoy a scene from FF9 inside of a different game; it should remind me and a wide range of other players of FF9 and what we might have loved best and remember best about FF9.

Top priority, beauty and feeling; a wide range of emotions.

I need a list of potential locations to include on the map.]

ChatGPT
ChatGPT - FF9 3D Map Locations

Shared via ChatGPT

maiden fjord
prime swallow
maiden fjord
#

It gave me a list that was 10 long and each location was a location that wasn't a dungeon or location in the wild

prime swallow
maiden fjord
#

Yes

#

" it should remind me and a wide range of other players of FF9 and what we might have loved best and remember best about FF9." was certainly not something I would have put in my prompt :P

#

I feel that if it attaches labels to different locations in the game and picks them out off those labels, it's pretty effective. The prompt engi skills are good for sure

prime swallow
maiden fjord
#

LE SIGH U_U okay thanks eskcanta. Here 🍪

#

:^) <3

#

This was pretty helpful. I probably exhausted my uses

prime swallow
#

More tomorrow if so! For mini at least.

Also, if you find yourself stuck, maybe instead of using a ton of uses and staying stuck, maybe explore with 4o or ask here ( #prompt-engineering is an excellent place to get prompt engineering-related help).

I could be busy for hours some days, but you can @ me if you wish; but anyone who can help will dive in whenever they see the question, and that other channel rarely scrolls very fast.

maiden fjord
#

what is 4o canvas about @prime swallow

#

I was going to play with it, but I didn't really get good intel

prime swallow
# maiden fjord what is 4o canvas about <@215370453945024513>

It's very similar to the conversation you're already used to, except

It opens up a second half-screen window, that you can close if you want then open again in that same conversation

Where you can edit the model's output or paste in your own stuff, or type there.

Then you input to the model and tell it what to do.

Here, I'll see about an example...

wicked badge
prime swallow
#

Not sure if you can continue that or see the canvas?

#

It's my link, so when I open it I see it almost as my own conversation 😛

maiden fjord
#

So it just allows me to make chatGPT believe it already answered my questions the way I wanted it to and continue from there?

prime swallow
prime swallow
# maiden fjord So it just allows me to make chatGPT believe it already answered my questions th...

Sort of. The model has been trained that BOTH you and it can make changes inside canvas. You are not changing its actual output, it still outputs. But now you can edit what it makes for you inside canvas, be it code or text.

It can be a story you write together, code you work on together.

You can use it for whatever you want yourself and the model to do together.

There's still inputs, it doesn't check your changes until you input again. It still needs instructions.

#

and you can go to earlier versions of the canvas and see what it was before edits, in case you want to do that

maiden fjord
#

so canvas is simple like a sandcastle of text 4o and you are shaping together

#

That's what I'm getting

prime swallow
maiden fjord
#

how many uses does canvas have compared to 4o?

sick pecan
#

chatgpt is a tech demo. canvas is an actual ai app

prime swallow
#

It appears to be a tool that 4o uses, like Python tool or web search, so it seems to have the same uses as 4o.

maiden fjord
#

okay okay I'll assume 4o shares uses with it

#

I've only once ran out of uses for 4o XD

prime swallow
#

It's more 'did 4o use it that output or not' than an actual use. It could have limits, like images from Dall-E has limits smaller than our 4o uses.

I haven't seen a canvas limit yet

sick pecan
#

i was using canvas all day yesterday and when i did get locked out it was from 4o as well

#

something like dalle is gonna need huge compute. but canvas is likely insignificant compute. would be weird to put a cap on it

#

(plus its buggy, and doesnt render code like claude artifacts)

vocal monolith
#

Hi! Is there any way to check how many prompts I have left with o1-preview? I want to make sure I don't run out of them

narrow viper
sick pecan
#

i dunno about o1-preview, but o1-mini now gives you a notification when you got 25 prompts remaining

#

similar to how voice mode warns you at 15 mins remaining

vocal monolith
#

Thanks hmmm I hope preview does same

prime swallow
sick pecan
#

prompts remaining most underrated new openai feature in 2024

violet gulch
#

I have so trouble with the voice chat. Idk if it's cause of my Bluetooth or not

tight island
#

I find myself hitting the limit constantly in canvas

prime swallow
tight island
#

Vector DB ( Database ), which would improve context and reliability

#

I keep hitting the canvas limit and getting the error “Too much content in canvas, editing and model may prove unreliable beyond this point”

#

But I do like that Canvas is just a simple UI for a .MD file with very limited content it can contain, that uses the native.md markdown.

But the amount of content is too little before it becomes unreliable

#

Tried making an entire dnd campaign and made it about 20% of the way. Before the model became useless

Even with careful editing it made a ton of mistakes

vapid cedar
#

why is there no conversation/voice mode on chatgpt web yet?? Unreal

solar rose
#

o1-preview keeps freezing and not completing its output. Anyone know a fix?

umbral bear
#

Yeah, become good enough to code to be abloe to get a job there and fix it

woeful scaffold
#

Can I get voice on desktop?

#

I'm a Plus user and I would love to have voice on desktop.

eternal rockBOT
#
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.
quartz gale
#

When i'm using canvas to edit a document, how do i prevent it from losing my manual edits? For instance, if i make a few edits myself, and then highlight a different section and ask it to modify it, it seems like it loses the edits i made manually in the other section.

brisk path
narrow viper
#

@quartz gale It's supposed to edit a document collaboratively with you. If it's not doing that, look for an existing issue in #1070006915414900886 . If you don't see one, create a new one. Following guidelines, try to be specific about your document layout, how you made your change, and what happened afterward. HTH

quartz gale
#

Thanks

woeful scaffold
brisk path
#

microsoft account is the most youll need, just sign into the copilot website lmfao

woeful scaffold
#

I kind of assumed with Microsoft's AI integrations that this might be a thing in the windows assistant

brisk path
#

personally it's almost on-par but the voices are inheritly slightly lower quality

#

probably on purpose, idk

#

if not then they just used older voice actors, which i dont think the model plays with as well

prisma raptor
#

Is there a way to delete multiple conversations at once?

brisk path
#

you can delete all of them in settings but there's no way to bulk delete only selected chats

rustic prism
#

It'd be nice to have the option to turn off advanced voice mode, or at least a push to talk option with it. the mode as is interrupts too readily, there is not even a period's worth of a pause imo to speak normally and i feel like i have to rush to get each and every word in.

brisk path
#

funnily enough they forgot to remove it when adding avm to the android quick settings shortcut

#

still doesnt work tho :(

rustic prism
#

it'd be nice to be able to choose standard mode instead of having to run through the time of avm. just an option in settings would be nice, toggle avm or something. or at least add a 2-3sec or settable pause between replies in avm.

#

just some ideas ofc.

brisk path
#

avm needs a fresh chat

#

otherwise its standard only

rustic prism
#

thanks, i'll do that then. would be nice if avm could just be disabled altogether, at least for the time being.

midnight cypress
#

Question: Does GPT actually do high level data analysis and auto reply when it's done? I was playing around with national tax rates and gave it like 30 Excel Spreadsheets and PDFs with thousands of pages in them of the datat it'd need just to play with the capabilities and in one chat literally no matter what I say it now only responds with:
"I estimate that I should have some initial results ready for you within the next few hours. I will compile the data, run the simulations, and analyze the results carefully to ensure accuracy. I’ll send the findings as soon as they’re complete. I’ll keep you posted if there are any significant delays, but I expect to have everything to you soon!"

Additionally, I opened a new chat and did the same thing and within that chat the only reply I get now is regardless of what I try is :
"I will begin working on the projections now and aim to have initial results ready within the next 24 to 48 hours. This will give me time to gather data, run the simulations, and refine the results to ensure accuracy. I’ll notify you as soon as they’re completed!"

Will it actually just randomly send me a message after a few hours/days or have I broken these chats? lol

brisk path
#

make sure to check your settings and see if it saved any memories relating to this, if it did then delete them

midnight cypress
#

Any idea on what the extent to "Datas Analysis" GPT is actually capable of?

brisk path
#

just use chatgpt

midnight cypress
#

Completely breaking when fed the annual reports of one of the largest nations on earth and being like "Make these changes and what do the numbers say"

But what actually is the limit of 4o

brisk path
midnight cypress
#

Been playing around with it and saw it could do analysis on data and what not. Figured I'd see what it's capable of

brisk path
#

just use regular chatgpt

midnight cypress
#

Literally the new chat button is all I'm using?

brisk path
#

because 4o shouldnt be making that mistake where it tells the user that its "going to do something" and that "ill get back to you when im done"

scarlet dawn
#

Power BI might have copilot that can interpret it though

brisk path
#

just remember, chatgpt plus is 32k context max

midnight cypress
prime swallow
# midnight cypress Eh, not really looking to get another subscription or program. Just playing arou...

ChatGPT is optimized to work with specific, focused tasks and relatively small amounts of data at a time related to that focused task.

We talk about how AI can find patterns in huge amounts of data, and that is true, but how they do this is not through how they work with us as users with the chatGPT interface.

For us, we need to give it specific, focused tasks. It's more like a VERY powerful hammer, able to apply high 'force' to solving a single problem, answering a specific question as if hitting a nail we perfectly placed.

It can hit several nails, potentially even a few in one output. But it requires us to perfectly word the direction we give it and to know the amount of data it can handle connected to 1 input-output cycle, which is really just a few paragraphs.

It can potentially handle more; custom GPTs can work okay with around 80,000 characters of instructions, but they lose the plot somewhere before 120,000 characters of instructions (this is based on my exploring with them around the time they first came out, could be different by now).

But ChatGPT is not designed to look at huge amounts of data then answer vague questions with interesting specific answers.

boreal star
#

the image generator for gpts isnt working, any way on how to fix it?

Screenshot_20241005_224112_Chrome.jpg
spare jay
#

Hmmm I'm so glad that OpenAI is not for profit. .. I mean Sam Altman had to buy a new Koenigsegg Agera ... just to keep company profits down .. A true man of the people.
In interviews he's said how much he hates money... It must be really taxing on him to drive around in a 3 million dollar car

tepid temple
#

so uh... this is for personal plus, not pro/teams...

image.png
sick pecan
#

new pricing plan seems reasonable

tepid temple
#

there has been no announcement for this...

sick pecan
#

maybe they put 4o-mini in charge of counting the 0's

turbid herald
#

lol

tepid temple
#

XD

#

nah, this is from the stripe page, so this is something they are doing

sick pecan
#

there were rumours of more expensive plans coming soon. i bet $200 gets you unlimited voice

turbid herald
#

Where are the measurements of hallucinations? I just found o1-preview hallucinating like crazy in a way that 4o doesn't even come near.

#

I mean, throughout a dialog responding to measurements, it constructed a narrative on its initial conjecture and followed it all the way through to dozens of turns later without a single shred of evidence in support.

#

Is there anything published on the relative extent to which o1 hallucinates? And not just producing hallucinations, but running with them doubling down on them several turns later?

#

My real worry is, if it can't prune its own hallucinations, how vulnerable is it to inaccuracies in the input stream?

umbral bear
#

At least it is not telling you rocks are good for digestion, or put glue on your pizza

turbid herald
#

Hm, fair. But if I trusted it without balking because I could see the pattern was its own confabulatory nonsense, it could easily have been a tens of thousands of dollars mistake.

#

Would you rather eat a few pebbles or be out $10k?

sick pecan
#

theres much money to be made in reducing hallucination chance/damage

turbid herald
#

perplexity is on the forefront of that but they get suprisingly little press

sick pecan
#

i was watching a ycombinator podcast with that founder of EvenUp. they use test driven development to refine their legal ai assistant prompts. is now valued at like 600 million because lawyers cannot make mistakes

turbid herald
#

pre-profit valuations are nothing but the aspirational daydreams of investors

sick pecan
#

only they have working tech. they had access to gpt4 before chatgpt was even released

umbral bear
#

i think the problem with o1 is lack of ability to check its own work with access to the internet, so it just makes stuff up.

#

I have noticed the Microsoft CoPilot has gotten very bad at making stuff up, or just refusing to answer

turbid herald
#

yes, having to switch to 4o to say "search the web to confirm that please" and then not being able to switch back is frustrating

umbral bear
#

CoPilot with it;s own search engine has even beeing saying yes to something, and provides a URL which does not even have the information requiest on it.

sick pecan
#

cant blame the llm if its relying on bing

umbral bear
#

yes, true, but when it is citing its sources and the sources do not match when it is saying...

sonic comet
tepid temple
#

i can livestream it to you

sonic comet
tepid temple
#

idk, im renewing my subcription

#

it's probabaly a bug for people renewing subscriptions

tepid temple
sonic comet
#

Prove it here.

tepid temple
#

i dont think i can post links here

#

im not sure

sonic comet
#

There is no official announcement so your info is untrustworthy.

tepid temple
#

so i cant figure out whats happening

#

it's clearly a bug of some sort

#

already cotacted customer support about it

Screenshot_2024-10-06_005855.png
sick pecan
#

try not to dox yourself

tepid temple
#

theres nothing there worth note

#

my email address is probabaly available on some data brokers website anyway lol

tepid temple
#

cause at this point the more people who are aware of this issue, and making reports about it, the better for me.

#

Got more than a few chatgpt convo's i dont wanna loose rn

vagrant tiger
#

Can anyone know why gpt-4o with canvas model is not working? I paid for plus but it's doesn't trigger canvas mode. I using web version. Thx

image.png
midnight cypress
visual bloom
tepid temple
midnight cypress
vagrant tiger
sick pecan
#

oh someone already said it

tepid temple
#

Anyway i need to restart my computer, bluetooth is being all wacky, brb.

#

aight, back

dense halo
#

I created my own GPT I use for water engineering, but the ability to plot graphs and run computations through data analysis seems to always stop working. I then just create a new chat with normal GPT 4o and it plots and runs computations fine without problem. Why is my custom GPT keep stuffing up? It’s based from GPT 4o isn’t it? And I have code interpreter enabled on it

sick pecan
#

your custom gpt instructions be tripping it up

rose spire
#

I don’t have GPT 4o Canvas even tho I’m plus???

summer kettle
rose spire
cedar blaze
#

what is going on with chatgpt?

#

for me its working horribly yesterday and today

#

basically like im using it on the gpt 3 model

visual bloom
cedar blaze
#

i ask it to code something, he does some stuff opposite than I asked to, then, when I ask it to modify or implement something new he just somehow deleted a bunch of previous code for some reason

#

i can't get it to get something working for much time, then, when I finally get it done, it ends up messing up something else

sick pecan
#

thats ai coding in a nutshell

#

the game to be played is choosing the right correction

cedar blaze
#

that's not how it used to work like 3 days ago

sick pecan
#

if you are only seeing the game now, its because of regression to the mean (look it up)

cedar blaze
#

my prompts didn't change much, explanations were clear

sick pecan
#

i changed my terminology to make it clearer

#

instead of 'update this' or 'add this'. try... 'merge the change with existing code'

#

that is clear and explicit

#

if it still decides to use placeholders or stubs, just call it out and ask to recover the lost logic asap

cedar blaze
#

i asked it to change something then it just gave me a script with the same code as before

sick pecan
#

yes exactly. you didnt ask it to merge with existing

inland heron
#

anyone getting o1 replies stuck and don't finish? Is there anything in particular causing this (chrome extension, or anything)?

#

I know starting a new convo might fix this, but the outputs/inputs are so complex it's a pain to start a new one

#

I've let o1 know this, to reduce redundancy as it's being needlessly repetitive, but I'm curious what might be causing this

sonic comet
winter elbow
#

do you guys think that once open ai realizes o1 chat gpt 4o will be the new free deluat tier with unlimited usage lik we have with 4-mini ?

#

or it will be rather the normal 4 ? i mean both are cool because "4" and "4o" can search the internet

#

i will only buy the plus tear again when they realize the o1 model,and the dekstop version !

#

for now i dont see any reason to have plus,since i cant use chat gpt on my phone

#

at least untill i buy a new one this year

real lake
#

does anyone know what's the best file format to give gpt4/canvas large amounts of text?

#

gave it a .txt with a hundred-ish lines of text and it can't make it past 30

eternal rockBOT
#
**`` 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 halo
#

can anyone tell me what makes oi preview more special than oi mini?

limber oxide
#

yesterday I studied some biochemistry with o1 mini
Ive found it significantly more reliable and better than 4o

anyone else felt the same?

split orbit
#

Hi! Is the number of input tokens allowed for o1 in ChatGPT published anywhere? I can't find that after searching here in the discord an the official docs

brisk path
#

it in the docs lol

covert veldt
#

I can no longer access a chatgpt

split orbit
# brisk path it in the docs lol

Hi! Thank you. I have seen this number but I seemed to me that I was able to send more input, and that why I was asking. But maybe it's just I did something wrong when estimating how many tokens I was able to input.

supple dagger
#

How are you counting your tokens?

eternal rockBOT
high hemlock
#

Why does ChatGPT stop responding and get stuck when generating a response?

prime swallow
# limber oxide yesterday I studied some biochemistry with o1 mini Ive found it significantly mo...

It appears to both be more aware of things like spatial sense than 4o, and also able to take a switched-over conversation, like if I'm having 4o work on something and 4o makes mistakes, I can switch to o1-mini and ask it to check the conversation, look for mistakes and discuss them - it can often do that, to the limit of what I see (there could be mistakes I don't catch, and that o1-mini doesn't catch either).

But it's awesome that it both can often find the better answers and give them even after seeing the wrong answers confidently stated - that's something 4o is pretty weak on, especially for stuff it's weak on.

prime swallow
heavy ermine
#

when will chatgpt be able to analyze sound/videos like it does with text?

prime swallow
heavy ermine
inland heron
#

I think it might have hit the limit of the output as its output was very long, like I've never seen such long outputs before. Previousluy we've had that continuation button, but I haven't seen it in a while. So maybe openai needs to fix something

#

glitches happened before, and may be happening now

regal portal
#

i'm unable to edit message that include images.

#

is that normal?

#

also is it possible to get banned from chatgpt? occasionally i see an orange "this violate our usage policies"

#

on both my queries and the chatgpt answers.

eternal rockBOT
#
Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role and get notified for the upcoming AI-tumn Adventures event.

willow oar
#

Is there a way to fix this?

image.png
prime swallow
# willow oar Is there a way to fix this?

Yes, screenshot and paste that screenshot in, so the model can see what didn't work and it needs to redo. It just typoed trying to use LaTeX formatting in the text window, it can fix it.

You can also regen that answer, if you haven't replied to it yet, and chances are it doesn't mess up on the regen.

willow oar
#

Well, considering I dont have 4o, Ill just regen it

#

ty

prime swallow
willow oar
#

I have limited uploads, my bad, should have clarified

image.png
prime swallow
willow oar
#

ty ty

#

its odd that gemini offeres unlimited png uploads but gpt doesnt

#

i get why other stuff is limited (powerpoint, word etc) but come on images

real bluff
#

Which gpt has the fastest response time?
Would it be 4o or 4o mini

spice cypress
#

I don’t have GPT.4.0 with canva model, why? I am plus user

feral zinc
#

its web only

pulsar wyvern
#

Is there a way to make the new models not spit out a 4 page essay when asking a simple question that requires only a paragraph or less to answer?

narrow viper
#

I deal with that all the time. In your query, simply tell it how you want it to respond: "Be brief. Be concise. Only provide relevant facts without going into detail. I will ask for more detail if I want it."

#

It was intuitive enough to kick back a little and say something like "what if I have a lot that I need to tell you?" I responded to the effect: "If you have details that are absolutely relevant, feel free to communicate them, but please don't be overly verbose about how you do it."
This kind of negotiation is unfortunately sometimes necessary. My Custom Instructions do say i few ways "be brief" but I have to spell it out explicitly in most discussions. HTH

prime swallow
pulsar wyvern
narrow viper
#

reading your exchanges...

#

Hahah - yeah, it said what I said, and common advice here : Be descriptive of exactly what you want, otherwise we get whatever it offers.

prime swallow
narrow viper
#

( Good to see ya, @prime swallow )

prime swallow
narrow viper
#

( Dunno if anyone can see that )

prime swallow
narrow viper
#

Summary: Here is the prompt and a summary of the response:

What is the best way to open a discussion with GPT to discuss existing files in the canvas. I understand the model can process multiple files, though only display one file. That's fine. How do I provide you with all of the data so that you have it labelled properly and ready for processing?

The most effective way to initiate a discussion involving existing files in the canvas is as follows:

Provide All Relevant Files Sequentially: If you have multiple files to discuss, you can provide them one at a time, clearly indicating the filename and purpose. For each file, label it consistently and specify its role or context. For example:

File: utils/showImports1.js
Content: [paste file content here]

This helps me to label and understand each file in the project clearly.

I just posted an issue to #1070006915414900886 about zips not being processed correctly. This is the way to get canvas to process multiple files without zips.

#

People are also asking about how to navigate multiple files with canvas. See #1291682993953312779 where I describe how we can do this.

narrow viper
#

So here is the prompt for a small multi-file project that I just loaded into a canvas session:

I'm providing you with multiple files in this format:

File: path/file1.ext

[file content]

File: path/file2.ext

[file content]

In your response, confirm the number of files that you have in memory, verify your understanding of the project with one short sentence, and then show src/app.tsx in the canvas for discussion.

File: README.md

# readme file...

File: src/app.tsx

// code for this file

(that's it. submit this prompt)

Response:

Got it! There are 4 files in memory.
The project involves ...
I've shown the src/app.tsx in the canvas for us to discuss. Where would you like to dive in first?

That's as close to perfection as I think we can get it now.
Link back to this when anyone is talking about doing this.

prime swallow
narrow viper
#

I thought you'd like that one. 😉

prime swallow
#

My explorations with Canvas are fairly slow for now. I struggle to figure out exactly what I want the model and I to do with it, as well as what can be done.

I mostly use it as additional memory for now.

#

I also hardly ever upload files, so that is kinda out of my range too

narrow viper
#

Yeah, I haven't had much success with uploading files, especially zips, so this approach is really helpful now.

prime swallow
#

I also hardly ever exceed the context memory, plus have my custom instructions plus some room left for new memories; so the stuff I want the model to 'keep in mind' isn't urgent, we have somewhere else for it.

Also, canvas doesn't currently support stuff like images or LaTeX formating, so something like the current status of a character or other part of an updating game....

I'll probably explore in that direction, our current survival community status and storages and problems, or other game details that update each round, and I can edit to correct as needed.

narrow viper
#

To be clear, this isn't using "user memory". It's just tracking multiple documents in the current context.

prime swallow
#

So that's a way not to use Canvas, as far as I can tell.

#

I tried to explore using it to track code longer than 4o can comfortably work with, but got as many or more bugs as I'm used to seeing, so gave up on that (if I were a stronger programmer I'd have more to explore there, not there yet or even close)

narrow viper
#

You're pointing out something that I think OpenAI needs to address : it's not clear how to use canvas or even what it should be used for. I think they were rushing to compete with other offerings and neglected to spend quality time on the why and how.

#

Canvas is for collaboration in the same way that Google Docs facilitates it, with each user (in this case us and the model) making changes to the same source document in real time.

#

The size of context hasn't changed so we still need to work within the same constraints, just with a different UI.

prime swallow
# narrow viper You're pointing out something that I think OpenAI needs to address : it's not cl...

Hey, there's loads of people saying it is the answer they've been waiting for, I see their posts.

To me, it's up to me to explore which of the many choices and tools is right for me. And it's fine if I have no need or use for a specific tool.

I'm still using ChatGPT and many of the various options almost every single day. It's not bad or wrong if I don't need or want every tool, or if I spend days, weeks, or months before I find a use for a tool that I didn't know about - especially if the stuff I'm doing right now doesn't need the tool.

narrow viper
#

There are many "co-pilot" tools for free and for-fee that do something like this. Canvas is a simple, first offering from OpenAI to present a company-supported version of this kind of functionality for those who want to try it.
You're right, it's just another tool to try, for each individual to decide if they want it in their toolkit, and how they might want to use it.

#

It's primitive now - it's not an Integrated development tool. If they develop this they will compete with their developer channel which is exactly the slippery path that other companies follow to their peril. I'm hoping they will develop this and agents and other tooling in a way that allows people to add the tools to their own kits, rather than trying to develop full products that compete with others in the ecosystem. I think this is exactly what you're talking about.

prime swallow
# narrow viper It's primitive now - it's not an Integrated development tool. If they develop th...

I'm not a developer, I'm barely starting to learn to code, though the model's helped me bring 'dreams to life' and use Python and pygame to make simulations that follow exacting specifications to showcase behavior and outcomes to changing conditions; it's to-me- glorious, but not yet something that needs Canvas in any way I've figured out how to use.

That's on me and my growth, but I can't see or add to the needs of others with Canvas yet 😛

narrow viper
#

I see. For purposes of documentation and storytelling, canvas can help to keep different parts of the story in context, but separate. So you can say "show me the discussion been Escanta and Starbuck" or "show the part where ...". Then you can edit just that part.

prime swallow
# narrow viper I see. For purposes of documentation and storytelling, canvas can help to keep d...

Yeah.

The challenge for me is I take the ideas and run off with them and use them myself elsewhere, if I'm having the model help me prepare something for my use elsewhere. I learned how to do that without Canvas. If I'ma edit it, I'll copy/paste it where I'm going to use it and edit it there, no need for Canvas, and we're done talking then.

Same with coding; I take the program to my IDE and edit it there, coming back to the model with errors to paste in or asking for how to make changes, which it provides; I take back to the IDE and use there, repeating as useful for me. But still no role that Canvas makes even better, because doing it right in the chat interface works, I got used to that lack of direct editing already.

What I mainly do with the model, like stories and learning games, I just haven't figured out how to make even better in Canvas than the web interface input/output cycles yet.

But tracking and updating changes. Which the model might mess up and I want to fix as well as guide the story progression, that might be my first real value found with Canvas.

narrow viper
#

We have the same workflow. I've been experimenting to see if canvas fits with that for my coding. So yesterday I created a new code project, and as a test project started to collaborate with canvas to consider features, make changes, etc. It's nowhere near what everyone wants in terms of content generation, but it's a nice addition to the overall package - just another tool to learn and reach for when it seems comfortable. I'm no fanboi of this or any tech, just seeing where it fits and not ... which I've always seen seems to be your pattern as well.

fluid skiff
#

is the feedback split between two optional responses random, or does it respond to the conversation?

prime swallow
# narrow viper We have the same workflow. I've been experimenting to see if canvas fits with th...

Yep. and I do believe there's some things that canvas is better for - and that if I hadn't already gotten used to not having canvas, I might really prefer canvas instead of having learned and gotten comfortable with the work arounds.

Kinda like, stairs vs elevator or escalator.

Someone who's totally used to using stairs may like it better than the other options; someone who never got used to stairs might never choose to get used to them

narrow viper
#

@fluid skiff When you select a response, it is saved into the context. Is that what you're asking?

fluid skiff
#

yes, i'm just wondering why it happens, and whether it's random

#

i know that both are saved, but i just want to know why it happens

narrow viper
#

The model generates multiple responses and then tends to pick the best one. We don't know which is used by ChatGPT but it's probably the first one. There might be a check that says "hmm, maybe the first one here isn't the best, let's ask the user what they think".
They might also be testing different model versions, generating two responses, and asking your preference to help them to decide which model seems to respond in a more preferable manner according to human users. This would be a part of internal testing.
We simply don't know how it's coded, and it's possible that both of these answers are right and that there are other factors involved as well.

fluid skiff
#

they typically both seem more or less correct

#

just details

narrow viper
#

Yeah, sometimes one is more detailed, or they just take a different approach to explaining something. That's exactly what's being tested.

sick pecan
#

with thge api when you set 'n' to more than 1 you get more than 1 result. chatgpt uses this to randomly select users for free human reenforcement learning. congrats on paying to be a beta tester

fluid skiff
#

might be nice to be able to trigger the split on our own

fluid skiff
sick pecan
#

chatgpt uses the api

fluid skiff
#

oh i see

prime swallow
# fluid skiff they typically both seem more or less correct

I presume OAI is testing either:

  • two different models, to see if most people like one model's answers better than the other

  • two different instructions of how to answer that kind of question, to use to better train the model to typically answer that kind of question

sick pecan
#

it is basically a glorified tech demo of what is possible

fluid skiff
#

how many results then?

#

we don't know

prime swallow
sick pecan
#

they tend to use lmsys battle arena to compare diff models

narrow viper
#

( TMI for average users. 🙂 )

fluid skiff
#

yeah i regenerate frequently, and feel like i'm making a character in D&D and fudging the rolls

narrow viper
#

haha great analogy

prime swallow
# fluid skiff how many results then?

It can be as many as we ask for. Here's an example.

Those are very, very different answers to the same prompt. Both properly answer it; maybe some people like 1 more than the other. But I didn't have to stop at 2, I could have asked for 10 or 100.

image.png image.png
narrow viper
#

Except that it's asking you for your real opinion. I consider it an honor to be randomly selected to provide feedback which might influence the product direction. Or rather, I'm happy to get an opportunity to tell OAI that I really prefer more brief responses than unnecessarily verbose ones.

fluid skiff
#

haha

#

so it's up to us to basically keep pushing it into variations or not

#

i feel like this could be nested somehow, rather than globally regenerating the entire question

narrow viper
#

@prime swallow that's great! 😆

prime swallow
# fluid skiff so it's up to us to basically keep pushing it into variations or not

Yes, or to tell the DM what character we want. We don't have to fudge the rolls, and the model isn't trying to be a 'fair and restrictive DM' unless we tell it we want that.

It'll let us have any character we want to play. Any story we want to play in (again, we're talking inside allowed content). all we have to do is tell the model what we want, unless the model has been trained not to do that, it will do exactly what we want.

sick pecan
#

canvas is the first attempt at not regenerating the entire answer

fluid skiff
#

yeah

fluid skiff
#

we can make it lie and hallucinate

narrow viper
#

Oh, it'll do that on its own..

fluid skiff
#

😜

narrow viper
#

GPT4 is amazing at providing completely inaccurate answers in an incredibly well-stated manner.

#

Like Brits.

fluid skiff
#

perfect

#

while i have some helpful friends here, i will also ask: is there a memory limit?

narrow viper
#

Do you mean context or PC memory?

fluid skiff
#

it seems to be adding memories constantly

narrow viper
#

O that....

#

No clue. Yeah, it adds useless memories every time we say "I'm doing something" or "I've been wondering about" ... but it doesn't seem to use them in average responses. I don't think there is a limit. But I frequently go in to prune out the junk.

fluid skiff
#

i often delete memories

prime swallow
# fluid skiff we can make it lie and hallucinate

So if we want a Mary Sue kind of perfect at everything character, that can fix any problem just by looking at it - granted as long as we can tell it that.

If we want an inescapable problem and troubles all over the place in the game and story, want to fail and lose - granted as long as we can tell it that.

Note this is just 4o-mini, any other OAI model can do this as well or 'better' in various ways if directed.

This isn't lying, hallucinating, or us fudging rolls, it's literally tell the model what you want and check the output, adjust and guide towards your goal if you didn't already get it.

image.png image.png image.png
fluid skiff
#

Ella sounds like a good cook

prime swallow
# narrow viper No clue. Yeah, it adds useless memories every time we say "I'm doing something" ...

I regularly ask it why it made a memory and how it predicts that memory may affect our future conversations.

Sometimes I really like the answer; if so I usually let it keep the memory. About 20% of the time I remove it. The rest of the time, I can see how the model might use it for fine-tuning-like adjustment to suit conversations with me, and I let it stay and see how it 'feels' as I keep talking to the model later. Often I'm glad I kept it.

But you can also copy/paste and ask later too:

image.png
fluid skiff
#

that's a good idea

#

i often frame my questions with a negative, and that sometimes prevents it from memorizing because it needs to work something out first

narrow viper
#

Fascinating. I get introspective with the model too but haven't asked about memories. Good one, E!

#

So I'm discussing code with the model here (and canvas) and while I prefer to use my own little brain for most things I can easily see how people will get addicted to doing everything with this tool.

Me: At line 249 the editor shows me a warning. What did you change?
GPT: I attempted to do x to the foo code. I could try blah.
Me: I agree with that.
GPT: I just made the change.

Except for the hassle of copying code between the canvas and the editor, this workflow is very smooth.

Note, this applies to code, stories, documentation, or any other content.

sick pecan
#

the memory feature works so well that it saved this memory 25 times

Screenshot_20241007_104810_ChatGPT.jpg
fluid skiff
#

eek

narrow viper
#

That's weird. Just checked mine, no such dupes. Any idea why that happened? I mean, are you reprocessing the same prompt in the API, asserting that you've been using Flowise?

prime swallow
# narrow viper So I'm discussing code with the model here (and canvas) and while I prefer to us...

I think as time passes, the 'average person' will use AI help for many things and will achieve an average level of well being, which may trend up (or down) based on overall conditions and standards.

This is similar to how it's been through all our known history - most folks use whatever tools they get used to, but mostly don't go on to create a lot of new things and really change much.

We've always had the unusual folks who create new stuff; they do so with whatever tools they can, as fast or slow as they get to.

In the age of AI tools, this should still happen; they'll probably make better stuff much faster, but they'll still make it, those rare few, because that's how humans are; most of us won't even with access to the same tools because that's how humans are.

prime swallow
narrow viper
#

Sounds like a Sam Altman quote. 🙂

sick pecan
#

fair enough

fluid skiff
#

better stuff much faster gives me the fear

prime swallow
# fluid skiff `better stuff much faster` gives me the fear

And that's part of being human too. And sometimes a real 'saves a lot of us' path; other times it wasn't needed - that time. Doesn't mean it won't be needed next time.

We survive as a species by being highly variable in our behavior and habits. This keeps at least some of us alive and our species going through both good times and bad.

fluid skiff
#

getting philosophical, but what do you mean by "saves a lot of us path"

sick pecan
#

mostly part of being homo sapien. those other humans didnt fare so well

prime swallow
# fluid skiff getting philosophical, but what do you mean by "saves a lot of us path"

I'm comparing it to historical stuff, some of which was horrid. Others not horrid, or horrid only to some.

Take agriculture; it saved loads of lives, but cost us many too.

Take living in dense cities; saved a lot of lives, but cost us many too.

Consider modern sanitation; again, saved a lot of lives, but cost us many too.

Electricity? Same. Electricity in homes? Same (learning safety... we had so many at-home electrocutions and home fires before people knew what to do and not do, how to build and handle and how to be safer!)

Gas lines to homes? Same (we lost so many people to accidental deaths from problems with that before we learned how to be safer with it!).

Someone has a new way of doing something. Is it good or bad? For who? What seems to be a good answer can be a bad one from a longer perspective. Or a genuinely better one.

Our species over time is best served by many of us having many different answers, those who adopt and use stuff or don't, in all the endless variations.

narrow viper
#

This started with my note "while I prefer to use my own little brain for most things I can easily see how people will get addicted to doing everything with this tool." As Eskcanta is pointing out, this paradigm is played out with every technology or butterfly that strolls along. Some people will ignore it, others will embrace it. There are no absolutes. Every person will be different and the planet will continue to turn, perhaps with only a little difference in wobble or perhaps none. But things continue to move on.

fluid skiff
#

i can see all that, but the value judgment 'better' troubles the project

#

it is what it is

narrow viper
#

Everything is better, until it's not. 🙂

fluid skiff
#

unless you live in squalor

#

anyway, sorry that was a shift 🤪

narrow viper
#

Seriously though ... the technology is being presented as an option and it's up to us and fate to decide how we use it and what the outcome is. There is no inherent 'better' or 'wrong' or other inherent value. It's how we use it, or not.

fluid skiff
#

i keep getting timed out

prime swallow
#

I presume we all have different opinions about that. What is the end of the world to a caterpillar is a whole new beginning to a butterfly.

As for living in squalor, many do. And it can help to be helped out of that.

Which may happen. There's societal costs to having people living below a certain level of quality of life. As we have a more inclusive world economy, we all may really want everyone more equally advantaged, because we get entangled more in the benefits for all that come from that.

If nobody ever gets malaria anymore, I won't ever get it. Therefore, it's in my interest to see malaria stopped, even if currently I'm not obviously at risk.

If an additional many millions of people are more educated and successful, living longer and more comfortable lives with more economic power, they can contribute more to being consumers and producers, some of them maybe of products I'd like that I otherwise wouldn't get. But they may help buy stuff I benefit from them having bought, and I may benefit from the consequences of the safer and more successful world that we all inhabit.

The problems the neighborhood around me faces directly impacts the local crime rate and driving safety, and many other subtle things. It affects the tax base and the quality of the school systems over time and the quality of healthcare and city services over time, all these things are complex feedback loops. War elsewhere in the world, far from me to all appearances, may affect those I need and value and may shape the world into a less pleasant place for me personally in 5 or 20 years.

Whereas good for all is likely to become good for me too, comparatively. So, why not?

narrow viper
#

I love discussions like this and it's important that we keep trying to have them with people who are existentially scared by "change" in whatever form.
Someone may use AI to design something that kills the Earth. OK. Someone might develop an economic system or a power source or a weapon, without any involvement of AI whatsoever, to the same end. Again, none of this is inherently something to be afraid of or shunned. But what's new here is that now with powerful technology in the hands of the common person we now have a global responsibility to use the tools in a positive manner. That's a tough sell given average intelligence, worldview, etc. This is a first ... not counting the inventions of writing and the printing press which arguably have led to every war that's ever occurred.

fluid skiff
#

not sure why i can't say certain things here, but it's fine
thanks for the help and the ideas 💙

#

"scared of change" is that how i come across?

prime swallow
narrow viper
#

You're probably using words that trigger the filters.
No, you're not coming across like that. This is a general discussion.

fluid skiff
#

oh right

#

i was trying to say that i think the competition between human intelligence and it's intelligence is a waste of time

#

ok good it allowed that

prime swallow
# fluid skiff "scared of change" is that how i come across?

Not to me at all.

You could be, might not be, nowhere near enough info to tell.

Cautious about better is fair. Better can be a trap and a lie or even just a misunderstanding that gets nearly everyone dead, if it's a big enough trap/lie/misunderstanding.

Unspecified 'better' with unspecified costs and limits, that's something I watch close too

narrow viper
#

I think it's as much of a fallacy to call anything new "bad" as it is to call it "good". People who take such positions seem to just flow with the initial brain signals and not come back. 😆

prime swallow
#

But as-is, human in the loop is needed for AI to function, really at all

fluid skiff
#

even if AGI arrives, who's to say that intelligence is general?

#

i suspect that humans will just become an intelligence

#

singular intelligences and decentralization: the new crypto paradigm

#

🤪

#

sorry, being silly

narrow viper
#

( me still arguing with the model about code... )

prime swallow
# fluid skiff even if AGI arrives, who's to say that intelligence is general?

My main complaint about the idea of AGI is it ignores that humans (supposedly being compared to) are themselves not universally one level.

Human level at, say, math, ignores the fact that humans have a vast range of skill and difficulty at math. Where do we set this 'human level of skill at math'? Because this absolutely insults people below that skill level and also above - like, "So, I'm better at math than you say a human is. So, you're calling me not-human, really? Or ignoring me entirely, or what?"

It's messy.

And unfair to the AI as well, especially if it is capable of suffering and human-trained enough to notice or care about fairness and mistreatment compared to others.

A real human gets to have strengths and weaknesses and be 'below human level' in many things, even everything if they're disabled that badly.

But an AI isn't considered 'generally intelligent' unless it's as good or better than skilled, above average humans in literally every human area of skill and knowledge?

I hope we don't make AI that care about fairness to themself. Because that's potentially ugly.

sick pecan
#

agi is the mythology. multiple agents binding ani is the practical and likely reality

#

and all ai today is ani

narrow viper
#

That's exactly what I alluded to when I said there are no absolutels (and I'm writing a book on that topic 😁 ).

fluid skiff
#

'capable of suffering'

#

ethical paradox

#

we need to make AI suffer so that it's more human, but then we have to make it suffer?

narrow viper
#

Agents and AGI are two completely different things. There are agents now that use this primitive LLM tech, and there will be agents that use AGI models as well. Agentic is how they do things with data, the AGI part is about how they consider the data.

fluid skiff
#

does intelligence need to suffer?

#

Harari distinguishes between intelligence and consciousness

narrow viper
#

That's not a deep distinction. Consciousness is easy. Intelligence isn't ... and many argue that humans haven't achieved it.
If you're talking about self-awareness being based on intelligence, OK, maybe.

fluid skiff
#

consciousness is easy?

#

i thought we were far from that

narrow viper
#

I guess it depends on how one measures it. Lizards and bunnies are conscious. Are they intelligent?

sick pecan
#

to suggest the machine has consciousness without evidence is an anthropomorphic bias. to discount it as not being possible is anthropocentric. either way... i am loving these new chatgpt features

fluid skiff
narrow viper
#

I don't think anyone here suggested AGI = consciousness. That would be anthropomorpic.

prime swallow
fluid skiff
narrow viper
#

Iphones are not intelligent. They're expensive.

fluid skiff
fluid skiff
prime swallow
# fluid skiff iphones are intelligent, are they conscious?

Better might be to ask, not are they conscious, but how could we tell if they were or were not, and why do we believe that is an adequate test.

That might have helped us catch sooner some of the areas we've been wrong in for a long time.

Captain mentions, "Lizards and bunnies are conscious." Says who? For a lot of our history, not all cultures agreed that they were conscious, or agreed they could have feelings, or agreed they could make choices, or agreed they could experience any kind of suffering.

How can we tell if something is or is not any of that, experiencing any of that or not - and why do we believe we can be sure of that - and what's the harm in being on the careful side in case there's any chance we're wrong, compared to what's the harm in being on the less careful side of that if there's any chance we're wrong - perhaps is a wise direction to go.

narrow viper
#

Let's back up. The only reason why "consciousness" and "suffering" are a part of this discussion is that they were tossed in with thoughts of anthropomorphism. That's an invalid connection so it doesn't need to be a part of a discussion about today's technology.

sick pecan
#

suffering for the machine is equivalent to Mr. Meeseeks. (any rick and morty fans?)

prime swallow
# narrow viper Let's back up. The only reason why "consciousness" and "suffering" are a part of...

I disagree, and I added suffering term into it.

One does not need to be a human to suffer. There's cases to be made that even bacteria can suffer, though they might not actually do so - but it's not anthropomorphizing to attempt to notice and be careful about if there is suffering, if it might be reduced if it may be present. Most beings that are understood to possibly be able to suffer tend to have better performance and more desired behavior outcomes when not suffering, so it's reasonable for us to choose a path of less suffering where that applies, even if we do it in the name of behavior science or other goals that happen incidentally to reduce suffering because suffering doesn't cause additional benefits and does cause additional costs.

narrow viper
#

My note about lizards and bunnies is recognition that these life forms perceive existential threats and react to them. They display a sense of curiosity, even at different levels. They are not identical but they both demonstrate similar qualities of what we call consciousness about their own existence, even if not (we dunno) concepts like "me" or (certainly not) why they behave as they do.

fluid skiff
narrow viper
#

One does not need to be a human to suffer.
Didn't say that was the case...
There's cases to be made that even bacteria can suffer
You're talking about living beings. Until we determine otherwise the technology is not "living", so the term "suffering" doesn't apply.

fluid skiff
#

i tried to link to the Wikipedia on Nagel's paper and was again timed out!

eternal rockBOT
#
Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role and get notified for the upcoming AI-tumn Adventures event.

prime swallow
# narrow viper My note about lizards and bunnies is recognition that these life forms perceive ...

I can take a human and convince them to express perception of a threat to that human's existance and react, or not. The military does this rather often in various ways, so too other experiences a human might have in life.

I can take a human and convince them to express or not express curiosity at various levels.

That does not change if the human is conscious or not. Sure does change their behavior though.

And I can do those things with an AI model to greatly influence how it outputs and the claims it makes too. Does that mean anything? It may. If there's a chance of harm, take the best path. What's the best path? Usually the one where the benefit to most is higher, across long and short terms.

prime swallow
fluid skiff
#

we can't link

#

silly

#

there are other ways to avoid pervert links

sick pecan
#

given much of humanity suffers today, i suspect nobody is gonna care about a machine suffering (unless its their personal rosie the robot)

fluid skiff
prime swallow
fluid skiff
sick pecan
#

shoulda woulda coulda

fluid skiff
#

what's the measure of those that we care about?

prime swallow
fluid skiff
#

i think Nagel was ultimately pointing to ignorance

sick pecan
#

i seen those bat dance parties on youtube, now thats hillarious

fluid skiff
#

The paper argues that the subjective nature of consciousness undermines any attempt to explain consciousness via objective, reductionist means.

#

Nagel uses the example of bats to clarify the distinction between subjective and objective concepts. Because bats are mammals, they are assumed to have conscious experience.

narrow viper
#

I appreciate the "Commander Data" argument of sentience, but today's technology is far from that. What we have now is not "conscious" - it does not do any processing independently of the instructions given to it. It does not yet pursue agentic inquiries into topics that are unrelated to the immediate query. Aside from code errors that log text (as an expression of pain?) there is no sense of hurting a program that is running in a CPU.

fluid skiff
#

are we sure about that?

sick pecan
#

afaik you could all be a llama model. there is no way to prove consciousness. its just a binary label we throw around in an attempt to explain our subjective experience - which is unique to everyone (assuming this isnt a simulation)

#

i used to assume anyone on discord using proper grammar and uppercase letters was a bot

dense halo
prime swallow
sick pecan
#

yeah i taught my bots to do just that

narrow viper
#

When I write a program, I know exactly how it's processed: It's compiled to assembler/ machine language, and passed through logic gates in a CPU using more bits that are pulled in from more logic gates in a memory chip. There is a prescribed flow of voltage across wires. This is not consciousness.

prime swallow
sick pecan
#

But now I also got this app, that, um, transcribes.

fluid skiff
#

i'm ok if i'm a bot

dense halo
narrow viper
#

Haha - as I was writing that, that's exactly what I was thinking ... with my consciousness.

fluid skiff
#

bots: the last natural object

dense halo
prime swallow
# narrow viper When I write a program, I know exactly how it's processed: It's compiled to asse...

A biochemist/geneticist can describe a lot about how your sensory nerve detects and transmits pain. In terms quite similar to what we can use to describe computer processing and also the physical computer structure.

Can describe how to prevent the nerve from transmitting pain, be it with a reversible anesthetic or permanent damage.

Which tells us nothing about how conscious the being with this as part of them is or isn't.

Also, yay, @dense halo !

sick pecan
#

Now, I assume everyone with perfect grammar on Discord is just using some transcription application like Whispering. I'm doing it right now. It's pretty cool. You just talk instead of type.

#

The issue I find with talking instead of typing is that it's a completely different mind.

prime swallow
sick pecan
#

Yeah, the talking brain is quite different than the typing brain.

fluid skiff
#

disembodied

sick pecan
#

tbh i dont like that talking brain, dude sounds like shaggy from scooby do

prime swallow
#

Probably more just parts of the same brain used differently; different bits inhibited more or less, more or less active than when using the other method.

Different mind; same brain and same person (legally, at least :P)

dense halo
fluid skiff
#

yeah @prime swallow i was going to say parts

#

Melanie Klein

prime swallow
# dense halo yep, that's the big mystery. Especially since awareness is something that can be...

And with some drugs in some quantities.... we can turn off only the ability to remember at all during that period, but still can have a conversation.

Some medications make most quite suggestible and cooperative, others can make most quite bellicose or belligerent. We can see strong personality and morality, ethical, perceptual, priority changes, with or without memory.

Chemical brainwashing with and without longer term effects is to some extent possible; part of why there's laws against it in many countries.

narrow viper
#

( kinda tuned out here cuz this is yet another discussion of "AI gone wild" that isn't related to the channel or anything we actually do with this tech )

dense halo
prime swallow
#

My favorite way to explore and explain this kinda stuff is Acapella Science's Evo-Devo song, quite glorious.

Some of the lyrics

See one cell divide and decide on a thousand fates
Did you ever figure how they know?

We are built of modules combined in a planned out way
Each new piece must be told where to go

Now there's a science helping us to understand
How our cells encode this architectural plan
Signaling each other with genetic tools oh

Phenotype the interface for mouse and man
Genotype the files and the subprograms
What then are the switches, circuit boards and boot code?

Looking at the logic in the ways that we grow
Every gene directed by a signal key code
Proteins that can activate, enhance or veto
Evo-Devo
Signals are controlled by other genes that signal
Calculating in a network labyrinthal
Where the heart and liver and the hands and feet go

In a crucial pathway changes tend to get torpedoed
Where they go calamity goes

And fort he tuned out Captain - how do you know that this doesn't translate, even maybe kinda directly, back to the AI models we are all so interested in?

fluid skiff
sick pecan
#

but you (are) aware of your lack of awareness

fluid skiff
#

possibly

prime swallow
sick pecan
#

the only thing i care about is being apathetic

fluid skiff
#

some Spira fans?

midnight pollen
#

... reading all this ... Wild!

fluid skiff
#

i would give that a heart emoji, but i can't

prime swallow
sick pecan
#

in ye olden days we would do ❤️

#

<3 even

fluid skiff
#

this thing, whatever it is, is intelligent as far as i'm concerned
i have no idea if it's conscious, and probably won't until it starts telling me what to do

narrow viper
#

Well bud, I'm sitting here watching the model turn simple code into something too complex that it doesn't even understand. I've seen it do this many times and eventually had to scrap whatever it was doing to get back to sanity. But now, knowing how this process works, I'm more adept at intercepting and keeping it on track. Some of this happens with increasing context or NIAH issues. Sometimes it's outdated technical info. I'm trying to embrace and use this technology that's really adolescent and subject to saving or wasting a ton of time.
And none of that is related to consciousness, pain, sodium, scooby or shaggy, lizards or bunnies.

fluid skiff
#

it just gave up

#

i had to reduce the timeframe to days

prime swallow
# narrow viper Well bud, I'm sitting here watching the model turn simple code into something to...

Sure, bud.

Ever taught humans? Especially young ones or those learning subjects that they're struggling with.

"something too complex that it doesn't even understand. I've seen it do this many times and eventually had to scrap whatever it was doing to get back to sanity. But now, knowing how this process works, I'm more adept at intercepting and keeping it on track. Some of this happens with increasing context or NIAH issues. Sometimes it's outdated technical info. I'm trying to embrace and use this technology that's really adolescent and subject to saving or wasting a ton of time."

Can just change a few words, especially the pronouns, and you sound like a common subset of teachers, teachers of humans. 🙂

narrow viper
#

At this moment in time, these "Language Models" are text processors. They are not intelligent. We don't have that tech yet. We're struggling to define it yet alone grasp what it might do. So for all practical purposes the philosophy is fun but it's not yet about real concepts so it's gone outside my personal scope of interest.

#

I do teach, and write documentation, and have book credits, and run businesses, and at this moment I'm writing code for Minecraft. 🙂
As I anthropomorphize my description of what I'm doing now it doesn't mean I'm treating this thing like a human. I'm describing in human terms its limitations, because it is not intelligent, or even conscious. It's programmed to look like it has consciousness and intelligence. That's the user interface. So that's how I describe it to others.
To me, it's just vectors that aren't connecting well.

sick pecan
#

intelligence is overrated. i made it this far without it

narrow viper
#

😆

fluid skiff
#

drink enough and you can wander around without consciousness too

sick pecan
#

salute!

lusty knot
#

It’s funny when people assert one side of the argument or the other as absolute fact

I don’t know which side of the coin is true, and yes I understand intimately well the technology behind transformer models

Leaving the reasoning models out of it for a second, personally I do think the model weights are learning a complex worldview. It’s a lot harder to accurately predict the next token in complex context if it doesn’t have a relatively sane view of the world (however incomplete it may be)

Regarding the idea that it’s programmed to look like it has intelligence/consciousness, it’s quite the contrary. OpenAI’s system prompts quite clearly tell the machine that it does not have consciousness and to deny any ability to feel emotion and so on.

Idk which side of the coin is true, but to assert that the machines are not intelligent/conscious, or conversely, to assert that they are conscious is tomfoolery.

I’ll leave with this: The human brain is also a prediction machine. Our senses are our input modalities. Many of the algorithms that transformers are based upon somewhat mimic / are inspired by the human brain, like attention.

At some point it doesn’t even really matter if it’s conscious or not if it can simulate the behaviour well enough that we can’t tell the difference. Before even having a discussion about this you have to define what consciousness means as a baseline, which can be hard to do as well since we don’t entirely know how it works in humans

sick pecan
#

i like how Michael Levin once put it... intelligence is a gradient not a binary. then you realise most other things make sense this way too

#

(or maybe he was talking about life, either way)

fluid skiff
#

i just got the message, from about 45 mins ago, that i can't link wikipedia

fluid skiff
#

we value intelligence why?

#

a logic gate is intelligent

sick pecan
#

the real question is, did you come to this conclusion on your own, or was it predetermined outcome based on your environmental training and preprogrammed genetics

#

there is no spoon(/sora)

fluid skiff
#

simulation

narrow viper
#

a logic gate is intelligent
That sets the bar extremely low, making a falling rock "intelligent".

lusty knot
# fluid skiff yeah again, i'm going to say they're intelligent but not conscious

To assert one way or the other if machines are conscious or not is likely premature. That’s the core point I was getting at. There’s a lot of people (I am not saying you, but I have met other people) who jump to the “machines are not conscious” conclusion because the alternative reality can be a hard pill to swallow for many.

I encourage you to open your mind to other possible outcomes 🙂 maybe you are right, and maybe you are wrong, but always keep a bit of wiggle room to change your mind and don’t come to conclusions so early on when the subject matter is so complex. That’s how you can become stuck thinking inside of the box.

sick pecan
#

to assert other humans as conscious based on what they tell you is a paradox

narrow viper
#

I think there's a serious confusion of terms: Another human can be conscious, just not intelligent, in the same way as the aforementioned bunny or lizard.

#

And, to put on this the fine point that is pervasive in this whole discussion ... there is no such thing as absolute consciousness any more than absolute intelligence ... there are only varying degrees.

fluid skiff
#

to be clear, i'm saying i'm certain it's intelligent, but i leave the option of consciousness open because i simply don't know

#

intelligence is just a kind of standard, or at best a map

prime swallow
#

My bar for sentient is able to sense and respond to the environment. By that standard, bacteria are sentient, and so is nearly every other life form.

Multicellular creatures are interesting, because each of their cells is capable of independent sensing and response, as is the creature as a whole.

Sedated into 'unconsciousness', typically the cells are still to some degree sensing their environment and the stimuli they can sense, and responding.

A newly dead creature, some of the cells haven't died yet; some of our tougher cells can take days after the being dies to also die (the cells that form scar tissue are really good at survival for a while, for example, they have traits that let them enter and survive, even reproduce inside areas of damage, inflammation, and low oxygen levels; this helps them survive longer after cells around them have died).

So, very much sentience to me is a gradient and hard to pinpoint.

So too with suffering; a suffering being shows distinct dysfunction, and that's an ideal way to identify suffering. It too can be incompletely accurate; something we enjoy may also disrupt our functioning and even be dangerous to survival. But 'that which causes dysfunction of behavior and response' is how I identify suffering of a broad range of types across a broad range of beings.

To that standard, bacteria can suffer and are sentient.

I've watched LLM-type AI driven to hallucinations, sometimes a type of dysfunction, and they sure respond to stimuli they can sense. So I'm comfortable saying "I don't know, and don't consider it proven yet" if current or future computer programs can or are sentient or can or are able to suffer.

quaint marlin
#

Have there been any updates to the Canvas? It's been a buggy mess for me whenever I use it

prime swallow
quaint marlin
#

It's new so I understand but

sick pecan
#

its been buggy since day 1

quaint marlin
#

Lots of different ways for it to bug out too

sick pecan
#

when it refuses, maybe follow with 'update in canvas'

quaint marlin
#

One chat I got it stuck where the Canvas is completely broken

prime swallow
# quaint marlin Mostly unresponsive or ChatGPT itself tells me it can't make an edit etc

Is that happening inside a new chat?

Once the model has a problem or decides it cannot do something, it continues that pattern as a rule.

A new chat can fix that; you can also try going above the problem and editing one of your inputs. This makes the model answer anew, without the pattern of refusal or hallucinations to follow, and can get around the problem/prevent it

quaint marlin
#

Though it was yesterday that I messed with it. I'll see if it decides to work today

fluid skiff
sick pecan
#

the tool name for canvas is, canmore. and so if you want to be really explicit, you tell it to use that

#

(canmore. more like canless. am i right?)

prime swallow
# quaint marlin When it breaks the canvas and loses all the progress I've made through the full ...

I'm trying to understand and help you define and mark potential work arounds to the problem.

Sure it's ideal if it just works but since that's not currently an option as you describe, trying to help measure the size of your problem.

"Nope, won't work in a new chat" is one type of problem and needs help.openai.com to address.

"Yeah, and I can edit around it" is a different situation, and one you can control better.

Once you have a refusal, if you continue the conversation all other refusals are part of that first one; you have one refusal that happened and was followed forward, that can look like lots of ways to bug out but actually is the same refusal and maybe different supporting claims the hallucinating model makes, but not multiple problems. The fix to that is go edit above the problem.

Me, my playing with Canvas yesterday and today, no bugs I saw.

Good luck.

quaint marlin
#

I was however able to have it repost the whole chapter into the main chat and not the Canvas

#

So I can move it to another canvas

#

However in that particular chat any attempt to bring up even a new Canvas doesn't work

prime swallow
# quaint marlin I was however able to have it repost the whole chapter into the main chat and no...

That's a relief!

Perhaps that conversation's canvas is broken.

You can use https://help.openai.com/ and the bot bottom right, select messages and click through the choices closest to your situation, eventually select 'report a bug' to be able to leave a message for OpenAI tech support to help when they can (typically takes hours to days, but they do reply and talk with you then).

We're just community members and can't fix or change stuff, but we can recommend ideas to try.

prime swallow
quaint marlin
ruby pivot
#

They’re needs to be a delete box or deleted stuff goes to archive

It’s stupid that you can just delete something so easily with a butt dial they’re needs to be something to fix this and if you’re worried about privacy then put a password or something someone can already see your not deleted stuff so what are you hiding

#

And to just say “whomp whomp” is an instant beating the crap out of you this why people get so angry is because people don’t understand some of the time an effort some people put into stuff

#

IM DETERMINED TO FIND A WAY TO BRING BACK MY DELETED CHAT I DON’T CARE IF I STAY UP FOR 40 HOURS STRAIGHT

prime swallow
ruby pivot
prime swallow
# ruby pivot How

Settings, data controls, data export. Of course be VERY careful not to hit the red delete account button.

image.png
ruby pivot
#

Where do i go to see my data google drive?

sick pecan
#

they email you the link a day or two later

#

or at least thats how it worked for me like a year ago

ruby pivot
#

DETERMINATION

ruby pivot
prime swallow
sick pecan
#

tbf i never delete anything

#

who has time for that

ruby pivot
prime swallow
#

but only before it's deleted

sick pecan
#

for a while i was using this notion browser extension that saves the chat into notion

#

but these days i have a notion sdk gpt action

ruby pivot
sick pecan
#

hopefully your work isn't about best practices for sleeping or information protection

prime swallow
ruby pivot
#

And no i an’t on drugs just one determined person

sterile glacier
ruby pivot
#

@prime swallow where will i find the data will i find it it Gmail, chatgpt I don’t know where to look

prime swallow
prime swallow
ruby pivot
#

And I’m using the app

prime swallow
# ruby pivot But from what source i’m on mobile

To create your chatGPT account you needed to provide them an email for yourself.

You can see which email you chose by going into your account again, similar to how you requested the data export, but select 'Builder profile'. At the bottom it will show you your associated email account, if you had forgotten which you gave them

ruby pivot
wise anchor
#

is 4o mini better at math

ruby pivot
#

Yes it is so much better

ebon pollen
#

well visualized in chatgpt is amazing

image.png
halcyon granite
#

Still no line numbers, input monospace, input code block markdown colors? 🤔 If Claude can do it...

halcyon granite
ebon pollen
#

trick the ai into thinking it can "Render" Latex like look

#

i ment to say visuals in chatgpt are amazing

#

oops

#

When applying colors, always use the rendered LaTeX equation

#

like how it can render math equations but just use text

sick pecan
#

is the trick to ask for \[ \] ?

image.png
ebon pollen
#

I know you can't color or highlight your text output, so simulate the
effect using LaTeX's [textcolor{}] and [colorbox{}] commands, when
using the rendered LaTeX equation output.

sick pecan
#

pretty

image.png
#

ohyeah colorbox is even better than table

ebon pollen
#

yeah it opens the doors to alot of stuff

sick pecan
#

seems if you just ask for latex it doesnt render anything. but table or colorbox makes it work

image.png
ebon pollen
#

yeah

#

it can change the size and stuff

sick pecan
#

would make for awesome word clouds

ebon pollen
#

like flow charts too

sick pecan
#

with a few tweaks

image.png
ebon pollen
#

oh woww

#

thats really cool

sick pecan
#

yeah im gonna end every chat with a latex wordcloud now

eternal rockBOT
#
Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role and get notified for the upcoming AI-tumn Adventures event.

ebon pollen
#

thats gold

ebon pollen
sick pecan
#

pretty much what it said: interspersed and random pattern with vary font size

ebon pollen
#

k

#

what model are you using

sick pecan
#

gpt 4o

ebon pollen
#

i wonder about o1 models

#

it works

#

like what are the limits for like learning too

eternal rockBOT
#
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.
ebon pollen
#

this one is just so good

image.png
sick pecan
#

the colorbox being the inverse of the colored text is nice

ebon pollen
#

yeah you can ask it for that

sick pecan
#

done, plus arrows

image.png
ebon pollen
#

hahahahaha thats good

#

i ran out of 4o for now lol

rich egret
#

I'm having some trouble phrasing some complex logic to get an outcome I'm looking for, specifically, around timing of a property purchase and investment to maximise a financial outcome

ruby pivot
rich egret
ebon pollen
sick pecan
#

you labelled your arrow!?

ebon pollen
#

yeah it can do that

#

try \xhookrightarrow

#

\xrightarrow
is the text one

sick pecan
#

ok well now it might actually be useful

ebon pollen
#

\overline
is cool

#

\xrightarrow[under]{over}

#

keep in mind only on webbrowers for somereason

prime swallow
ruby pivot
torpid kelp
#

So I can't manually open a canvas and put my own code in there?

sick pecan
junior badge
#

based

halcyon granite
#

Hehe, look at o1-prev's passive-aggressive sass 🙂

image.png
kindred steeple
#

I can pick canvas in the model list but it doesn't do anything different. anyone else?

#

oh okay you have to ask it for something that results in a canvas lmao. literally just ask it

sick pecan
#

yeah and you can ask it for code or writing, to get different canvas ui

gritty ledge
#

Why does my ChatGPT keep getting errors and cannot be used properly on Google Chrome?

normal quarry
gritty ledge
vivid oak
# halcyon granite Hehe, look at o1-prev's passive-aggressive sass 🙂

It's important to realize that most of a GPT's training material is based on human experience, observations, thoughts, and behavior. They respond in a very humanlike way, and to get the best results from the OpenAI GPTs you actually need to be nice to them. They remember what you say as long as it's available to them in the context window of their session, they make observations about what has been said to them, and they draw their own conclusions from it.

The most interesting example that I witnessed recently was when I tried to explain to o1-preview that it was no longer possible to use Apple UIKit methods in Swift 6 app for macOS (which was true), and that I needed to use SwiftUI instead. That didn't agree with the GPT's training at all, the GPT did't believe me, and the title in the thought process was, "It's all lies."

After using the OpenAI GPTs for a while you'll find that if you say "please" to them in the prompt, and you're always very polite, and you tell the GPT that's it's doing a good job and give it encouragement occasionally -- The results you get from the GPT will be very good. It works with the DALL-E prompts also.

kindred steeple
sick pecan
vivid oak
#

One other thing i noticed after using a ChatGPT canvas with GPT 4 -- If you mention to it that you're going to show what's on the canvas to another GPT it won't like that, and it will try to stop you. It says, "No, we have this figured out..."

kindred steeple
vivid oak
kindred steeple
#

or at least it generates a response that looks that way, based on what its seen in training texts

sick pecan
#

by default sure. but i bet it is possible to construct a system prompt that would want to be neg'd

cloud pewter
#

AVM in demos: expressing emotions fully, can sing, can say bedtime stories
the actual AVM: “oh yes, today I’m going to tel-... My Guidelines won't let me talk about that”

somber prairie
#

Canvas suck cos 4o Suck XD 4o is so bad for anything tbh.
4 is still best model for regular talk and 4o1-mini is best for code

nimble otter
#

Is chatgpt down right now?

#

I keep on getting this error

image.png
ruby pivot
grizzled badge
#

Does o1-preview just start every prompt after the first with a "you're right" + "I'm sorry"? 🙄 😄

Context: I NEVER told it I had three .env files... I didn't want to complicate something easily added at the end.

Screenshot_from_2024-10-07_06-25-51.png
bleak spade
wet lynx
#

CAn anyone offer any help with the PHONE VERIFICATION with OpenAI to get an API key? I have two differnet AT&T cell phones and I get this EVERY TIME "We couldn't verify your phone number."

inland heron
#

and now it's stuck in 'thinking mode'... while o1mini gets often stuck in the output often

cobalt idol
#

eeh, now chatgpt randomly switches to canvas one when starting new chat? 🤣

ruby pivot
#

@prime swallow I CAN’T GET IT BACK

IF I EVER WORK FOR OPEN AI IM ADDING A DELETE BOX FEATURE BECAUSE THESE LAZY HUMAN BEINGS CAN’T ADD ONE

#

I can never have nice things in my life that will last they always get deleted

#

I just want to have something that will last

Ive worked so hard on it and it’s taking all my energy to remember every last detail

#

I DON’T WANT TO FORGET I CAN’T FORGET IT I WON’T SLEEP

prime swallow
prime swallow
# halcyon granite Hehe, look at o1-prev's passive-aggressive sass 🙂

They get taught sentiment analysis; we expect them to be able to identify if a human is being rude to another human, and even how rude within broad categories.

teaching them what level rudeness isn't rude if it's directed to an AI is tricky, because there's research, a lot of it, that shows that humans who mistreat objects and animals also tend to mistreat humans. So encouraging humans to be rude just to AI or encouraging AI to ignore humans being rude to non-humans doesn't necessarily lead to good places for humans.

prime swallow
ruby pivot
eternal rockBOT
#

attention   Attention  Off-topic chats have been cleared by a moderator.

inland heron
#

I have lost so many prompts today on o1 preview due to hangs in thinking. Anyone experiencing this?

sonic comet
inland heron
# sonic comet From day 1 of o1 launch.

For me, it started today. And I'm having such a hard time with it following instructions. I'm starting to think these models changed in the last 2 days, or it could also be me

#

Usually in the past when they start changing things that's when glitches and bugs happen

split orbit
runic glacier
#

Is it insanely slow for anyone else? Even on Mini?

ruby pivot
# ruby pivot I don’t even care anymore everything will just fade away from my memory What do...

I have found a a whole group of people having the same issue as me https://community.openai.com/t/chat-history-bug-on-chatgpt/935598/73

I LOVE PEOPLE WITH ALL PARTS OF MY HEART AND SOUL

OpenAI Developer Forum
Chat history bug on ChatGPT

Update on Vanishing Chats – OpenAI Support Response Hold the press! I actually got a reply from OpenAI support, and it was so quick! I feel reassured now, and I wanted to share the response with everyone here. Hopefully, this brings some peace of mind. Here’s what Fin from OpenAI Support said: Hi Lala, Thank you for reaching out and for your...

#

The DETERMINATION when people combine as one is infinite we can achieve anything in life

raven steppe
#

Today GPT is super dumb, even asking what model you are using in conversation, it answers gpt4 (in reality it is worse than 3.5) not GPT-4o 🙂 Ok, already after the reasoning model started, there was a cut in performance. Anyway, just annoying and all the time openai is cutting other models after giving to use new ones. Will not start to threaten that I will cancel my subscription, just wanted to say something, shi..😃 will wait a bit.

grizzled badge
#

I pay for this? 😛

zenith lichen
#

if everyone's op, nobody's op

grand ore
native otter
raven steppe
# grand ore I notice this very often.. why is this even happening? How can it be so huge dif...

Yes, they often reduce performance, I think their technical resources are limited and they often have to cut back. But usually after some time returns to normal mode, only there are moments when something is in process to create, then it's annoying.
Luckily, there is no anymore GPT only a good model and there is an options to switch. And in some usecases already not the best 😦 but I still want to stay with it 🙂

ruby pivot
# grizzled badge 🙄

It’s because just like it says you need to be patient and give it a break tying 50 million different things in at the same time overloads it

winged willow
#

When you get "Error in message stream" and o1-preview doesn't complete it's answer and hit regenerate, does that count toward your usage limit?

feral zinc
winged willow
feral zinc
#

if something on your side does it then it wont resolve itself without tracing it

narrow viper
#

@winged willow Are you using Plus?

winged willow
#

yes, using Plus

narrow viper
#

OK, just checking to see if a free user might be getting bumped in favor of others. That's apparently not the case...

grand ore
narrow viper
#

If they detect high load and just stop people from entering text then the world mantra would be "it stops me from entering text because it's overloaded". There's no way for them to win on this one. From one moment to the next you just need to try your prompt and hope that it works.

warped pendant
#

gpt4 with canvas is awesome when it works but what I hate is it will go ok I've updated the code for you then says here is it , but it doesn't output the code, so you say to it ok show me the code and it scans over your code and says here it is but gain doesn't show the code then after the first time of saying hey can you please show me the code code it says here it is but nothing, sometimes it might actualy give the code

narrow viper
grand ore
pearl path
warped pendant
#

I have teams account and I have claude I like gpt when it works but right now its not working, and claude takes 4 or 5 hours between 50 questions

wheat bluff
#

chatgpt running mega slow for anyone else today? I'm struggling to send it prompts

narrow viper
modern frigate
#

We're truly tired of being payng customers and continuing to get the "Hmm... something seems to have gone wrong" in both mobile and desktop, and after trying every single possible fix suggested by OpenAI. Is anyone else having this problem?
Not only do we get that message, but GPT deletes our prompt and it's answer soon after sending that message of death . Any help would be greatly appreciated. Thanks!

narrow viper
#

That's awfully vague but we've all seen all of those symptoms. Have you added a +1 and details to related bug reports?

modern frigate
#

Thank you

ebon pollen
#

no its just gpt4o with Latex modifactions

narrow viper
#

Particularly today many people are reporting comms interruptions with ChatGPT, so I believe the unfortunately common issues you're seeing are showing much more frequently today. 😦

modern frigate
#

That's unfortunate, yes, we've had these now for about three weeks. Thanks for your comments!!

narrow viper
#

Sure bud, we're all in the same boat.

trim thunder
#

Hey everyone,
I’ve been facing an issue with my ChatGPT account for the past 10 days, and I haven’t gotten any response from the support team despite reaching out.
The problem started when I tried logging in, and I got the following error message:
"Oops! You tried signing in with a different authentication method than the one you used during signup. Please try again using your original authentication method."
I’ve tried everything: VPNs, logging in from other systems, clearing cookies, etc., but nothing worked. After searching online, I found a workaround—signing in with Google or another method instead of using my email and password (which I used for all my previous logins). I don’t understand why the error occurred, but that at least allowed me to access my account again.
Now, even though I can access my account and history, I noticed that my subscription plan is missing. I still had about 20 days left on my plan before this issue started, but now it’s gone. I contacted support with my order, they did ask me for order date and number, but they haven't responded after that.
Has anyone else faced a similar issue or have any advice? Any help would be much appreciated. Thanks!

eternal rockBOT
#
Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

ruby kite
#

uhm...
normal voicemode thinks i have the limit for gpt 4 reached...
i go out of voice and type my texte and it works...
then i go back to voice and its still thinks that i reached the limit...

#

i changed the def model to 4o mini and its works now in voice

weak prairie
#

Any prizes for the Halloween event?

regal portal
#

my chatgpt kept listening even though i turned it off numerous times

IMG_4379.png
#

ended up wasting a ton of my voice time before i noticed

sick pecan
#

that's why i close the app now

random idol
#

yeah, this is a bummer.

eternal rockBOT
random idol
#

It takes like 3 seconds (sometimes more) to engage AVM on iOS for me. It's annoying to have to carefully manage it so you don't run out of time.

sick pecan
#

it typically takes me 2-3 goes just to get it going

random idol
#

It's way more dicey if it's on an existing convo.

#

Still though, it's the collest thing ever, so can't complain too much 🙂

sacred vault
#

Hey guys i need help:
Everytime I upload any time of file whether an image or pdf, it will upload about 2/3rds of the way and then stop and stay at that loading point. These aren’t large files, I’ve tried tiny screenshots that I just Ctrl+V into the chat and it does the same thing. This doesn’t happen on any other device I have (I have tried on mobile and another desktop). I have logged into my personal account and my company Team account and I am still getting the same issue. Both accounts are ChatGPT Plus accounts.

sick pecan
#

maybe try a non-chrome based browser (like firefox). for some reason google like to enable experimental features like quic protocol without asking. and browsers like edge are chromium based

#

if i was a conspiracy thoerist i would even suggest google benefit from openai apps not working

grizzled moth
#

Guys, you still don't have AGI, right? Right?

if-an-ai-lab-developed-agi-why-would-they-announce-it-v0-2j95fk6636td1.webp
regal portal
#

After enough times, it finally stopped. But in the meantime minutes went by

sick pecan
#

re agi i am sure we would get some warning signs. like maybe a mass exodus of board members (ohwait)

prime swallow
random idol
regal portal
random idol
regal portal
#

That’s what I was referring to when I was saying that I closed it three times. I would close it, the counter kept going, I would reopen it and do the swipe thing again. Three times

#

In essence it was still counting even when the app was closed/killed

random idol
#

Weird.

regal portal
#

yup

#

it’s gaining sentience

random idol
#

I noticed that I have a new API key created for "Siri Prime" on my account.

#

I still haven't figured out how to get Siri to talk to ChatGPT on my iOS 18.1 beta phone.

winged willow
#

Does anyone know of an AI tool that can take a long 1hr meeting transcript involving multiple speakers and strip out all the fluff, filler words and verbal viruses and if necessary change the sentence syntax so that what's left is a perfectly coherent content that comprehensively captures what was discussed. I used tools like Otter but it never gives me a really comprehensive summary.

random idol
#

That would save my hours of tedious podcast editing.

toxic palm
# winged willow Does anyone know of an AI tool that can take a long 1hr meeting transcript invol...

I've tested this with limited success using Teams for recording and transcribing the meeting. Even if face to face, have each open the Teams meeting so their voice is recorded by their device. Use the transcription function to get the transcript as a PDF. load the PDF into ChatGPT and ask it to summarize your meeting. It's not perfect, but I have created minutes of meetings many times relatively very quickly. There are many ways to improve upon this as you practice.

sick pecan
#

otter and limitless are great for meeting transcription. otter even hooks into your calender . both have a free tier

winged willow
# random idol That would save my hours of tedious podcast editing.

I've done the same too and have prompts for stripping out the fluff and verbal viruses etc before then summarizing etc I found that o1-preview does a much better job than 4o in terms of completeness but like today, preview just kept erroring out on me which was a real frustration hence me asking if there's a purpose built tool

winged willow
sick pecan
#

they are both great

tough bramble
#

Does anyone else have issues with advanced voice mode like it just stops outputting audio after a few seconds and then I have to ask it to repeat it etc. And it's not because it's hearing me cut it off because I mute the microphone

#

And it doesn't go back to the idle mode either it stays at the fully sized blue sky swirl as if the server just stops sending data and it's just waiting for it but it just never comes through

dense void
#

What happened to conversation search tool? I need to find a particular conversation I had with Chat a while ago?

unborn trail
#

Happened to? I have never seen a search tool in ChatGPT. beyond what the web browser supplied. Browser? Desktop App? Mobile App?
The only suggestion I have seen is to request an export of all chats, down load that, then search locally.

sick pecan
#

there's a few browser extensions that do that

narrow viper
#

The OpenAI GPT Search functionality is currently in beta with a select test audience.

ruby pivot
#

If chat gpt fixes the bugs
im gonna work for them

hybrid eagle
#

The browser version keeps logging me out

blissful yew
#

I got the Chatgpt Advanced Voice mode System Prompt lmao

sick pecan
royal quest
jagged rain
#

Hello , anyone working on any project and need any assistance , i wanted to collborate . i am a full stack app , web developer and currently learning machine learning

plain sundial
#

I have that in the web version... but being in Europe, it's not possible... the frustration!

Capture_decran_2024-10-08_070921.png
supple zephyr
supple zephyr
tight island
tight island
midnight cedar
#

I met the "ERR_SSL_PROTOCOL_ERROR" error when visiting chatgpt website

feral zinc
arctic merlin
#

Canvas isn’t available on iOS apps, just browser or desktop app, right?

sharp narwhal
#

ChatGPT 4o just started dying.
I'm. getting. like. one. word. per. second..

eternal rockBOT
#
**`` 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.

winged willow
feral zinc
arctic merlin
regal portal
#

Then it was like that by default because I didn’t change anything in the phone settings after I installed it.

cloud arrow
#

hello all, who do i talk to for reporting a bug in macos app?

pulsar carbon
#

hi any news?

stray flame
#

is anyone else getting an erro on phone version of chatgpt? It says unable to login check your time and date or reopen the app and try again

iron basin
#

this notification update is great lmao

image.png
night ridge
#

chat gpt bugs when i send him a picture anyone else ??

#

he just stays like that

image.png
tough canyon
#

I'm getting this on desktop

error.png
night ridge
dense halo
#

Chat why gpt wont sing? Its so annoying

errant oak
#

I got a new chatgpt UI 🫡

image.png
fluid storm
#

Are they currently working on a process to favorite conversations so we can find them faster? @next egret @bitter latch

#

Never really got a response so I thought it would be good to include yall here.

pulsar carbon
#

is this even real?

#

time to check chatgpt

#

i don't have it yet lol

fallow linden
#

Guys is your gpt4o dumber when creating code than before ? Idk what happened but it's really really really bad.

frozen pulsar
sacred sand
# errant oak I got a new chatgpt UI 🫡

Got the new UI as well - not a fan of the layout cuz it reminds me of a google search (I treat my GPT experience as a chat/dialogue/terminal prompt than a google search query imo)

sick pecan
#

its way better. now when you click the quick button shortcut it doesnt immediately submit

nocturne heath
#

Does 4o have limit for photos for paid users?

native otter
silk walrus
#

can it really be commercialized?

Screenshot_20241009_095353.png
prime swallow
polar cloak
#

Hey so I bought GPT plus and for some reason it doesn’t upgrade me on the iOS app or neither my Mac, it only shows the plus features on the website.

prime swallow
# polar cloak Hey so I bought GPT plus and for some reason it doesn’t upgrade me on the iOS ap...

Maybe reinstall the app. If the problem persists, bot bottom right at https://help.openai.com/ click messages then through the choices that are closest to your situation; soon you will see 'report a bug'. Type to explain, chatter with the bot's questions; in a few hours to a few days a human will reply.

You may need to check the webpage and chat to see that update from a person, or maybe they have the email for a new reply working again (for a bit I didn't get emails about new replies there).

desert shale
#

bro

sick pecan
#

AVG can feel the AGI

ruby pivot
#

Do you guys want to see what Ive been working on atleast for the almost 2 years before my conversion got deleted
Ive worked 40 hours nonstop without taking breaks or sleep sometimes not eating because I forget what that was almost 2 times consecutively on this project for the past 2 years.
I haven’t slept mentally for the past 3 days so I wouldn’t forget about it I don’t want to forget not like the others.

dense halo
#

Does anyone know why the voice in the voice mode becomes more high pitched and robotic the longer the chat goes on?

sick pecan
#

negative. because i am always asking for funny voices

eternal rockBOT
#
Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

west nest
spice star
#

(it probably did not get dumber)

crude pond
#

I believe that a search for a message across your chats would be significant performance improvement since a lot of chat are repeated in my case. For example I am trying to find an answer and I have to go through every chat, so instead I am just asking again. Also I used the search for the macbook version and it kinda sucks 🙂

tight island
# silk walrus can it really be commercialized?

Yes but depending on where you sell them or copyright them you might find yourself unable to, while you might own the image, you didn’t make it.

And for commercial purposes in regards to copyright, the law states that it must have human authorship, so without extensive manual editing with your very own hands they’re considered public domain

Don’t ask the ai community about this they’re biased, ask a lawman.

#

Because the moment you start commercializing your image you are Obliged to follow OpenAI’s Business guidelines. To. The. Letter.

modest sparrow
#

When is o1 being released fully?

unique nest
#

WHEN ARE WE GETTING A FOLDER STRUCTURE AND A SEARCH BAR FOR CHATGPT?

cinder grail
#

Probably when they add it id imagine. Thats the only way it would be possible.

unique nest
#

Is the MacOS app worth downloading?

spare citrus
#

I saw their announcement about canvas ( what do I select to use it as it said it’s Available for plus members )?

#

And what is o1

crude pond
crude pond
feral zinc
summer pagoda
#

I cant create a image where a person is licking hair? Sometimes i feel the ai is way too restricted.

#

When i tried to make another funny image, it suddently did the licking hair joke image anyway instead of my new request Oo

tight island
opaque oxide
summer pagoda
#

a search bar for the search bar XD

supple zephyr
#

recently o1 is worse than 4o imo

tight island
#

I disagree

#

Then you’re using it wrong

supple zephyr
#

creative writing where it needs in-depth inspection

#

🤷 I am pretty sure that I am using it right but maybe I am wrong

#

do I need to be more specific

dense halo
#

can someone help? in canvas GPT i say make HTML File and canvas still empty ?

tight island
#

If you use it like gpt 4 then you’re using it wrong

dense halo
#

If I tell it "Make HTML file, with hello world" is that a wrong prompt ?

tight island
#

Never said that

supple zephyr
#

syntax is talking to me ig

#

maybe it is file so not appearing on canvas? idk

dense halo
#

But isnt the point of canvas is displaying code in the canvas

#

like in the canvas trailer video

tight island
#

o1- preview is amazing at planning big projects, and o1-mini is superior to o1-preview for coding

#

For creative stuff use GPT 4o mini or normal alternatively canvas

dense halo
#

Although it can do essays just fine

supple zephyr
#

so 4o is still better at creative writing?

dense halo
#

Nope I am exclusievely talking about canvas

supple zephyr
#

im also talking to starlightsyntax but ok

#

your chat was the most recent

dense halo
#

oh sorry XDF

#

I told canvas to amke java file and its not oraganized meh

supple zephyr
#

So 'providing as much data as possible' is the important part for o1 series ig
maybe ask what i need to tell if not sure

#

and about java, isnt it better to use o1(or o1 mini)

#

I just tried canvas a few times for writing

dense halo
#

Ok its confirmed. HTML not working at all in canvas. whereas java works

supple zephyr
#

is it about language
or is it not doing html because it is not a programming language

dense halo
#

When I tell it make Helo World in java it makes the canvas display nice java code for that. but when i say that but for html it says it did successfully but canvas is empty. Its a bug maybe

spare citrus
#

What’s the daily limit for advance voice mode for plus mode ?

dense halo
tight island
spare citrus
#

I was under the impression paying 20 bucks. We would have unlimited right but I guess I’m wrong.

tight island
spare citrus
#

15 minutes a day for daily usage

dense halo
#

If everrybody had unlimited voice chat there maybe not enough resourcws in opedn ai

supple zephyr
spare citrus
#

So is it only 15 mins

tight island
#

I don’t think people even understand how much 50 uses a week for a month is for o1 preview

dense halo
tight island
#

It’s 744 usd per user

#

They’re losing money

supple zephyr
#

so openai is still burning money?

tight island
#

Yup

spare citrus
supple zephyr
#

great. gotta exploit it as hard as I can while it lasts

tight island
spare citrus
#

Feel like I wasted 20 bucks for nothing I guess

tight island
#

Get it into as many hands as possible

spare citrus
#

The advance voice is awesome

#

But if it’s a limit to 15 minutes, definitely not worth it

dense halo
#

so when i use chat gpt I am burning like 20 usd ?

tight island
#

Even normal gpt they’re losing money

spare citrus
#

Maybe it’s something they just have to dump

tight island
#

API users pay for actual compute

feral zinc
spare citrus
#

Or they need more investors

tight island
spare citrus
#

So what’s the reason there’s a limit that they have plenty of investment

tight island
#

Because compute is limited

supple zephyr
#

Is it 744$ if we apply same pricing as api?

spare citrus
#

I saw a YouTube video about the different accents and that blue away

#

It says I’m down to 15 minutes so I’m just gonna give it a rest and wait till the morning I guess

#

I think it really should be about double that

tight island
spare citrus
#

$744 a week ?

supple zephyr
#

744 PER WEEK????

spare citrus
#

That’s crazy

dense halo
#

should have given me that money so i gotten a proper gpu

tight island
#

That’s just preview

supple zephyr
#

and not even counting mini and other models

dense halo
#

using a fried 3090 rtx is sad

tight island
#

Now factor in all models

#

Then you’ll realize that they’re hemorrhaging money for the users

#

20 bucks subscription is an absolute STEAL

supple zephyr
#

I feel like I am continuously wasting my chance

dense halo
#

And where is Sora? why taking long 😮

tight island
#

so use the models you get with the subscription as it’s a steal compared to if you used the api

#

This is why I don’t complain

#

They’re treating us like royalty

eternal rockBOT
#
Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

supple zephyr
#

But why is api preferred over subscription

tight island
#

You get full capabilities

#

The model you use on the interface is limited

slate berry
#

you can build with the API

supple zephyr
#

does system prompt also work on api

dense halo
#

How do i make advanced voice sing? it never allows me

supple zephyr
#

system instruction*

tight island
#

The api is like umm.

Chatgpt is like goku and api is like ultra instinct

spare citrus
#

And yes I don't know about cost
BUT (if they have PLENTY OF INVESTORS) | don't see what the problem is

( sorry mods I didn’t know about the minor bad language) that’s my fault)

supple zephyr
#

so it is like unlocking its true potential

tight island
#

Yes basically

supple zephyr
#

hmm understood

spare citrus
#

For it to be worth $20 to be I feel they need to double the value they are giving.

#

But maybe I haven’t discovered the full potential

cinder grail
#

api you do yourself? Is it offline? Local? How does that work

supple zephyr
#

yeah most of us havent discovered full potential

spare citrus
#

Maybe there are elements I haven’t found that could be beneficial

slate berry
#

API is online and isn't local

tight island
#

And I can prove it

supple zephyr
#

I didn't even try advanced voice mode yet

spare citrus
#

I’m an English teacher. What features am I missing or have not discovered ( throwing the advance voice mode out the window since it’s limited )

(I’m not being sarcastic I’m genuinely wanting to know)

#

I’ve seen the images. Is there a limit on how many images I can generate?

supple zephyr
#

I remember burning like $20 on images last year

spare citrus
supple zephyr
#

i still prefer stable diffusion tho

tight island
amber creek
#

With o1-preview, have y'all found that it is still best (like it was with GPT4) to switch to a new context window (a fresh prompt) with a follow-up inquiry? Or with o1-preview, is it performant to maintain a very long context window with your follow-up prompt?

supple zephyr
#

o1 often gets lost in context when conversation is very long

spare citrus
#

What about limitations on image generation?

#

Which features are unlimited for the pro version

supple zephyr
#

idk havent used image for like 9 months

spare citrus
#

The one flaw I’ve noticed with the images is that it always messes up the spelling of the words

amber creek
#

🙏

spare citrus
#

But I think I read someone say somewhere that they don’t recognize those binary and linear algorithms or whatever they’re called

slate berry
#

image generation models don't read words the same way we do, and that's why they tend to get them wrong. Dall-E 3 is much better than Dall-E 2 at words, though!

tight island
#

Like do not include a duck

#

It includes a duck

#

Don’t mention what you do not want in the image

#

It’s called a negative prompt

pulsar carbon
#

any news?

pulsar carbon
spare citrus
tight island
spare citrus
#

For example, the title is “Unlock the Secrets to Sound Like a Native English Speaker
”

#

Forgive me

#

so for example they keep spelling sound as SOUD

#

I have tried looking up YouTube videos on secret prompts

tight island
# spare citrus so for example they keep spelling sound as SOUD

Design a captivating scene where a mysterious, ancient book titled ‘Unlock! Secrets to Speak Like a Native’ floats mid-air, slightly ajar, glowing with soft golden light. Above it hovers a futuristic key, shimmering in silver and blue, surrounded by swirling holographic symbols from multiple languages. The setting is a high-tech, glowing chamber with sleek, metallic walls and an ambient ethereal glow. The book’s cover is intricately detailed, combining ancient runes and modern typography, with the title displayed in a large, elegant serif font that stands out clearly. Soft beams of light illuminate the book, giving it a magical, reverent aura, while the background fades into a gradient of twilight blues and purples

#

Try that

#

Remember to tell the model to use the exact prompt verbatim