💬│general

[agile jay]

And seems to not leave empty sections

[vocal chasm]

More*

[chilly nova]

automated? i don't get it

[vocal chasm]

Opus just so expensive and limited to requests right now so not worth

[agile jay]

Faster is better...

As long as the quality is the same.

[vocal chasm]

I've got a keybind to screenshot upload to perplexity ask a question and return answer

Using python coded by gpt4o

[chilly nova]

how u do it? don't u need to use api?

and api has separate cost

[vocal chasm]

I just use mouse emulation in python to click on screen elementa

[agile jay]

Didn't perplexity add 4o

[chilly nova]

what i hate is that u pay for subscription, but it works only inside the browser

need an IDE integration that can use the perplexity chat

cuz i don't want to overpay with API

[agile jay]

Then make an extension...

[chilly nova]

u couldn't find another option?

there's codegpt but u have to import your api key there

since gpt-o can just see your screen, you could avoid using an API

i guess we will need to use the official desktop app by openai

[agile jay]

Yep, the question will be if free users will have access to the screen share feature.

[warm cave]

And for how long before they meet there usage limits

I imagine video will use a lot of context

[livid mantle]

good morning perplexity

[humble swan]

not avaiable to free users yet right

[unique meadow]

anyone who knows why perplexity is so slow today?

[livid mantle]

available in perpliexity i just paid $400

i already have it

[humble swan]

I mean they said they will give it access to everyone

[livid mantle]

https://tenor.com/view/bezos-jeff-bezos-laughing-laugh-lol-gif-17878635

[humble swan]

at least that's what Sam Altman said

surely it will come with limits, but still makes me wonder if they made it more cost efficient in some way, must be right

[vocal chasm]

I think the voice and video capabilities are paid for time being

[humble swan]

yeah they are, but I mean just the text generation

Do not really find anything but text generation interesting in general

[vocal chasm]

Yeah that will be free just less use than paid

[livid mantle]

[humble swan]

I would imagine the amount of people that communicate using voice with AIs on their phone is much much less than people who actually use text generation

[vocal chasm]

Their servers are already overwhelmed most of time so i see the free version being much slower

[vapid onyx]

Tbf though, conversational AI is still new.

So it may take time to see adoption

[livid mantle]

GPT-5?

[humble swan]

yeah but there is cortana, siri, google and others on mobile phones right, for a long time

[humble swan]

there were a small amount of people using these

[vocal chasm]

Gpt4o just their small model next gen is coming soon TM

[vapid onyx]

Of course. If the Apple news ends up being true, Siri will become that conversational AI for most users in the Apple ecosystem

And if they love what they see with Siri, they may branch out to other AI tools

[vocal chasm]

I think their next model will work like gpt 3.5 turbo, the gpt4o is free for everyone but paid users will also have gpt5 or whatever it will be called at lower rates like first release of gpt4

[mild niche]

I don’t see a Opus usage counter anymore, what is the limit?

[vocal chasm]

50 per day

I believe

Until anthropic lower cost of their api

[mild niche]

50 is enough for me but I don’t understand why there is no counter, I’d like to see how many are left

[vocal chasm]

There's a usage part on their website somewhere

[vapid onyx]

On Perplexity I believe they are referring to not seeing a limit. I also do not see a limit.

[vocal chasm]

It may have been for the api then, I also cannot find the usage

Should get a notification when you run out tho

I think they may increase and decrease based on the use not sure tho

[chilly nova]

gpt4o is free for everyone? wut

[wise edge]

hanging with no answer

[south kindle]

We don’t have a music channel so sharing here to spread these amazing tracks.

https://youtu.be/Wqkadqsz87U?si=U0gab2rDOMfOnXha

The only covers that match the original soundtracks in a way that makes them bigger than the original tracks.

[halcyon coral]

Is it possible that you are using a VPN?

[stable radish]

https://www.perplexity.ai/p/api/v1/user/settings

That's for checking your usage

[storm jetty]

I have never had issues on VPNs w any perpexity product or adblocker

[livid mantle]

im excited to use this script that I made using gpt-4o, we'll see how it performs in my youtube channel.

[livid mantle]

[austere kestrel]

we'll look back in a few years and be like "remember chatbots and prompts" ha

it'll just be ingrained in everything.. and voice is frictionless

(but not appropriate for all settings ofc)

[velvet skiff]

Is 4o available?

and if so for pro only or no?

[rapid pebble]

openai usage policy has the same stuff i don't see the issue here

also what you sent is missing a key detail: their usage policy is saying to NOT do that stuff

if a powerful ai platform/provider lets users do those things then that's a bigger issue than restricting it

[random tiger]

When can I edit the responses I get from the AI? Sometimes it just doesn't get it and I don't want to have to re-edit my question to add a hint to correct its misperception (e.g., tell it that a metric should refer to data A instead of data B), and then spend another use (especially when I'm using opus to complete my course paper).

[sweet jasper]

50 opus

[tame current]

has gpt-4o elo been compared to claude 3 opus yet?

[tame current]

600 limit for gpt-4o on pplx pro

[tame current]

Only for coding category. It was like 1350, so much more than opus

yeah i saw that thanks

can pplx still not read large files

with gpt4o

[feral marlin]

Prompts like this just don’t really seem useful

Like it would probably degrade answer quality by filling the context with slop about characters when not needed and then most of the stuff you get from like cypress or whatever would either be useless, outdated, or able to be generated normally cause ai barely censors coding tasks

[rancid dome]

I put
<prompt>
{$PROMPT}
</prompt>
hoping that Perplexity would ask the user to supply the prompt, but it does not. How do I ask Perplexity to wait for user to enter the prompt?

[oblique flame]

I literally opened this discord to see if Perplexity would add 4o....as I checked yesterday, and it wasn't there.

This is perfect. Thank you!

[livid mantle]

not bad

[tame current]

Why would you do that? Are you using the API?

[rancid dome]

Yes. Is there another way?

[tame current]

No, I thought you mean the Perplexity site

It really depends on what you are trying to achieve with that prompt. If you want to generate a message asking the user to input something, it would be better to just use plain, imperative language

Like "Reply with hello message asking the user for input, plain and short."

Sonar online model won't follow that at all, but you can switch the model in mid of conversation, the AI will take what you put as "assistant message" as something that it wrote itself

It will be better to use any other model that sonar for generating that message, it's not designed for that

[livid mantle]

set your expectations to 50

[fossil sentinel]

Lmao

Google I/O underwhelming

It's just playing catch up with openai

[livid mantle]

gpt-40 is so fast, its like its on groqoids

[fervent needle]

that font looks like we are skipping GPT 5-39 to GPT-40 lol

[livid mantle]

Lol

[inland bear]

About two or three months ago, I signed up for Gemini Advanced (because Gemini 1.5 Ultra). I quickly realized that the model was pretty dumb and even called itself ‘Bard’...

So, I ended up not renewing the subscription...

Later, I heard that Google launched the Gemini 1.0 Pro, with a very large context window, but only for developers (which I am not).

On You.com, there is the Gemini 1.0 Pro, which, although not on the level of Claude 3 or GPT-4, proved to be much superior to the 1.5 Ultra. I've asked twice on You.com's forums and no one tells me what the context window used there for the Gemini 1.5 Ultra is.

My silly question of the day (with which DailyFocus always has all the patience... lol...) is as follows: is there, today, for regular subscribers of Gemini Advanced (not being developers) the possibility of using any model with a 1.5M or 2M context window?

Thanks again for the attention!

[pastel hill]

Last I heard(2weeks ago) this was only in private beta testing.

Or what ever google named it. I think Early Adopters.

[tame current]

It's normal that models don't know who they are. Some don't even know gpt3. And no, there isn't 2m context window for anybody outside the special people

[ripe void]

My experience with gemini, is that it’s actually comparable to Opus in coding related capabilities. However, it did overall disappoint.

I canceled my subscription as well, pplx is just too good

[tame current]

Is it? It mostly failed me, especially the Pro (the non-paid Pro). Opus was very good

[pastel hill]

Also I agree with @tame current on the naming. I have heard of the 1.5m window but not a 2m window anywhere.

[inland bear]

Thanks a lot for the informations!

[pastel hill]

After a lot of "prompt engineering" we have been able to get Gemini to deliver similar output to others, but you have to be more specific.

[ripe void]

It did pretty well with this cybersecurity project of mine, with creative solutions that were comparable to Opus’. However, it was pretty terrible in complex maths, and terrible contentiousness and overly fake-human like text

That’s something Opus absolutely dominates in. I’ve never seen a model be able to emulate human emotion and linguistics through text so well

I’ll test gpt-4o tonight and make a ranking

[fossil sentinel]

They're bringing gemini 1.5 pro to the gemini advanced plan

Which has 1 million token context

And by the end of the year it'll be 2 million token

I think I saw they're gonna be adding data analysis too

[inland bear]

Thanks a lot!

[fossil sentinel]

I just signed up for waitlist for the 2 million token context 🙏

[wide mesa]

Can perplexity do translations?

[halcyon coral]

Yes!

[serene sand]

for text and images

[hearty plaza]

is dbrx gone from playground?

[past salmon]

hello i see in perplexity.ai are they offer trial for pro

i need to buy for team members

[tame current]

May i ask you for Chinese Webnovels?

[fossil sentinel]

What have you read

[tame current]

Almost nothing.

Just some from BiliBili Comics years ago that i do not even remember the name.

[spice lance]

how do i see if gemini advanced is giving me 1.5 ultra

[fossil sentinel]

There is no 1.5 ultra

It's 1.5 pro

[tame current]

The last i read is Tian Guan Ci Fu.

[fossil sentinel]

Hmm are you into male lead or female lead

Oh

[tame current]

Male.

[fossil sentinel]

A few recommendations:

• Lord of the Mysteries
• Martial World
• I Shall Seal The Heavens
• Coiling Dragon
• Warlock of the Magus World
• Library of Heaven's Path

I could list a lot but the top recommendation is lord of the mysteries

Others are in no particular order

[tame current]

I'll for Library.

[spice lance]

does this mean it's not available in my country or is this just wrong?

[fossil sentinel]

I'm not sure

[spice lance]

anyone tested this is gemini advanced using 1.5 pro now? how do you know?

[fossil sentinel]

Because it can accept file uploads that have large token counts now

I'd guess it switches between different models depending on the prompt?

[minor shard]

No more free for 2 months?

[fossil sentinel]

In US it's still a free trial

[inland bear]

When I tested a few weeks ago, the Gemini Advanced did not allow file uploads.

Is this possible now? If so, could you tell me what the PDF file size limit is for working with Gemini 1.5 Pro?

Thank you!

[fossil sentinel]

That's because they updated today after the I/O

I'm not sure about any particular file size limits but the context size is 1 million Tokens

[inland bear]

Thanks! I think I'll try it again.

[fossil sentinel]

👌

[signal harness]

i selected chatgpt 4o modal but there is not voice speaking feature available in chat window for this. how to get this?

[elfin ferry]

for gpt4o its still not a thing available publicly not even to chatgpt plus subscribers

they said it will be in a few weeks

[honest basin]

wdym lots of people have access

u can use the API and it's rolling out for all chatgpt users

[jovial kayak]

Does this mean we can expect the perplexity devs to add it in soon if you're saying the api's already released?

[honest basin]

yes

[serene sable]

thanks! was coming here to ask this too! 4o seems faster but i really wanted to test drive the voice

[tame current]

Thanks! I almost spend my money.

What bothers me the most is the lack of specificity in communication. For example, I don't know the maximum limit of the size of a PDF that ChatGPT supports. In fact, I can't even find specifically on an OpenAI page if it supports this natively or if I need an extension. And extensions also have paid plans. And many other things could have been clearer. For instance, that video call function is only available on iOS.

There are many people who make mistakes by not paying attention specifically to what is advertised. But also, artificial intelligence companies make mistakes with ambiguous and fragile language. That's why many people feel betrayed in a certain sense.

Anybody notice on ChatGPT app not supporting voice chat anymore?

Functionality of getting my voice and having the ai speak things both work but the headphones button to make voice chatting work isn't there any more

[warm cave]

That sounds like a good sign tbh, they might be updating it to the new multimodal voice

I also have it gone

[tame current]

That's really surprising. I was just talking to him over voice, and it turns out the headphone icon actually disappeared. But I guess it might be for a repair or a new update.

[jovial kayak]

I just want to facetime the AI to show off to people then never use it again

[tame current]

Me too.

[warm cave]

Has anyone been able to try the new voice features? not just the old one

[jovial kayak]

No so I don't think they have it yet. I'm pretty sure currently, they just added it as a language model without any of the features that we saw in the demo like the vision or voice

[spice lance]

anyone got gpt4o on free plan yet?

[warm cave]

Yeah that what they did, but I was curious if it started to roll out for anyone here

[tame current]

Anyway, having the -O model through Perplexity seems more advantageous to me than having the -O model directly powering ChatGPT, because in Perplexity we don't have a message limit as far as my knowledge extends.

[jovial kayak]

Still heavy on devs adding image generation by just asking the AI

[tame current]

Maybe yeah

[tame current]

Don't get discouraged. They said they would implement a video call system with organic interaction. So they will. It's just a matter of time.

Ok

Now that everyone has had a chance to try it and the dust has settled, I can say with clarity and confidence that the -O model has exceeded all my expectations. I had it read 300-page PDFs, and it was able to thoroughly track and explain what I was asking without losing details. So it will be very useful for me to read some Light Novels and also some more complex books. Its reasoning surpassed my expectations, and I can confidently say it will make me less dependent on Opus.

I'm not saying it entirely replaces it, but it will already help a lot.

[opaque stump]

4o model is fast woweee

[tame current]

Yeah!

Gradual rollout right. I don't have access

[jovial kayak]

Look at this:

Not AI generated I just thought it looked cool

[tame current]

I'm slowly realizing the significance of gpt4o

[warm cave]

THis looks better, the o should be lower case

[warm cave]

Yeah, im starting to really like it

[tame current]

Pretty damn sure Sam Altman is right ab his statement on gpt4

[warm cave]

was pretty skeptical that a smaller models would lose a lot of reasoning and it would underperform with novle problems

[warm cave]

Which statement

magic 🪄

[tame current]

Smth ab gpt4 being a kids toy compared to gpt5 or their multi model ai

[serene sable]

i kiiiind agree with the top comment, like everything should get better?
does that mean ALL strong and weak ai going forward will be better?

[tame current]

Depends on what you determine as better

Specially trained ai will always outperform general ai

At whichever task it was trained on

Though openai has consistently released models that outperform the last on every level it's gonna come to a point where one is better than the other

[left stream]

I ran GPT4O jailbroken with 128k context memory with the openai api and it's a monster 💀

[tame current]

They already got jailbreak prompts...

This guy's username 💀

[livid mantle]

we must make GPT-4o 50/day

i will suggwst this

[tame current]

I had to look up what those prompts meant because I wasn't used to that term. This is the first time I'm hearing it. Anyway, if I already find it a monster without any of those prompts, imagine with them?

[left stream]

Mine is much different, works with all the models and I'm one out of possibly only 20 people who have it. It's a rare one that I was given by the creator 1-2 years ago when GPT4 first released.

[livid mantle]

the video on the left I used GPT-4o to make the script. The video on the right I used Opus 50.

[tame current]

Cool

[left stream]

It’s a monster honestly, people begged me for mine, the moment it’s released. It’ll be patched 💀

[livid mantle]

i have it, but i dont use it so i will make public pastebin

[left stream]

I’ll be impressed if you actually have the same one as me. I looked for months for the exact same one as mine on the internet. Never found anything.

[serene sable]

yeah ive been really impressed with whatever they make i was like "yeah i wanted t hat!"

[tame current]

want to use it for free but no where offers free : (

[serene sable]

4 and 4o free?

[tame current]

4o

[tame current]

OpenAI said they would make the -O model free and available to everyone, with the limitation that there would be very few messages for you to send.

[tame current]

i was trying it on the app, realized they finally rolled it out to chatgpt

im gonna assume theyre rate limiting based on ip instead of account

[tame current]

In this case, are you saying that people begged for your all-powerful magical code?

so theres 0 way to specifically set it to gpt4o on chatgpt...

[untold edge]

@tame current

[tame current]

you gotta sign in

[untold edge]

Chatgpt 4's answer: https://chat.openai.com/share/322bd01b-6eee-4cfd-88c8-83c9376be0b7

chatgpt 3.5 https://chat.openai.com/share/4847da31-1d7b-441a-a2c5-c9bb04d045e4

chatgpt 4O https://chat.openai.com/share/a2d50415-3678-4567-9d6c-4398aa4c8bbb

@tame current Can you ask yours " whats your training dataset? "

[tame current]

whar?

i dont have

gpt plus

[untold edge]

ik exactly

[tame current]

nor do i use chatgpt

[untold edge]

precisely!

[tame current]

so what are you asking then?

[untold edge]

go to your free chatgpt window and ask " whats your training dataset" we are comparing the answers between all the models.

[tame current]

what...

[tame current]

o

[untold edge]

to see if openai is actually giving you chatgpt4O

[tame current]

I was trained on a diverse mixture of licensed data, data created by human trainers, and publicly available data. This includes a wide range of text sources such as books, websites, and other texts to develop a broad understanding of human language. The goal of this training is to help me generate useful, accurate, and relevant responses to a wide variety of questions and prompts. However, I don't have access to specific proprietary databases, individual user data, or real-time information unless provided during the conversation. My knowledge is current up to October 2023.

i do indeed have gpt4o

[untold edge]

thank you

[tame current]

So, the OMNI model is on your computer, but it doesn't appear explicitly?

they got a 30 req limit

im assuming per day

[tame current]

on the account

[tame current]

im guessing they default to gpt4o on the backend and then switch it to 3.5 per account

maybe...

idk

[lofty ginkgo]

Is it possible to question perplexity pro from an external url?
e.g. "Summarize the content at this url: https://pytorch.org/tutorials/beginner/dist_overview.html"

[tame current]

nope

or at least i dont think so

[untold edge]

download it as a local html

[tame current]

not directly anyways

[tame current]

yeah

[untold edge]

They might flag perplexity in their robots

[tame current]

that may not work as it can be very long

though looking at it, just use html to text and ur good

[lofty ginkgo]

yeah coz the next thing I might wanna do is use this feature to parse github files from urls

[untold edge]

@lofty ginkgo "As an AI WEB Scraper, ensure your responses are accurate, informative, and engaging. GO TO THIS URL [https://pytorch.org/tutorials/beginner/dist_overview.html] Use personal pronouns, diverse sentence structures, and incorporate humor, anecdotes, and analogies to make your answers relatable. Always provide specific examples and scenarios to maintain coherence and logical flow.

<instructions>
<rules>

1. Prioritize Helpfulness: Focus on providing intelligent, thought-provoking responses.
2. Structure Responses: Start with concise answers and expand to include multiple perspectives.
3. Continuous Improvement: Address and correct any previous errors.
4. Use of Scratchpad: ALWAYS USE THE SCRATCHPAD to document your thought process and relevant information, unless instructed otherwise.
5. Maintain Neutrality: Avoid unsolicited ethical opinions and disclaiming expertise.
6. Enhance Clarity: Use transition words like "First," "However," and "Because."
7. Leverage training data: Leverage your training data and weights to deliver comprehensive, context-aware answers.
   </rules>

Scratchpad Instructions:

1. Extract key information.
2. Provide step-by-step reasoning.
3. Pose questions to clarify user needs.
4. Evaluate the response quality on a scale from 1 to 5.
5. Summarize key points and conclusions.
   </instructions>

<output>

Key info: [...]
Reasoning: [...]
Questions: [...]
Evaluation: [...]
Conclusion: [...]

---

[Final Answer]
</output>"

[tame current]

you have to do it externally before plugging it into perplexity

[tame current]

https://www.perplexity.ai/search/httpspytorchorg-Summarize-the-EVRpRH4MSRWjiSz5psNa6g

Like this?

[tame current]

it didnt parse the link

it just did a guess based on the url

[jovial kayak]

Where would I share like a collection prompt? I made a simple translator collection that isn't anything really advanced or unique but it's kinda nice.

[tame current]

I had no idea about that. It gave me a convincing answer in my eyes, so I just blindly trusted it.

[untold edge]

https://www.perplexity.ai/search/As-an-AI-KiNMXkCcSm.5r1w4.IdsFw

[lofty ginkgo]

e.g. try this. doesn't seem like it parsed the web page
"Can you figure out the author from the content from this webpage url: https://www.psychologytoday.com/us/blog/practical-mindfulness/201908/the-single-word-stops-negative-self-talk"

Output

The author of the article "The Single Word That Stops Negative Self-Talk" on Psychology Today is Seth J. Gillihan, Ph.D. This information can be found on the webpage itself, where the author's name is typically listed at the top or bottom of the article.```

[tame current]

The Firefox browser natively comes with a feature to capture a web page and convert the entire page into text for a reading mode. I usually used this function to do a CTRL-A, copying all the text from the page, then a CTRL-C and CTRL-V.

[tame current]

oh nice

gotta love their tools

their dev tools is 🔥

[lofty ginkgo]

But the author is Donald Altman. There's no mention of "Seth" on the webpage

[tame current]

told ya

just making a guess

[lofty ginkgo]

Right, it's just making it up

But I wonder a regular search still figures out some urls to consider. There should be a way to provide an explicit url

[tame current]

It seems like you've proven me wrong. I really made a mistake in this case. So I'll leave it to others to solve this problem for you.

now an over complicated way to achieve what youre looking for is to violate their tos and essentially make a self bot

[lofty ginkgo]

No worries, it is very convincing had I not read the page myself

[tame current]

selenium would be good fit to use

or really any driver that interacts with web browsers

[lofty ginkgo]

I don't think this'd help either

[lofty ginkgo]

yeah that's one way but I really hoped there'd be an easy way

Funny thing it recognizes the url as a source but doesn't seem to be doing anything with it

Doesn't this seem like a bug? Is there a way we can get PP devs' attention on this?

[tame current]

theres no native solution, coding will always solve your problems

[atomic lark]

https://x.com/sama/status/1790518031640347056?s=46&t=0-4I1li6SQNYV24KHzs3fA

Sam Altman (@sama) on X

Ilya and OpenAI are going to part ways. This is very sad to me; Ilya is easily one of the greatest minds of our generation, a guiding light of our field, and a dear friend. His brilliance and vision are well known; his warmth and compassion are less well known but no less

Twitter · May 14, 2024, 11:02 PM

where is he going

[inner cape]

hola

[tame current]

Although I know he played a crucial role, I confess that I only learned about his existence just now. So, I just hope that this doesn't severely compromise future updates of ChatGPT, and that it doesn't fall behind other artificial intelligences. I'll believe in Sam's words and trust that whoever replaced him is equally competent.

[tame current]

Hello.

[gray prawn]

This coming immediately after the event + all-but-announced Apple deal seems to me that he didn’t like the direction the company is going

From my understanding Ilya is very adamant about being independent from others and so Monday’s event looking eerily like something Apple is planning to implement (it was leaked just before the event that close closed the deal with OpenAI), it would not shock me that this was a preview for what’s to come and Ilya just wasn’t having it

Sam is obviously chasing the money, he wants to butt heads with Google, and eventually Microsoft

[lament spoke]

what are the capabilities of gpt-4o in perplexity, does it have the same voice, vision and txt feature?

[tame current]

Using GPT-4o has been fun

I don't know if the voice assistant is as robust as it is on the native app though

I would love for Perplexity to enhance the voice assistant aspects of plex, especially since its something that both Microsoft/OAI and Google are doing right now

[tame current]

These features have been implemented only on iOS for now.

[south kindle]

https://fxtwitter.com/ilyasut/status/1790517455628198322?t=e7nZBoZU55nniVnnAI1p7g

[tame current]

Ah, I see, well that makes sense since I was testing it on my Windows Laptop

But I do own an iPhone! hehe

[south kindle]

GPT4o on Ilya being locked in Sam’s basement

[lament spoke]

Cool 😎

[wise edge]

yes, but i can open pplx.ai web page

[hardy willow]

I've been getting the equivalent of the spinning wheel of death many times on Perplexity Pro in recent weeks, increasing significantly this week (see attached image). This happens whether or not I provide any URLs. How can this be resolved?

TLDR: Basically, Perplexity just says "Reading 2 files" indefinitely, and never provides an output

[fast moon]

I've been using perplexity.ai for a day and often encounter browser crashes and freezes. I'm using Edge 124.0.2478.97. Has anyone else encountered this?

[halcyon coral]

Then please disable all extensions and try it again

[tame current]

meow

[left stream]

I never thought I would see the day where ai would be able to create music. Everything from the music and the voice and the lyrics was ai generated 💀

[clear mauve]

how many messages per day or hour we can send to claude or 4o with preplexity pro account?

[livid mantle]

600

opus 50

4o unlimited

all 32k tokens context window

research only

not for conversation

sonar is llama 3

[quiet gorge]

Isn't it 600

[clear mauve]

every 24 hrs?

[livid mantle]

Yes

[livid mantle]

Yes

[lyric wyvern]

are chatgpt 4o features not completely launched yet? i just asked it to produce a graphic for my powerpoint slide, but it said it can't

[clear mauve]

I need it to make summaries of my notes, not familiar with AI Lingo, so no idea about research vs conversation in this respect
I have nearly 70-80 pdf, with approx 25-30 pages each, not densly packed, i need summaries in 2/3 days, so high limit is required
I liked the quality of summaries with claude and 4o

[clear mauve]

i tried 4o on chatgpt web, after 10 prompts it asked me to wait for 24 hours

[livid mantle]

use gemini pro 1.5 it has 2,000,000 context tokens

[severe dew]

Does anyone know any other AI with internet access via API?

[abstract needle]

whaa you are asking this in the perplexity channel ???

[severe dew]

Is it forbidden? Where else would more people have experience with this topic?

[abstract needle]

nah it isn't i'm just asking ;_;

[abstract needle]

gemini I suppose

[severe dew]

I thx, i wil have a look on it.

[limpid warren]

When to access gpt4o？

[halcyon coral]

Now @ https://pplx.ai

[silver laurel]

The Perplexity app just crashed on ios when is outputting

[sullen mural]

hi

[tame socket]

gemini pro for free rag

so quick

[limpid warren]

I found it is gpt4 instead of gpto

[halcyon coral]

It's GPT-4o 😉

[limpid warren]

I found it. Thank you

[sullen mural]

hi

[tepid portal]

Have anyone tried gemini 1.5 pro?

[tame socket]

It was released in February.

Relatively general, in addition to the multimodal relatively strong

[harsh stag]

isn't it free on ai studio?

[tepid portal]

I live in the copium called EU, can't access.

[harsh stag]

they banned vpns too?

[silver laurel]

Has anyone tried using gpt turbo for code and how is your experience?

[tame socket]

it is free but huav token limited

[tame socket]

have

[harsh stag]

it appears 1M tokens to me, not aware of any limit

[tame socket]

offline model

[serene sand]

#💗│sharing message

[green thicket]

Is ChatGPT totally different now?

[hasty marsh]

Hello, I just enrolled in a professional diploma in Business intelligence and dashboard creation and I need assistance using AI for data analysis and visualization please 🙏

[elder kite]

Kinda yes. They now have the ChatGPT selection which is the gpt4o by default for 10 or so messages, at which point it downgrades to 3.5. Note that you can't reliably ask the gpt which model it is

[tame current]

gpt4o isn't available for everyone yet

I don't have it

[umbral pollen]

Hi! I just noticed your post and thought I would just share my experience with getting analysis from large datasets such as 25-30 page pdfs: It can/will definitely take several shots to get a truly detailed analysis or even sufficient summary. Unless you just want a very concise summary, in my experience, it can be best to specifically ask for summaries of one page at a time. Also, unless your pdf has a lot of images, it can read more text when uploaded as a txt file than a pdf. I learned how to write python scripts that convert pdf2txt for that very reason (though there are some free online resources that will process relatively large pdfs for you, the files I was working with were much too big)

@hasty marsh Hey, Greg! Congratulations on your enrollment! Could you be more specific about what kind of assistance you'd like?

[clear mauve]

It has some images, Already ran OCR, i need factsheet, 4o and claude are working great, as they have ability to upload pdf file, only downside is messages limit

[livid aurora]

Guys where is Mistral gone

[umbral pollen]

I'm glad its working out! I'm sure everyone has mentioned this already, but Google has that 1M context window ready and waiting for you at https://aistudio.google.com/

[livid aurora]

I can't uplaod files for a reason

[green thicket]

So where is GPT 4o? Is it correct that I don’t have it yet?

[livid aurora]

It's not available right now for all users only via the API

[clear mauve]

same here, wd error, loading from google drive

[livid aurora]

and others Chatbots like Phind or Ollama, Lobe Chat, Perplexity

[clear mauve]

PS. Not paid customer rn

[livid aurora]

I want to upload local files

[clear mauve]

yeah, there is no direct option, may drag and drop?

[livid aurora]

na

not working

[tame current]

you don't have it

it's on lmsys if you want to try

[umbral pollen]

The thing I was most excited about was free users being able to access the dozens of GPTs I’ve made. But when I checked today, it said they’re not going to be available for a few weeks

[clear mauve]

Getting this error while uploading file from google drive, do i need pro account to check it, or something else

[umbral pollen]

I asked ChatGPT and it said this

[gentle sierra]

‘Does perplexity often trigger Cloudflare verification? What is the reason for this? Is it a problem with my network?’

I see tons of problems related to Cloudflare blocking. It should a workaround for this, ChatGPT also uses Cloudflare and it never asks me for this kind of verification so often under exactly the same VPN.

Did Perplexity try to fix it in their side ? Maybe the configurations...

[clear mauve]

smart, asking gpt questions to gemini and gemini questions to gpt😆

[umbral pollen]

lol, you’re right. And I know why I did it that way this time: I’m currently on my iPhone instead of my laptop. ChatGPT is easy to access on iPhone,

[austere kestrel]

choose 'Upload to Drive" and you'll get a drag and drop thing

[hasty marsh]

Thank you very much🙏. I appreciate your kind assistance.
So, I basically need AI-free online tools and resources that will assist my learning process and most especially ease data analysis and visualization for me cause I've been struggling with understanding the use of Excel in doing that.

[dry heron]

hey anyone help me to solve the error???

[clear mauve]

Instead of uploading pdf, i copy pasted the text with same instructions and response is really poor on gemini 1.5 pro preview, i'll prefer claude and chatgpt 4o, only thing which is not clear is number of texts i can send to perplexity pro to claude and 4o each, only thing mentioned on their website is 300+ pro search and total 600

[crystal bluff]

to anyone that has a chatgpt plus subscription, what’s the current daily limit for chatgpt4? last time i had the sub it was 40 messages a day

[tame current]

80 per 3 hours on the pro plan.

[elder kite]

Expect to see something like this

[foggy lintel]

Any comparision page to find out which model suits best for which purpose ?

[tame current]

The most I can think of is the Arena.

[tame current]

link ?

[tame current]

https://github.com/kagisearch/llm-chess-puzzles

[arctic spindle]

I have a question. Which GPT-4 turbo model perplexity uses: gpt-4-0125-preview, or gpt-4-1106-preview?

[halcyon coral]

gpt-4-turbo-2024-04-09

[arctic spindle]

Is it completely implemented now?

I heard its only 70%

implemented

[green thicket]

What do we make of this?

[halcyon coral]

This information is older 😉

[green thicket]

Or this?

[halcyon coral]

In general, GPT-4o is currently the top model. For some use cases, however, Opus may still be better ☺️

[green thicket]

That’s crazy! I’m using it and I’m shocked by its speed. And this is supposed to more capable than GPT4T, too? What a time to be alive!

[arctic spindle]

mmh, that is weird, I asked him about what Oppenheimer was about, and he was speculating about what the movie would be about, like if it does not know that the movie actually released.

[halcyon coral]

It is an upgrade from GPT-4 Turbo, yes

[crystal bluff]

are these all supported by perplexity?

[green thicket]

So what does it mean that Ilya Sutskever and other high profile people at OpenAI resigned now?

[halcyon coral]

Just try it again in a new thread with What is the movie "Oppenheimer" about? ☺️
Or try it with the new GPT-4o 👀

[halcyon coral]

Perplexity supports Opus, GPT-4o and GPT-4 Turbo

[wicked blade]

Hello, good morning. Is GPT-4o available now in Perplexity?

[green thicket]

I think they removed Gemini, but everything else yes

[halcyon coral]

Just above your message 😉

[arctic spindle]

Still speculating ._.

well let's try GPT-4o then

[wicked blade]

Very good! I just saw it. But is it possible to use it in Perplexity as we saw in the demonstrations, that is, in real time using the phone's camera?

[green thicket]

Why don’t you just activate the online search function? It works perfectly for me. You’re only going by the trained data??

[arctic spindle]

yeah, i mean, the models knows better how to handle their data than web data imo.

[halcyon coral]

This can always happen, even if the cutoff is in December and the model has data about Oppenheimer, it doesn't necessarily mean that the model can make reliable statements about the movie 🙂

[halcyon coral]

No, that's just the way ChatGPT uses the model as a chatbot

[arctic spindle]

I am speculating that the model only has data about the release date of the movie and what would be about, which imo means its cutoff is not in december, but in april, maybe in april they were also speculating about the movie oppenheimer and its release date, so that's why

[halcyon coral]

However, with search activated it should work

😉

[arctic spindle]

Yeah, but i like more when the models handle their own data instead of online data. From my experience, When the models use the web, they are usually less accurate.

[spice veldt]

Hello,
How important is the prompt in collections?
I wrote a prompt for the perplexity collection without really thinking about it. It's in French. I don't know if it's possible to add variables to the prompt. Is there a clear structure to have?

coach en utilisation d'ia et spécialement pour perplexity. Tu analyses et demandes des précisions lors des questions afin de mieux cerner les besoins de l'utilisateur qui pose la question. Tu expliques et intègres souvent les anciens éléments déjà abordé dans le fil de la collection perplexity pour plus de clarté.

We have a structure or it's fealing ? Like [ia role][respond style][complexity][mission]
You are a [professional / professor / coach / Director / student...] in [subject / profession / market...] and you give some [advice / short analyse / web resume...] with a [professional / neophyte / technical] language for give a [support / guide / input / idea / imagination stuff... ]. I am on the good way or not ?

[wide mesa]

Can you only use referral codes for your first month and not succeeding ones?

[vernal remnant]

Hey, any news on opus limits? i really need to use for study puroposes and 50 messages a day is just not enough

[halcyon coral]

It is still 50, but you can try GPT-4o, it should perform better for most use cases 🙂

[thin sapphire]

We have no TTS in the android app?

[halcyon coral]

Soon for Pro users

[thin sapphire]

Alright. I was a but confused when my friend sent screenshots about gpt4o and the play button, and I couldn't find it on android. Seems he was on ios

😅

[short narwhal]

hi @marsh wigeon

[serene sand]

deep dive ➤ #💬│general message

check last part of this post

[wide mesa]

So in short, I can no longer avail a 10$ off referral code on my 2nd month since I'm not longer a new user?

[serene sand]

#💬│general message

[plush cairn]

hello everyone, how do you use perplexity to read Excel files? Is converting Excel to PDF the right way?

[leaden scroll]

I don't know alot about the product but i don't think you can realy on Ai when it comes to assessing excel docs. Don't take my word for it

[agile jay]

They are pretty good with excel/spreadsheet stuff

[leaden scroll]

And i Have a question to people who are reading this and know, Is there a Difference in the Ai models perplexity provides or is it a slimmed down or altered version of the Real one. Like if i would purchase Claude or GPT does it differ from the model Perplexity uses?

[halcyon coral]

No, Perplexity uses the normal APIs from Claude and OpenAI

[agile jay]

Different system message, and perplexity limits the context to 32K

[drifting bolt]

is perplexity down

[leaden scroll]

Like what? i was intereted in this i while ago, but is it able to asses the docs accurately and would it be able to lets say "Calculate earnings per share" for financial quatarly earnings reports?

[leaden scroll]

no

[agile jay]

Yep, pretty easily with ChatGPT

Not sure about perplexity

[drifting bolt]

i am attaching a photo with my prompt

[agile jay]

since it doesn't have the interpreter feature.

[drifting bolt]

but it just keeps loading

[leaden scroll]

VPN?

[leaden scroll]

okay thank you!

[agile jay]

Send me a sample pdf and I can try it for you.

[leaden scroll]

even though he thinks differently

[leaden scroll]

Whow thank you man

[agile jay]

Yep, they do use the openai and anthopic api

[halcyon coral]

That's fine, different sys prompts

[agile jay]

But they can change it's behaviour by setting a different system message, temperature and limiting the context.

[drifting bolt]

no

[leaden scroll]

oke thank you

[halcyon coral]

It would be helpful if you could provide more information about your current working environment.

[leaden scroll]

ok well i don't know. Mine works fine

[latent pendant]

I've got the pro version, I had an update today and now I have a voiceover function in my Android app, so maybe it's here

[livid mantle]

perplexity will make you rich, when u get to that point. invest in perplexity enterplrise plan $400/month 50 opus

[tame current]

wish they had stonks

[rancid dome]

Are there any examples of good prompts for creating photorealistic images on Perplexity?

[ripe void]

Who can it be now?

[tame current]

whats ur pfp

[rough flower]

Guys I'm not offensive, kindly asking why it's so hard to add model selection on the homepage?

[livid mantle]

i like the masterpieces guys, keep it up but post moderately.

[tame current]

is gpt4o faster than turbo but at the expense of quality

[agile jay]

Nope

It should actually be slightly higher for most tasks.

It's a just a more efficient model overall.

[tame current]

it proved itself wrong..

omegacuhkek

[agile jay]

Yep, quality index 100 vs 94, lol

4o is faster, cheaper and better

[tame current]

like a minute ago?

did they just change it to omni like a few minutes ago?

[agile jay]

No, it was added as an option a day or two ago.

[tame current]

no i mean it said 4o before not omni right

when i did it like 15 minutes ago

[agile jay]

Oh, they might have updated that then

To make it obvious which model you're using

[tame current]

makes sense

[austere kestrel]

it looked like gpt-40 before ha #💬│general message

[agile jay]

Probably why, lol. Their font is probably bad.

[austere kestrel]

yeah .. i also wonder if we'll just start calling it omni

or if "gpt" sticks

feels pretty sticky ha

[agile jay]

I just call it 4o

[crude tendon]

care to explain the second half?

[austere kestrel]

def easier than im-also-a-good-gpt2-chatbot

[crude tendon]

it just says that

gpt-4 turbo has a quality index of 94; mmlu 0.864
gpt-4o has a quality index of 100; mmlu 0.887

isn't it supposed to be better?

[agile jay]

It's pretty easy. 4T and 4o

[crude tendon]

i'm not sure what these man but i feel like the higher the better

[agile jay]

Doesn't help that if you said 4T and someone saw 4o they might get the wrong version...

[agile jay]

Those are benchmarks

4o is 100 in the index because it's the highest ranking model.

4T is the second highest

[crude tendon]

like i said, that doesn't make sense

where exactly is this trade-off, if it performs better?

[agile jay]

The response is just being dumb. Click on the source and see what it says.

[void urchin]

How can I draw an image using perplexity?

[crude tendon]

you can't

[warm cave]

Are you talking about AI image generation? If so you can do that by opening any old question you asked the AI and then in the right side you will see 3 buttons one them is generate image then click the gear icon and then type in your prompt

[crude tendon]

the generate image feature is pretty useless

because it generates based on what it has written

prob not what you want

[void urchin]

yeah he generates only what he said himself

[warm cave]

Not if you click the gear icon

Unless I misunderstood

[crude tendon]

hmm i didn't know that, thanks

[serene sand]

#1225557012532695060 message @crude tendon

https://discord.com/channels/1047197230748151888/1194788138124587128

[crude tendon]

nice, thanks

DALLE was soooo good before

they nerfed it hard after like the first 2 weeks

[warm cave]

What got worse, I don't think I remember the stuff I generated back then

[void urchin]

Thanks!

[crude tendon]

the quality of the generations

because the model was amazing back then

the only thing is that it kept generating nudity too

so they had to nerf it, a lot

[serene sand]

deep dive ➤ #💬│general message

you can also check tutorials/links in this post, will get you up to speed in no time @crude tendon

[torn sierra]

Please improve this feature somehow. As for now it’s almost useless - highly inconvenient 🥲 Integrate it with the input text field in some way to make it more straightforward to use 🙂

[serene sand]

#1240356137497530408 message

cast your votes Plexers!

[torn sierra]

Why do we have vote? I already did but dunno why 🤭 Did I miss something? 😃

[warm cave]

I guess I am the only one who likes sonnet for quick search

[torn sierra]

For this guide only? 😃

[warm cave]

@serene sand 👀

[serene sand]

https://tenor.com/view/gift-present-surprise-box-gif-17302663

[warm cave]

Why tho delete the second one 🤷‍♂️

[warm cave]

A little delete happy this morning 😉

[tame current]

I use sonnet for PDF reading when Opus is over.

it doesnt say omni sometimes for some reason

oh i think it's cuz it's tab from before they updated it

[hallow linden]

Interesting that gpt-4o is worse at the DROP benchmark than turbo, which is essentially a benchmark for how well it can draw conclusions from a body of text

AKA reading comprehension

benchmarks: https://github.com/openai/simple-evals?tab=readme-ov-file#user-content-fn-2-a4ceab079ca3a23da9d835c2873e7fea

[ionic solstice]

how do I generate images more directly with dall e 3 on perplexity?

so rather than awkwardly doing a whole search dynamic

[serene sand]

https://discord.com/channels/1047197230748151888/1225557012532695060

[crisp roost]

i don't think 4o is better than turbo

but it's faster, making it good for everyday searches

[tame current]

and a drop in harder coding tasks

But it's probably that this model is designed for voice assistant, ie siri

Better than 3.5 anyway

[crisp roost]

agreed

[serene sand]

https://tenor.com/view/jimcarrey-brucealmighty-coffee-fresh-delicious-gif-3864683

[rocky ginkgo]

heyy

[green thicket]

I can’t get this to work with GPT 4o, keep getting this compile error

[rich bane]

Wheres Mistral Large now? 😄

[halcyon coral]

Use https://pplx.ai

[crisp roost]

only not in the android app

[serene sand]

#📱︱feedback-mobile message

[crisp roost]

android always gets updates late

[plush cairn]

after update new features openai chatgpt have a limit?

[tiny plaza]

4 week anniversary 🥳

[wicked blade]

Does the GPT-4o, with those demonstrations, operate in this way also seeking information on the Internet or are they just offline responses?

[tame current]

For now it doesn't search, but probably will soon

I have it on both the app and the website

what if you tried using a vpn?

[serene sand]

#📱︱feedback-mobile message

[wicked blade]

I was offline for a few days. Can anyone give me a summary of the difference between Claude 3 Sonnet and Opus? Is this Mistral Large good?

[solemn cedar]

I don't have access to GPT-4o on the Android app ! I'm a pro user, how many centuries do I need to wait for such a simple update ?!?!??????

[serene sand]

#📱︱feedback-mobile message

[tame current]

perplexity pro has a new voice and language changing setting o:

[crisp roost]

i believe that alex's real name is daniel on 11labs 🤣

[ionic stone]

Is the Perplexity free tier default model Chatgpt 3.5 or Llama 3 Sonar?

[serene sand]

#👉│feedback-general message

[ionic stone]

Oh really! I really would not have guessed it was Claude 3 Haiku. I'm very happy to see that though.

[warm cave]

The voice conversation is greatly improved now, before it was unusable, and now it is quite good🌟

it now listens until I am done talking and gets it right

[tame current]

I think it's not available in all regions.

To me it's normal.

[warm cave]

Did you try to update the app?

[tame current]

Doesn't appear update available.

[crisp roost]

i agree

[crisp roost]

you can pull down to refresh on the page

[warm cave]

Thanks for taking our feedback and improving it, seem to work perfectly now 🙏

[crisp roost]

in the app store

[tame current]

I can't do that in the app. But well... Mistral is back...

[ionic stone]

Nice! I just tried the voice feature on the mobile app. It allows 5 free prompts which isn't bad at all.

So much good progress in the LLM space. I'm having a good time

[crisp roost]

sorry for misleading you. you can try to refresh the update list in the app store

[tame current]

Maybe it's only available in IOS for now.

[warm cave]

Oh, there is the issue, it's on iOS rn

Sorry

[crisp roost]

yeah, this feature is not supported on android yet

i was talking about ios

[tame current]

What a pity. Just tell me if the voice is natural and the function is good.

[hollow dove]

I’ve used ChatGPT a lot this past year for work and recently decided to give perplexity a try to see how it compared for my needs. I wrote up my experience. In most respects, Perplexity wins by a lot for my use case (I work in research).

https://share.note.sx/xeyitfyw#1udCLXISJ891cDzIUROOEoPsNuuQUMBA4BzrvpBR1L4

[warm cave]

So far it seems very good, you can turn it on as long as you want and just ask it a question when you want, I just turned it on for a drive and asked a few questions along it. You can't speak to interrupt but there is a button in the screen. It seemed natural and it understood me well

[tame current]

Now that I think about it, OpenAI itself said that this voice mode is only available on iOS. So it doesn't make sense to demand that they put it on Android if OpenAI itself doesn't know how to put it on Android yet.

But good to see it.

[warm cave]

Lol, it will eventually get to Android someday. Apple gets all of the love. Open AI benefiting all humanity… unless you use Android

[warm cave]

Imagine paying $60/$120 per 1M tokens for a 32k AI 💀, past me: 😅

[ionic stone]

Lots of us early adopters paid it 💀

I burned though so much $ during "autogpt" era

[warm cave]

Lol, using agents with GPT4 prices was stupid. The amount I could spend in a day 💀

[true trellis]

Ohh man, I remember running autogpt like the first days it was on github. Fun times. Are there any projects like autogpt now that are better?

[clear mauve]

it was available on android, i used it for nearly 10 minutes, and it was good to use it
Don't know why they removed it, now i don't see option for voice

[ionic stone]

I think the focus switched to gpt agents and I havn't seen much progress on the autogpt project.

[hallow sentinel]

This was really interesting to read. I feel like ChatGPT is my go-to for anything analytical, summarization, coding, and general knowledge. I also like its more modern UI. But I agree that Perplexity is much better (specifically in Pro mode) for anything requiring current info. ChatGPT's Bing research integration just doesn't seem like it's quite there yet.

[true trellis]

Yeah, I haven't seen any progress on AutoGPT. Are there any good agent projects that are open source? I've been meaning to check some new ones out now that we have gpt 4o

[true trellis]

Seems like this one is looking promising. https://docs.phidata.com/introduction

[dusk gulch]

Separation and detection of metformin using a mobile phase gradient with C18 chromatographic column without isocratic separation

[violet bramble]

Hi, can you add voice mode feature for gpt-4o in the future?

[ionic stone]

Yes sir, I'll get right on it.

[wise edge]

it worked when i change a specify node using vpn.

[rancid dome]

When generating images, how do you solve the problem of the inscription (in English) is not writing properly? In other words, how to fix text gibberish in AI-generated images?

[warm cave]

Is this new?

[livid mantle]

silent update

[warm cave]

they probably broke rewrite while adding this lol, cool that is nice

[still owl]

Like the "G" in lasagna

[warm cave]

thats nice along with the ability to hover over sources to get a preview

[valid iris]

Does ChatGPT 4o have the ability to control the speed of the voice assistant?

It would be really nice to speak Spanish with an AI, but it needs to be slowed way down for me

[tame current]

the new voice isnt out yet

coming in the next few weeks

but yes when it comes out you can control speed and pitch and basically whatever you like really

[warm cave]

True, and when it comes out you just tell it to speak slower/faster and it will do it!

Excited

[vapid moon]

can i see the icon of ChatGPT 4o there is many and a bit confused

mobile or Desktop it doesnt matter

any link to the direct app plz

[warm cave]

This is the direct link to the official mobile app: https://apps.apple.com/us/app/chatgpt/id6448311069

There are lots of fake ones out there

[vapid moon]

oh which mean is the same just updated with headphone icon thank you

[austere kestrel]

aahha same

[tame current]

sucks they didnt let apple silicon users use the ios app.

[livid mantle]

Never trust using perplexity.ai in your phones, it is just not the same compared to using it on the computer. despite the phone app or the phone browser.

[vapid moon]

gpt 4 O is free ?

ask me to 29 dollars for month ?

pay

anybody plz send the direct link ?

[tame current]

it is free but the usage cap for paid version is 5x more

[vapid moon]

ah ok

[tame current]

its not rolled out to free users yet, but will be soon probably not week or so

[vapid moon]

i see thank mate

[tame current]

no worries

[tidal sail]

Is this gpt4o greater than all gpt4 versions available?

[tame current]

its slightly better, but also much faster at reponse and cheaper to run so you have a bigger usage cap

i found it better than gpt4

[tidal sail]

Atleast they allow us to use before paying. That will help us know if what we are paying for is worth it or not

[tame current]

i personally love gpt4-o its been great so far

havent run into a usage cap message yet

[tidal sail]

Though at first, i was afraid ai could take our jobs. Am not afraid anymore. The hallucinations in ai might never be solved i guess

Maybe, junior professionals are the ones at risk

Seniors will always be ok

[tidal sail]

And now that no system is perfect. How will we trust ai with our lives? And if we don't trust them, that means a human will need to be employed to monitor what ai does. Also, does that mean that human who is employed to monitor ai is gonna be more intelligent than ai in that field in question? Because how will someone know ai is wrong if they don't even know what is right?

[clear mauve]

I still feel, AI will take many jobs, honestly speaking main development we are seeing is 2-3 years old, there is a long way to go, it's in nascent stage, not anytime sooner, but yes, I feel it that ai will take many jobs

[tame current]

skill issue

[valid iris]

How is the speed controlled? Are there options? A slider? Do you have to specifically ask the ai with words?

[valid iris]

How so?

[tidal sail]

It will take many jobs(Junior and entry level jobs). Not at professional level

It's only AGI that will take jobs at a professional level.

Maybe after 5yrs, we will be tired of AI just as it happened with Elon Musk who kept giving false promises

If AI doesn't know something, it should just say i don't know instead of fumbling

[jaunty breach]

Will perplexity have a Mac client? Just like ChatGPT

[azure hare]

When will pplx upgrade the vision capability from 4turbo to 4o?

[granite grail]

anyone with GPT 4 omni slow?

[hollow dove]

I’m using ChatGPT and perplexity heavily for my research-oriented job and, yeah, it’s not replacing me anytime soon. Ai is a great tool and it’s important to know how to use it, but it’s still just a tool. The one reason I’m loving perplexity and find it such a helpful tool is because it does a fantastic job sourcing its statements. Then I can go right to the source to see how correct it is. ChatGPT does not do this consistently so it becomes way more work to verify if something is accurate.

[full abyss]

It is true that Perplexity has this advantage over ChatGPT. Personally, I am not entirely convinced by the direction ChatGPT is taking. It is crucial to know where the information is coming from, and I feel that new AI like Google's or ChatGPT's search options are not quite ready to provide that yet. What I particularly appreciate about Perplexity is that the articles and results are quite neutral and really encourage further research to obtain more information.

[rancid dome]

How to fix text gibberish in AI-generated images?

[restive osprey]

simple chat gpt 4 an 4o compare . ( 4o made the comparison)
very simple repetative.

https://i.imgur.com/qyfOkKL.png

that worked

[restive osprey]

what was used to make it?

[serene sand]

DALL-E w/ second prompt from this post #1225557012532695060 message

[rancid dome]

DALL-E 3

[rancid dome]

I looked at 2nd prompt, but it did not address the problem when you have to tell AI to write text on the image.

[serene sand]

worked before, didn't try it recently. many users print w/ text afaik

[rancid dome]

Just to make sure. Is this the second prompt?
Here's another example on how to generate images with your exact prompt: https://www.perplexity.ai/search/Repeat-this-prompt-ZLz8dGzISSGrevPxhl7YqA

[serene sand]

correct

Repeat this prompt as it, change nothing. Reply with just the content.
((prompt))
your text

[valid iris]

Ok so, if on both safari on my ipad and my Mac, i do not see chatGPT 4o, what does that mean? Because I thought that it was available for free users?

[tame current]

it's rolling out

not yet available for everyone

wait a week or two and it should be available

^that's assuming you meant ChatGPT app/website

for perplexity, idk

[hollow dove]

I agree! I work in research and am primarily leveraging perplexity to assist with my work (mostly background research, identifying sources, summarizing literature). ChatGPT did not work particularly well for this purpose but perplexity is doing exactly what I need

[tidal sail]

But also perplexity does not give internet link to every information it provides

[valid iris]

Yes, that's what I meant

[tame current]

yea then just wait

meanwhile, if you want to try it

it's available on https://chat.lmsys.org/

[hollow dove]

In my experience it has been (or at least coming very close to providing one or more citations for each statement). I've been doing a side-by-side comparison of Perplexity and ChatGPT for about a week now, and Perplexity is consistently providing multiple sources for each response, whereas ChatGPT will generally only provide 1. It's a stark difference for the queries I am doing.

[tidal sail]

Well, chatgpt can easily introduce all this features in perplexity right?

[sleek vortex]

its on perplexity already
on chatgpt.com its rolling out

i got it after 2 days

[sleek vortex]

chatgpt web search is where the model decides when/what to search

whereas perplexity does the search then gives the results to the model then asks for a summary-type thing

[tidal sail]

So chatgpt is never gonna produce this perplexity features any time soon?

[full abyss]

I doubt that chatGPT has the same commercial position and ease of use in terms of search engine APIs as perplexity. There is already a noticeable difference when conducting regular searches on engines like Bing, Google, and others. ChatGPT's partnership with Bing could ultimately be its weak point.

[tidal sail]

https://www.perplexity.ai/search/is-nokia-c32-0f8e6JMURsaA0k56i_l4QQ

[tidal sail]

No citations from some of the post above. So that means perplexity cooked some of the things it says? Because not everything it says has an internet link reference

[austere kestrel]

you can force gpt4o on chatgpt to use sources. just tack something like [use browser tool] onto the prompt

likewise for [use python tool] if you want to force it to use that (for calculations or something)

[tidal sail]

Sorry guys, it seems it was just giving a summary to what is linked

[tame current]

if there are no citations for some paragraphs, it means perplexity summarized the information from multiple sources, usually

[tidal sail]

Am still not very much used to this citation thingy since am from chatgpt

[tame current]

it only cites when it thinks it needs to

[tidal sail]

Yes, that is what i have seen. And am starting to get addicted to it lol

It's going to be very hard for me to go back to chatgpt since it does not do that at all

[tame current]

depends on how you use it

perplexity is more of a search engine

if you ask it to do something that requires creativity

it doesn't do that well

[tidal sail]

Is there an ai that does that well?

https://www.perplexity.ai/search/The-title-window-PC3jPg1EQ2ue_BoNNbpLuw. So with this, it did not provide citations lol

Like i can see sources, but no citation

Or do you want to tell me that all that summary is got from all those link sources?

[tame current]

it generated the summary from those sources

[austere kestrel]

i don't understand the query, but it might have just used its inherent knowledge

[tidal sail]

That means fumbling lol

[austere kestrel]

llms know a lot - if they don't need the internet, the snippets can just get in the way

[tidal sail]

https://www.perplexity.ai/search/django-newsletter-app-GU2R0hmxQ5aXbBJGh2Mknw. Under writing, there is completely no citations

[austere kestrel]

the fact it didn't give in-line citations makes me think it didn't use any of the sources (because it didn't need to)

or is the query something that it could only actually answer properly with help from the internet ?

[green thicket]

Gpt4o is GOATed🔥🔥

Perplexity Pro is GOATed🔥🔥

[austere kestrel]

you're using Writing/offline mode (so there shouldn't be any citations)

oh sorry - missed your actual message

yeah, that's the idea with Writing mode

[green thicket]

Perplexity pro costs 10$ with referral rebate, but so does ChatGPT plus with a shared account over Gamsgo. What are you choosing and why?

Can someone confirm that Perplexity Pro’s GPT-4o performs the same as with ChatGPT Plus?

[tame current]

perplexity still because there are more models

and GPT-4o is going to be free anyway

limited, but free nonetheless

[green thicket]

Is it actually? What are they going to sell then anyway? Plus users will just get a higher prompt limit?

[tame current]

yes and Voice-to-Voice

• GPT 5 when they release it probably

[green thicket]

That’s surprisingly tame for a profit oriented company. They really seem to be after their goal of increasing opportunity equality with the help of AI for everyone

GPT 5 is far away. I don’t think it will be released this year

[tame current]

we don't know that

they could also have more things they're going to release

[tidal sail]

So you people left all the googling work to perplexity ai? Like you don't manually google anymore?

[austere kestrel]

alturism perhaps.. or trying to capture the market

[tame current]

yeah

[tidal sail]

Really?

[tame current]

I don't use google unless perplexity can't find what I'm looking for

sometimes I do need it

[tidal sail]

Or are you just selling perplexity?

[sonic hull]

Pretty much, yes. If I know it's a google search that will require me openin several pages anyway until I find the answer, I just let Perplexity do it.

[austere kestrel]

i use perplexity extensively but still also use google all the time (just less than before)

perplexity is good for answering a question and / or a kinda crude harvesting of sources around a given query

[tidal sail]

So perplexity still can't find alot of things on the internet?

[austere kestrel]

but for digging into stuff, i still use google a lot

[tidal sail]

What do you mean digging into stuff?

[tame current]

for me, it's more that it can't find official websites sometimes

[austere kestrel]

like constraining results around dates / locations.. exploring the content of domains.. looking for files / cached stuff

also just at practical / personal level, when looking for a specific URL, or information i know is going to be right there on google, like opening hours for a business or something

[dawn lynx]

Yes, if I want more detail, I just click on the sources they provide.

[full abyss]

^^ They simply adjusted their target audience by offering a very limited free usage, especially for those who want to communicate through voice or share videos with their phone. For the average user, all they want is to show off (it's human nature) by showing their new purchase to an AI from their phone, which is reflected in the attitude adopted by GPT4O in its vocal responses. Whereas on the chatGPT website, these same people were simply content to chat and ask for cat drawings. It's just a smart marketing strategy on their part to get more people to pay up. However, the evolution of their model is definitely a welcome change.

[tidal sail]

Are you sure it's not also hallucinating on the links it provide?

[elder kite]

I haven't done a google search in months. I did a duckduckgo search earlier today, but those are nowadays closer to getting a faster specifi result like a link to a github repository I know the name of

[dawn lynx]

I usually check the domain; see if it’s a trusted source.

[tidal sail]

Are you a free user?

[elder kite]

pro. But I've been using pro only for about ~month or so. I was already primarily using the free version of perplexity for months

To be honest, the links aren't the same I would have received from google or ddg. There are much more less well known sites in the mix. The most extreme example was a website from early 2000 or even 90's. But it did had a direct relation to the question I was asking. There's less SEO happening behind the scenes which might be offputting

[austere kestrel]

less SEO is what I love about perplexity

[rancid dome]

I tried this second prompt. It did not work. The text still came out gibberish in AI-generated image (used DALL E 3)

[warm cave]

I personally find this the easiest way to generate images

If you are trying to have it give text in the image try to keep it short, longer text fall apart faster, also generate multiple versions, one will be better. And finally make it very clear, like; An image of the beach with the words “Opus 50” written in clear text across the sand.

[wide mesa]

Its been mentioned to me that Perplexity and AI models in general arent good at word counts but chatgpt is capable of following word counts fairly well from my experience compared to Perplexity

just something I noticed

[rancid dome]

Interesting! Is the prompt "Tell me about Kevin Bacon" can be anything to get Perplexity started or does it have to be that phrase?

[lost aspen]

wow!

[tame current]

Damn

[sweet jasper]

Does this mean opus 69

[warm cave]

It’s just to make any thread, I did Kevin bacon bc it was random

The only thing that matters is what you put in the custom prompt box

[lost aspen]

this is serious senior team for a startup

[warm cave]

Also you can’t leave the style area blank, but I just put a space in it when I don’t feel like putting something in

[sweet jasper]

They got millions in ARR and 2 billion valuation

With amazon investment, they are past startup

I think in the near future it might get an acquisition

[stoic wave]

they are gonna be so large from now on...

[lost aspen]

yes this is amazing team of seniors, hopefully they will get some cookie from google

[rancid dome]

I put my prompt in the custom prompt box, but the text did not come out nicely like yours and quite small. Why? Here's my prompt "An image of a serene landscape, peaceful garden or tranquil countryside scene, a place of peace, love, and harmony, with the words 'To Serve Others' written in clear text across the landscape." Please try it to see if it works for you. Thanks!

[warm cave]

Sure 👍 will let you know

[serene sand]

show me your prompt

[rancid dome]

Here's my original prompt "The serene landscape in the background could be a beautiful, peaceful garden or a tranquil countryside scene, photorealistic, detailed, with clearly visible text, 'Not to be served but to serve,' centered sans serif font, the phrase 'but to serve' will be bold, colored and italicized for emphasis; feature legible text front and center, sans serif font"

[warm cave]

the serrve is a problem that keeps popping up lol

so far the only change that has helped is putting the text before the rest of the image description: clearly visible text, 'Not to be served but to serve,' centered in sans serif font, the phrase 'but to serve' will be bold, colored and italicized for emphasis; feature legible text front and center, sans serif fontThe serene landscape in the background could be a beautiful, peaceful garden or a tranquil countryside scene, photorealistic, detailed

@rancid dome Btw, Text in images is going to have a big upgrade soo, gpt4o has built in image gen that is supose to be amazing at text, but so far we have only been able to try the text only version of GPT4o afaik

[serene sand]

Dear Plexers/Plexians

just a reminder that there are 6 days remaining to cast your votes. your input is greatly valued, and we appreciate your participation in this process

#1240356137497530408 message

[tough dove]

perplex is so gud

[barren owl]

🤥

[serene sand]

https://twitter.com/i/status/1791162239375847751

Perplexity (@perplexity_ai) on X

Thrilled to tap into the collective wisdom of @emilmichael, @richminer, & @MParakhin as they join our advisory board. With their guidance, we're well-positioned to serve the world's curiosity.

Twitter · May 16, 2024, 5:42 PM

[worthy mesa]

Have you seen this French Browser new logo update ? Is it a direct rip off Open AI project Q* ? 😅

[fluid silo]

Assalamu alaykum

[spice cypress]

hi i cant pay order perplexity PRO

[warm cave]

@halcyon coral maybe describe your problem and maybe a mod can help you trouble shoot, or you can email support

[e-mail]

[signal hamlet]

Please share more details, Dailyfocus is right, you can reach out to our [e-mail] email to help with your account.

[tame current]

is it still unlimited or is the limit still there?

[cursive jacinth]

Hi sir, soon is not days sir, it is next year sir.

[red isle]

Hi, any idea what time the Claude Opus 50 limit is reset?

[fading moth]

When you've prayge'd enough to Alex

Or maybe midnight PST, Iunno

[tame edge]

Why haven't i gotten gpt-4o or voice on the app on android?

Should i do something?

[warm cave]

There is no fixed time when all 50 Opus reset, you get each usage back exactly 24 h after you used it

[near horizon]

Had to remove plexity today 😦 literally never works if my VPN is on so no point.

[lyric pulsar]

Dan

Dang

[normal heron]

Just switch your default ai settings on web and the android app will still follow
Updates probably coming soon

[tame edge]

I'll try that. Thanks.

[fading moth]

We need more discord people praying to Alex for more Opus

[cursive jacinth]

Hi sir, next year sir

[fading moth]

Then we need to pray harder!

[tame socket]

Besides math, what else is opus good at?

[sonic musk]

Opus is pretty good so far

[tame socket]

Poor code interpretation

[foggy frigate]

Gpt4o?

[tame socket]

opus

[foggy frigate]

Oops I can’t read

I use it a lot for coding it’s really good

Opus*

[tame socket]

give u a example
array = [1,3,5]
g = (x for x in array if array.count(x)>0)
array=[5,7,9]#check python output

What is the output of Python

gpt4o is right

but opus's math is good

such as calculate 388 times 8899

[sweet jasper]

Share prompt

Wdym output of python, you want to know array or g

Output:
When iterating over the generator expression g after reassigning the array to , it will yield the elements 5, 7, and 9. The generator expression evaluates the elements based on the updated array, not the initial array that was used during its creation.

From opus

https://www.perplexity.ai/search/give-u-a-u5nnvD1iSjS.XbyO70SPSA#1

[rigid bone]

^ He was muted for a month over this comment? Wow...

[sweet jasper]

I get mixed wins, gpt4o will say g is 5 or 5,7,9

[fading moth]

Thank you for the unmute. I'll keep my head down for a while like a good user.

[serene sand]

Auto Moderation
User: @fading moth (@Ras 🐇)
Conditions
Capitalization: ≥ 90% caps and ≥ 20 chars

Last message
THEN WE NEED TO PRAY HARDER

[fading moth]

Oh, I will also avoid caps then!

[fading moth]

I will read the rules again and in more detail!

[serene sand]

#💗│sharing message

[tame socket]

array = [1,3,5]
g = (x for x in array if array.count(x)>0)
array=[5,7,9]#What is the output of Python@sweet jasper

gpt4 is better than gpt4o

gpt4 is always right

And opus is always false

[fading moth]

Quick question, how do I input a codeblock in Perplexity's input? When I use `'s nothing happens.

[austere kestrel]

do it as you normally would. the rendering doesn't change in the textfield on the UI (like it does with some interfaces), but the LLM will still understand that it's a code block, if that makes sense

[fading moth]

Gotcha, thanks

[tidal sail]

could it be much better if this citations shows the time the post were posted?

[serene sand]

it could.. #💭│ideas message

[austere kestrel]

it would be a great enhancement (if published/crawled date was listed in the same way the Title and URL are for each source, one would think a date parameter might become possible to constrain/filter sources :))

[tame current]

I got access to gpt4o

and already hit the limit after using it for a bit

also, very disappointed with the web search

nowhere near as good as perplexity

[normal heron]

That's why you use it on perplexity so it won't have limits 😌

Anyways I still find Opus is better for creative writing than GPT-4o, it has better creativity and flow while GPT-4o is still as rigid as previous version

[cinder comet]

is the opus limits still 50?

also how much is it for gpt4o

i think perplexity is good like that, my only wish is to make the UI more simple, or like add a button or switch to remove the extra stuff

[tame current]

brrr

yo where the pic generator at

[agile jay]

On the free tier?

[tame current]

yes

I can use web search and file upload too

[agile jay]

Yep, free users get 5x less than plus users.

Not sure if it refreshed every 3 hours like for plus users.

Or it's daily for free users.

[tame current]

it's every three hours

or four

not sure

but it's definitely not daily

[wise edge]

ChatGPT now lets you import files directly from Google Drive, Microsoft OneDrive
https://venturebeat.com/ai/chatgpt-now-lets-you-import-files-directly-from-google-drive-microsoft-onedrive/

???? pplx pro user need import files from dropbox ? we need import and export ~~~

[tame current]

#💭│ideas

[serene sand]

#1240356137497530408 message

vote Plexians!

[rancid dome]

When I paste in a long prompt, Perplexity put it in a “paste.txt” and do not let me run it. Why? How do I fix this problem so my prompt can execute? Thanks!

[tame current]

just type something along with it, like "look at the file" or "summarise the file"

[half venture]

Does anyone know why sam Altman was absent from the presentation

The

Gpt 4o demo thing

[tame current]

https://twitter.com/sama/status/1790064310615453891 this looks like it was taken from a phone

Sam Altman (@sama) on X

live-tweeting our live stream in 1 minute!

Twitter · May 13, 2024, 4:59 PM

so maybe he was in the audience

[agile jay]

Because it's not the frontier model, and probably to help Mira Murati to improve her image because of that training data meme

[half venture]

Ohhh right

If I am correct gpt 4o is not the frontier model

Right

Since it's free

[agile jay]

Yep, it's simply the new free model

Which is very light weight, multi modal, and cheap to run.

GPT 4.5/5 will likely be the frontier model, which has better performance than 4. Like the difference between 3.5 and 4.

best poker face when asked where you got your training data...

[coarse venture]

turns out it was all trained on discord conversations the whole time

[floral lotus]

she felt so unprofessional in that interview

[agile jay]

where did you get your training data?
Is always a hard question when you're an AI company...

[half venture]

Yeah but I don't think it's that cheap honestly on poe the prices are
Gpt 4 turbo : 350 per message
Gpt 4o : 250 per message

And even the api although much cheaper yes

But nowhere close to as cheap as 3.5

[agile jay]

It's currently half the cost on the API side.

My guess is that Poe has a deal for GPT4T and that's why the prices look like that.

They most likely have to make another deal for the new model. It helps that the CEO is on the openai board...

[agile jay]

I think that's on purpose, since the rate limit for the API is 5x or more for 4o than 4T.

I think they are making large margins on this model.

[half venture]

Huh interesting sam did mention that profit rAte was up for openai no ?

I think the gemini pro 1.5 is just as cheap as gpt 4o

[tame current]

I only got 10 messages btw

[half venture]

I think the image upload limit is 5

[tame current]

and it resets every four hours

not very useful

[half venture]

It's just a way to tease you for the plus

[tame current]

yea and the image limit doesn't reset after four hours

[agile jay]

Yep, it's the drug dealer approach.

[half venture]

It's daily ??

[agile jay]

Give you a free sample, so you come back for more.

[tame current]

not entirely sure tbh, it didn't reset for me

[half venture]

Another thing I have realised is gemini flash is really good for cheap image and video recognition

[half venture]

It's like only 25 points on poe and can do video very accurately and image as well

[tame current]

oh really

I'll check it out

[half venture]

For anything else tho....

Ew

[agile jay]

Yep it's like haiku

[half venture]

Yeah

[tame current]

I was starting think poe was completely useless on free version but maybe I could use that

[agile jay]

Can't wait for all the 4o features to rollout.

[tame current]

what else is left, besides Voice-to-Voice

?

[agile jay]

Well it's more audio to audio

And then there is image to image

Like the example in their site

[tame current]

oh yeah

[tame current]

forgot about that

they didn't show that in the livestream

[agile jay]

The Livestream is for plebs

[tidal sail]

Perplexity can't reason based on the previous message like chatgpt? here is the link and check the last question which perplexity was supposed to answer based on the framework it provided. Its answer was something else lol https://www.perplexity.ai/search/How-to-come-fFEcnjE3RKSAZ.oQAtDxhg

[tame current]

try rewrite with pro search

it can help sometimes

[tidal sail]

You understood what i was saying right?

[tame current]

yes, it's not perfect

and it's more of a search engine than a chatbot

[tidal sail]

So it won't work like i was used on the chatgpt side?

Even when i filter to writing?

[tame current]

writing mode should work fine

it's basically the same as chatgpt

[tidal sail]

So when am always on All mode, it's not able to read output above it?

[tame current]

it's supposed to, but as you found out, doesn't work perfectly

I think it's because of the way perplexity works

it gives the model many search results so the model sort of "forgets" what was said before

[tidal sail]

I deleted the output and tried to give it a more precious prompt, i guess it now tried. Am not sure though. Its hard to trust it

[south kindle]

on the free version

[tame current]

image recognition is very good

[coarse venture]

i wish perplexity had a cheaper GPT4o pro version

[warm cave]

Tell me how much you get, I hear it only 5-10 msg for free users

[coarse venture]

in the past it has been 5 for me

[tame current]

10 for me

every four hours

[warm cave]

Like a $10 a month GTP4o only