💬│general

[somber geode]

Yeah... or like 500 words a day and you'll have a novel by the end of like 9 months or so.

But then how much money are you going to make off of that book. That's also a scary thing to think about!

I personally believe that books, right now, server more as a personal branding thing, that I'm an author, than an actual money making tool.

But I'd like to you get your opinion on it, cause I haven't personally dabbled in the area and it's just an educated guess from my end.

[agile jay]

Depends on how consistent your release rate is. I know a lot of authors who make a lot of money, but they are publishing a few books a year.

For most authors, they just need one book to do really well to propel their career.

Marketing also helps...

But if you wanna publish fast, doing the refining method tends to be a lot more effective.

[somber geode]

whats the refining method?

[agile jay]

You just write a book really fast and don't focus on the quality, just the main plot. That ends up being your first draft.

Then you edit the first draft adding a lot more details, fleshing out characters and scenes etc.

This part is normally repeated a few times until you and your editors are happy.

But also means your first draft can be a lot smaller than the final one.

Normally around 20-25K words.

And then the following drafts increase it by many times.

[somber geode]

ahh... cool.. so basically like a thought dump

and then editing

[agile jay]

Yep, it's far more effective.

Writers are a lot better at improving text than at writing it...

Also speeds up the whole process a lot

[somber geode]

Yeah... felt that as soon as ChatGPT hit the market... We are all editors in that sense! 😅

Hey man! I gotta go rn... mind if I add you as a friend and chat later?

[agile jay]

Okay, np

[fallow raven]

Ohh hey sorry I deleted your directory accidentally

We need <Code>

[agile jay]

Rip, looks like it was recursive too

[fallow raven]

wait i think i can fix it

--no-preserve-root

... oh

Yeah, we're not installing arch by the way again. alright haha that's enough of my lame jokes, just love seeing others with clever names too ❤️

I went too far with mine..

I own linux command domains

[agile jay]

Just a basic JSX component...

[fallow raven]

fsck-the-rules
Sudo-rm-rf.me

mine are basic, it's just, fun names afterall

Though JSX I see hides python repos

How do you feel about snakes?

[agile jay]

Lol, pretty sure Google domains added .zip and a few others before selling out.

[fallow raven]

No... t-- there has to be more

though google domains --- I think I'm okay with my mk.directory..

Are they that bad tbh?

[agile jay]

Haven't used JS and Python in a while.

[agile jay]

Well now you can make a .zip domain which is super confusing since file hosters already use .zip, so I can make it look like a file download when it actually points to a site...

[agile jay]

Most of the time I'm using Linux containerized.

Alpine etc, for hosting web servers.

[pine valve]

perplexity down again?

[harsh stag]

exactly

https://tenor.com/view/ah-shit-here-we-go-again-ah-shit-cj-gta-gta-san-andreas-gif-13933485

[fallow crystal]

I'm curious. Is there a big difference between the enterprise pro version and the personal pro version? If I am now a personal pro version, do I need to consider the enterprise pro version?😺

[atomic mulch]

it is down for me right now, it was working 5 minutes ago...

[random flicker]

Yeah is down again

[proper sage]

So what's happen if we have pro plan and cannot use it ?

[hexed grotto]

Hello

What’s happening with the website !!

[hexed grotto]

Same here

[pine valve]

#🔔│status message
It's been announced. It's not available at the moment.

[fallow crystal]

Will there be compensation?

[halcyon coral]

Please check the #🔔│status channel. We appreciate your patience.

[hexed grotto]

Same issue !!

[fallow crystal]

@halcyon coralI'm curious. Is there a big difference between the enterprise pro version and the personal pro version? If I am now a personal pro version, do I need to consider the enterprise pro version?

[halcyon coral]

Please check this: https://www.perplexity.ai/hub/blog/perplexity-launches-enterprise-pro

[fallow crystal]

There is no difference in the behavior of language models (e.g. number of model uses, number of page citations, etc., retrieval speed, etc.) between Enterprise Pro and Pro, right?@halcyon coral

[neat flame]

Has Perplexity been publishing post mortems anywhere?

[sleek vortex]

No

The only difference is compliance and SSO if required

[fallow crystal]

Thank you. Then I don’t need to choose the enterprise version.

Which language model is stronger during use? I have been using claude3 opus. Maybe there is a better option?

[serene sand]

https://discord.com/channels/1047197230748151888/1240356137497530408

[fallow crystal]

Sorry, my English is not good. I use Google Translate. I want to say which one (or which brand)

🤪

[cinder comet]

😜

[fallow crystal]

When will real-time voice chat be available? No need to type. I want to just put on my headphones, close my eyes, and have a conversation. That's good. I'm looking forward to it.

[fluid flint]

I am getting a weird query on my perplexity AI app using on Android whenever I try to shift my AI model to GPT it again goes back to default please help

[lofty yoke]

How does Perplexity work when it comes to finding sources? That’s not something ChatGPT really does and I’m surprised how well perplexity does it. It’s not searching the web in real time I think, because when GPT/bing does that it’s so slow

Or is it really fetching articles live and feeding the context into the AI models in realtime?

[fluid flint]

Can anyone help with my bug

??

[lofty yoke]

Ok what! I can actually ask it anything

It just crawls the web on demand

That’s insane what search engine is it using

[fluid flint]

..

[tame current]

no it's not, chatgpt is very fast when it searches the web. bing is slow but I doubt the bottle neck is fetching the pages.

perplexity most likely fetches cached pages by the search engines

(and apparently has its own index but I highly doubt it's used much...)

it's like asking why Google is so fast...

[tame current]

#📱︱feedback-mobile

[elder kite]

Perplexity has their own spider that they use to spider the websites. They create a vector index out of the pages. They do not use bing or other search engines for queries

[tame current]

source?

[quiet monolith]

Hi everyone, if I create a thread on perplexity and it gets a bit long, to the point that scrolling to as far as possible doesn't get one to the first message, does it mean that the old messages are deleted, or is there a way to reach them?

[tame current]

shouldn't be deleted

[somber frigate]

Hello everyone.
I can't chose the AI Model.
When I chose "GPT-4o" then it just automaticlly back to "default"
And i can't find the content for Page funtion ?

Anyone have the same problem ?

[tame current]

I think you can use "home" button on your keyboard to just jump to top?

[quiet monolith]

Does absolutely nothing 🙂

[tame current]

hm

it shouldn't be deleted tho

[harsh stag]

how "long" are we talking about?

[elder kite]

• https://spectrum.ieee.org/perplexity-ai
• https://www.zdnet.com/article/5-reasons-why-i-prefer-perplexity-over-every-other-ai-chatbot/
• https://blog.emka.web.id/2024/03/how-to-block-google-bard-and-perplexity.html
• https://thenewstack.io/more-than-an-openai-wrapper-perplexity-pivots-to-open-source/
• https://darkvisitors.com/agents/perplexitybot

[elder kite]

someone mentioned elsewhere that you have to restart the browser, not just reload the page. Something about cloudflare

[somber frigate]

thank you sir

[quiet monolith]

Well, the thread itself was started a few days ago, but there were lots of responses. I wanna confirm if perplexity simply deletes old messages to keep space?

[quiet monolith]

Long story short, when I scroll to the top, I get to the middle of it, not the very first message that started the thread. AI itself says that it doesn't delete messages but haven't found anything else to confirm it online

[daring hinge]

why can't I choose a model, I'm constantly being transferred to a regular one

[daring hinge]

this does not work

[tame current]

from the first link:

“When trying to excel in up-to-date information, like news…we won’t be able to retrain a model every day, or every hour,” says Yarats. But crawling the Web at Google’s scale also isn’t practical; Perplexity lacks the tech giant’s resources and infrastructure. To manage the load, Perplexity splits results into “domains” which are updated with more or less urgency. News sites are updated more than once every hour. Sites that are unlikely to change quickly, on the other hand, are updated once every few days.

it could literally just be using the crawler for news sites and google for everything else

[elder kite]

It could, but you can also verify this with the results you get on perplexity. You are less likely to get results from more heavily SEO optimized sites than you would from google or bing. This is why the sources often feel "weird", they are so different from what you would have from google or bing.

How I understand the statement is that they have a staggered crawling on domains. Static sites like regular brochure sites are indexed less often than something like new york times or reddit

You can also see it in the delay of results. I have seen multiple people confused why perplexity can't summarize a tweet (x?) given an url. Because the crawler hasn't seen it yet.

[somber frigate]

DM the support but they haven't answer yet

[tame current]

the less SEO optimized pages just means they rank the results differently, it doesn't say anything about which search engine they use

and the LLM looks at the page

and finds the answer

[harsh stag]

is restarting the browser doesnt help?

[tame current]

so obviously, the sources are going to be higher quality

[daring hinge]

doesnt help

[somber frigate]

yes, i change the IP, restart my PC, clear cookie ... but it keep change back to "Default" AI Model

[tame current]

it even says that in the link you sent:

“You can think of it like the LLM does the final ranking task,” says Yarats. “[LLMs] don’t care about an [SEO] score. They just care about semantics and information. It’s more unbiased, because it’s based on the actual information gain rather than the signals Google engineers optimize for whatever reasons.”

[daring hinge]

the best spent 10$ XD

[tame current]

just wait for a mod to reply or someone from the team

[harsh stag]

are you able to send any messages? or is it just you can't change the model?

[daring hinge]

We can't change the model

at the same time, the image generation model can be changed, but the text one cannot

[somber frigate]

Yeah we just can't change the model. Still can send messages as normal

[tame current]

they've been having some trouble

since yesterday

[daring hinge]

Are you serious? Should I try it tomorrow?

[halcyon coral]

Thanks for reporting everyone, the model change bug will be fixed. Appreciate your patience 🙏

[tame current]

this kind of issues are usually fixed in a few hours

[austere kestrel]

They say they use BERT as part of their approach to ranking..
you know SEO means optimised for Google, right? so half the internet is filled with junk because of it.. so many 'articles' are just a bunch of very specific keywords placed in certain positions all designed to rank higher in google search results

like sure perplexity is ranking results differently, but they don't really have to do anything - websites haven't been 'optimised' to rank higher in perplexity's rag/search process.. it's beautiful and hopefully it stays that way

tbh the only evidence you need of the fact they primarily use their own index is the fact that there are parts of the internet that it can't reach

it would be a better product now if they just used third parties for the search results imo.. it's impossible to index the web like google / bing do.. but over the long term, it will leave them in a better position imo

i'm pretty sure weather and finance data are provided by third parties

no way they're crawling forecasts and stock prices by the minute

[tame current]

such as ?

[austere kestrel]

this doesn't resonate with you at all? #💬│general message

[halcyon glacier]

I wish it would come back naturally

[austere kestrel]

the current lead article on the front page of CNN has been indexed (lhs)
yet another CNN article, which was published an hour earlier but is tucked away in the Business / Economy section, apparently has not been indexed..

[fallow raven]

iy does in fact, do this actually.

You just need to give it the correct prompt. I can give you what I use in my custom system prompts If you would like to try this out, Works really well with the new model. Gets kind of annoying though. cuz it'll do it for everything that you ask it now. I'll show you a screenshot or if I can post a screenshot.

[fallow raven]

Now now. that's assuming this even uses Google which I'm pretty sure it doesn't But yes, SEO of terms do indeed exist. and they've been exploited by many other websites. What are you trying to get at?

Hence the acronym search engine optimization.

[wide stratus]

"LHS"?

[austere kestrel]

aha i don't quite follow.. i'm saying / assuming google isn't involved (at least for the majority of the sources/snippets pulled by perpelxity and used to inform the responses)
I think this is doing the heavy lifting
https://docs.perplexity.ai/docs/perplexitybot

[austere kestrel]

left hand side?

[wide stratus]

What about the left side?

[austere kestrel]

left hand sidescreenshot

[wide stratus]

ah ok. 👍

[austere kestrel]

with hindsight it was both unnecessary and confusing to have added that 😂

[wide stratus]

🤔

[covert aspen]

This might be a strange question, but when I ask Perplexity to produce an image, it gives me a detailed description of how to draw it myself. When I question it about doing it via the image generator selected in the Settings, it states it cannot do that. So why is it that there is an image generator selection box in settings if we can't use it?

[halcyon coral]

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

[south kindle]

LFG Godspeed Starliner

[shut plaza]

Claude is finally available in Canada

Interesting

[rancid dome]

What do you use in place of Claude 3 Opus?

[fading moth]

I'm exceedingly curious if we're going to see any new sweeping bans tomorrow

From Anthropic

That's when their new ToS policy kicks in

I'm very curious if they will revoke access to sites like Poe that are known to also be NSFW

There is a service called moemate that's been around and serves a service identical to character.ai... This is an announcement from MM's head developer

https://cdn.discordapp.com/attachments/1223037723770622036/1247823695511293992/Screenshot_20240605_005906_Discord.jpg?ex=66616de0&is=66601c60&hm=2de125127e118bb3b3fb6b9637a32b98c265662f7d80a635a49b17a2cc2a75ee&

Thoughts?

[toxic lodge]

Interesting: https://searchengineland.com/perplexity-pages-showing-in-google-ai-overviews-featured-snippets-442936

[feral marlin]

idk ive never had any significant issues with opus censorship and it performs consistently well for complex tasks that require multiple steps, i like barely use ai currently though ive just used it to help with reverse engineering and some other stuff

[empty siren]

hi everyone, looking for something open source that can generate synthetic fine-tuning data from corpus to QnA pairs.

[valid yoke]

Claude is available in Canada now

[austere kestrel]

“We observed that our users were already sharing their search pages with others. So we wanted to make a more presentable format,” Modisett told TechCrunch over a call.
I can understand the thinking here. A Thread can be messy. But it doesn't really feel like the original Thread is preserved when it's 'converted' into a Page - it's more like a classic LLM summary of the original thread, rather than the original thread tidied up for presentation

that techcruch article cites this tweet [seo.....]
https://x.com/kristileilani/status/1797634277498290352/photo/4

Kristi Hines (@kristileilani) on X

Perplexity pages get indexed quickly and now appear in featured snippets and AI Overviews.

It should be interesting to see the domain start ranking for more keywords with Wikipedia-like AI content.

#seo

Twitter · Jun 3, 2024, 2:19 PM

which includes this screenshot – how does Indonesia account for 40% of traffic lol?

[valid yoke]

Isn’t it the fourth most populous country on earth?

[austere kestrel]

sure. and India is the most populous country on earth - and yet Indonesia (43%) has 8 times more traffic than India (5%).. according to whatever that screenshot is anyway

• Indonesia 40%
• US 8.5%
• India 5%
  that isn't normal lol

and it doesn't even support indo as an interface language lol

[valid yoke]

It would seem somewhat unlikely but not impossible

[austere kestrel]

sure - not impossible

but utterly implausible

[austere kestrel]

oh fair - on the mobile?

why wouldn't they have it on the web too - like seems they're missing a trick there

anyways, agree to disagree - i don't even know where the screenshot with the charts was from or who posted it (other than that they are into seo).. i just found it striking

maybe it's accurate, maybe it's some glitch or something else - i dunno..

[valid yoke]

Ya

[austere kestrel]

@valid yoke hmm yeah this sounds a bit extreme the more I think about it (in the context of mobiles)

i'm skeptical.. but maybe you're right ha

[valid yoke]

Maybe there are some bots running in Indonesia to milk Perplexity

[vapid onyx]

I do wonder when we will see image generation come to the mobile app

[sleek vortex]

never™

[south kindle]

upcoming UDIO changes. I need to test out their Pro for this "15 minute long" song....

[native thicket]

Is it down again?

[peak dawn]

Is it just me. Or did Perplexity only recently let the Pro search show step by step what it is searching

[agile jay]

Yep, that is new. Around a week old or so.

Basically using an intent system for more agentic like execution.

[agile oracle]

why is perplexity not reading my files

like i am not paying for pro for it to be useless

[foggy tapir]

Does perplexity support PDF files ?

In my case, I can see pdf as allowed file types. However, perplexity is still unable to read them.

[wide stratus]

Why do you pay for it at all? 🤔

[royal niche]

Mine works perfectly fine. I regularly use it as a way to quickly explain some of my client project proposals so I can just ask the question without explaining my entire scenario and context.

I will say though, all of my pdf documents are very text heavy with few images and little to no styling. I haven't tried with super pre processed versions where the pages are essentially just pictures of all of the content, styling, and background images merged as one.

[royal niche]

Not sure why others pay, but I pay for the ability to switch the backend model, upload files, and unlimited chatting. I use it quite often, no way the 5 pro searched a day would suffice. I probably do around 200 queries a day when I'm doing research.

[shut plaza]

Perplexity Pro is one of my most used subscriptions

also I kinda just like asking it random questions that pop into my head

[wide stratus]

Wow ok. I'm totally fine with GPT-3.5 so far, it's good enough for my own occasional query. 😝

[fallow crystal]

Has anyone used the Sonar large32k language model? What is the effect?

[bitter hazel]

you only think you are fine with it, until you try something new. my friend wasted a week of work using 3.5 out of principle. then claude solved it all

[past sparrow]

Hi, guys.

I am looking for a developer who can build the MVP what generate the Video from Text.
Budget is $100 for MVP, and we can discuss about the further process after we have done the MVP.

Please send me DM

[royal niche]

Probably the wrong server to be asking, but just check out Comfy UI. Could probably do it by yourself.

[north magnet]

$100 budget is crazy

most devs here are on $100 a hr lol

[agile jay]

Especially for a task like that.

[ionic stone]

"Hear ye, hear ye! I require a stalwart developer to construct an MVP capable of transmuting mere text into wondrous moving pictures. My vast treasury extends to a whopping $100 for this endeavor.
Once you've achieved this miracle on a shoestring budget, we shall convene to discuss my plans for world domination...err, I mean further development. Prithee, send me thy most desperate digital samurai via carrier pigeon."

[gentle dirge]

Google lens api combined with an llm is pretty cool

[warm cave]

Gave me a good laugh 🤣 thank you

[grim pumice]

[elder kite]

I was gonna say that the requirements were easily satisfied in an hour. In 10 minutes. Then I realized it wasn't text from video but video from text 🤦‍♂️

[humble breach]

Why did Perplexity remove Haiku from labs?

And so many others too.

[solemn cedar]

I can't tell you where it will go, but I can tell you where my money will not go very soon

[wide stratus]

My theory is for saving money, Perplexity has to pay Anthropic for letting us use Claude Haiku on labs website, or?

[outer glen]

lol i think i broke perplexity https://www.perplexity.ai/search/Zdravo-Zam-moram-foAW4rklQI6LCkCsbBfNKg

[outer glen]

broke it again 😭

[blazing nexus]

Hey everyone!

Thanks to Telekom, I now have a year of Perplexity Pro, and could now theoretically generate images, but unfortunately I can't find a way to generate images with Perplexity... can someone please tell me how I can generate images with Perplexity using Dall-E??

[maiden elm]

We'd like to invite you to explore Contradictory Scientist, a unique tool that utilizes the Dynamic Logic of the Contradictory (DLC) to analyze and characterize complex phenomena across various scientific disciplines. Contradictory Scientist deconstructs phenomena and systems by identifying dualistic and antagonistic dynamics, providing a deep and structured understanding.

The DLC methodology allows for the exploration of phenomena through dualistic, structural, and systemic analyses, facilitating the comprehension of their interrelations and internal dynamics. This approach not only offers detailed insights but also encourages ongoing exploration and questioning.

We invite you to experience Contradictory Scientist and discover how it can enhance your understanding of the world around you.

https://chatgpt.com/g/g-CoeDgcQcX-contradictory-scientist

[summer crescent]

I attempted this with a picture from a drama show and both lens and ai was like "I'm not allowed to identify humans"

[gentle dirge]

💀

[granite grail]

also happened to me earlier today lol

[outer glen]

its happening a lot 😔

[tame current]

Perform a search. Wait for it to complete. Then Generate Image button will appear. Click it.

[tame current]

Here is better guide #1225557012532695060 message

[sleek vortex]

lol

[sleek vortex]

probably

[frozen tundra]

Do you think they’re going to add the ability to edit collections any time soon to the iOS app

[balmy barn]

Hello everyone
What is the new coupon code

[warm cave]

The only ones I know of are individual referral codes for $10 off your first month

Hello 👋

Nice, that will be helpful

@tame current We traverse a Discord server teeming with individuals who remain unaware of the treasures they forgo, their lives bereft of “complexity”.

Lol, that I good thinking and cool!

[warm cave]

Its on

[agile jay]

Nope, it's just an extension

Oh, for discord

[warm cave]

I don't

[shut ore]

I'd like to create a perplexity page. I have already seen their videos about Pages. I can't find where to click to create my first page. Any clues?

[agile jay]

You a pro user?

[shut ore]

n

[agile jay]

https://perplexity.ai/page/new

[spiral belfry]

Me too

[shut ore]

Is Pro required? I can join if it is. I'd like to stand up a page by end of day
ah thanks

@agile jay 🍻

[agile jay]

Pretty sure you need pro, But you can check the link to see if it works on free too.

[shut ore]

"Come back soon
While in beta, Page creation is available to select users."

oh darn

Anyone here in the selected "select users" ?

:: sniff ::

I wanted to try something new instead of Notion

Also , is there a channel here for Pages?

[atomic talon]

Heyo 🙂
Is there a way to get the latex code out of equations written by perplexity ?
It's so nice to have the formulas, but I can't get them in to my obsmd vualt :/

[foggy pecan]

https://www.youtube.com/watch?v=hkHCnZ5GrNc

[fading moth]

[agile jay]

Obsidian should already support latex.

[hardy willow]

Is there any way to only have Perplexity utilize/review/reference an uploaded PDF, JPG, or URL...... ONE TIME only, in order to provide additional insight, and then continue the conversation as normal, without ENDLESSLY "reviewing file" in EVERY. SINGLE. PROMPT. from that point moving forward!?

This disrupts coversational flow, prevents forward motion, and stagnates the LLM as if it's stuck in the past, continually invoking whatever image or info was included in that picture or PDF, even once the conversation has drastically changed or progressed.

[south kindle]

https://www.kaggle.com/models/google/paligemma-ft

https://ai.google.dev/gemma/docs/paligemma

[distant aurora]

What timezone does perplexity follow for its 24hr bracket?

[distant aurora]

Just start new thread?

[agile jay]

Depends where the server was made

This one is +2 I think

[distant aurora]

And how do I find that out?

Is it common for everyone?

[hardy willow]

That's ridiculous and non-helpful. "Hey, your thumb hurts, have you tried cutting off your arm?" I didn't ask "How can I make my life and work much harder and disorganized by jumping threads every time I upload a file, because this system is flawed," I asked how I can continue making progress and get the AI to stop obsessing over an irrelevant file I uploaded for TEMPORARY CONTEXT.

[distant aurora]

You do you buddy. LOL.

Wait.

Just to be clear.

Is the continued conversation related to the attachment or not?

If not, then you are probably just being a crybaby.

[south kindle]

The suggestion was still helpful. Longer threads having context issues is a know pain point for the platform. same goes for threads started with files/photos. the model will constantly use the wrong files/information as context/sources for it's reply. currently the best way around it is to start a new thread.

[distant aurora]

I do remember it getting stuck on a particular question and spamming the same response even on different questions.

My conversation was related to the attachment though.

I was searching answers for assignments on some pdf.

[south kindle]

what, and then 10 messages later it's still using the PDF as a main source instead of moving on?

[distant aurora]

It was supposed to cuz I had to ask 50 questions from a chapter.

But it got stuck on some question.

And kept sending same response even on different questions.

[halcyon coral]

Please check this: https://discord.com/channels/1047197230748151888/1237163951750385705

[warm cave]

That would be good if it worked for the next message, last I tried its still broken

The only way to remove a source is to have it rewrite it after removing the source like shows in the video

Clicking clear sources dose nothing for me

[wheat tartan]

Is 4o the best for extracting tables from PDFs?

[warm cave]

GPT4o and Claude Opus should both be good at it, but one thing to note: If the tables are imbedded images, then it wont be able to extract them, you would have to upload those as seprate images

Basically afaik if you can highlight and copy and paste the table data yourself then the AI can read it, if not it will be skipped

[wheat tartan]

I've tried a couple of different methods, it seems that taking screenshots of the PDF and asking it to make tables yields good results

[warm cave]

Yeah it dose, I have found that GPT4o is very good at it, but sometimes it seems the vision on Claude is a little better, so if it fails maybe try a rewrite with Claude

[wheat tartan]

screenshot then "convert all of the tables to a spreadsheet friendly format"

[warm cave]

but GPT is more consistent with formatting

[wheat tartan]

I also tried uploading the entire document and prompting this: "Generate tables from page 2" but it uses the PDFs pagination vs absolute printed page, and it seems to stop to limit tokens

[cloud sigil]

how would ppx work in school assignments/multiple choice questions/ short answers?

I’m free btw I no pay

[tame current]

#📢│announcements message
is this something new or just promotion?

[devout geyser]

initial glance feels like a promotion of some kind, I'm about to briefly watch it to see what it is

oh it's not live yet

[tame current]

it premieres in 3 and a half hours

[devout geyser]

I'll be sleeping by the time it airs then 😛

[tame current]

considering the name "The Know-It-Alls" and the release date, it sounds like something like Windows Recall

[devout geyser]

yeah, could be

[young holly]

if you click the link in the video description it starts calling you know-it-all lol

[grand sand]

awww man im gonna miss the premiere :(

comes out at 2am for me lol

[gloomy shore]

CSV files work apparently. I found this out last night

[tepid compass]

Anyone knows an app for windows for TTS like the built-in one (win+h) but with an option to use external tts api like whisper?

[snow hare]

how do you have the ability to change models from the prompt screen? 😲

[young holly]

https://greasyfork.org/en/scripts/490634-perplexity-model-selection

[snow hare]

ohhh cool, thanks!

[agile jay]

Or use Complexity

[south kindle]

#⭐│starred message

[livid aurora]

What is the announcement?

[warm cave]

Idk, we have to wait

[warm cave]

Yeah, complexity is the best

[livid aurora]

I dont see the time

[livid aurora]

What is it?

[warm cave]

Extension made by feline for perplexity

#⭐│starred message

Way better than the user script I made, haha 😅. They always broke

[jaunty dove]

7pm PST or in 50m

[livid aurora]

thx

[warm cave]

What do you guys think, exciting or letdown?

[livid aurora]

thx

[jaunty dove]

Expecting nothing, just a promo video.

[warm cave]

Last time we got excited over enterprise, haha

[jaunty dove]

I would love it to be about the new pro search that some folks got, so we can see what it is.

[livid aurora]

I subscribed to Monica.im to see what it can do

[warm cave]

What is that?

Have mot heard of it

[livid aurora]

https://monica.im

[valid yoke]

in 58 mins

https://www.youtube.com/watch?v=QfoulVr6UU8

[south kindle]

it's gonna be related to the new Pro Search (i'll put 10 dollars on it)

[olive ermine]

I'm starting up FFXIV again after a few years, and I'm a few expansion packs behind. I asked Pro Search some questions like, "Without spoiling anything, I'm at this main story quest, what is the story up until this point?" and "What has changed for white mages and what's the new rotation like?" https://www.perplexity.ai/search/I-am-a-.9Wxh.lpTBWJtw2vJrYzVQ

This is why I find pro search so valuable, to build resources like this specifically tailored to my individual context.

I can ask simple, obvious questions I wouldn't be able to ask any search engine like,

Is it even feasible to complete the amount of story content I have ahead of me before Dawntrail launches? What benefit is there to preordering?

and get nuanced answers like

Playing 2-3 hours per day, it would likely take a couple weeks per expansion to get through the MSQ. Rushing is doable but may lead to burnout.

[agile jay]

The agentic intent thing?

[warm cave]

Guys, Imma post Opus 50 under the youtube comments

Too bad we can't post opus 50 images under it

[jaunty dove]

LOL, can you imagine if it is a Rick rolling video of them saying unlimited opus usage is back.

[agile jay]

Nah, they are announcing that they are increasing it to a wopping, 51 opus/day.

[warm cave]

Lol, the promise fulfilled of increasing the limit, but only technically

[agile jay]

The children in africa don't even get 5 opus a day...

[warm cave]

Every 60 seconds in Africa a minute passes 😭

[agile jay]

And every day, 50 Opuses pass too.

[calm swan]

yo why is perplexity down

[agile jay]

They are too busy getting ready for their event.

[warm cave]

Hmm let me try

[agile jay]

Which starts in 10 minutes

[warm cave]

Works for me

[warm cave]

Every person working 24/7 to get us opus 51

[agile jay]

Yep, they are in cyber space, fighting those hacker bots.

[warm cave]

With the latest cyber war grade VR goggles

[agile jay]

Yep, they are probably part of the governments new cyber warriors program.

To protect the world from rogue AI trying to steal your jobs,

[rocky radish]

i just saw perplexity logo on "the know it alls" in an nba finals commercial

what is that

[agile jay]

Oh, an event in 7 mins

Wow, they paid for an nba match commercial?

[rocky radish]

an nba FINALS commercial

[agile jay]

I was right when I said they put all their money in marketing.

[agile jay]

Who watches NBA anyway?

[rocky radish]

... me hahaha

[agile jay]

Only the Americans and Chinese.

[rocky radish]

europeans are now too

[agile jay]

Not really...

[rocky radish]

right now there's a european dominating

which is growing the sport in europe

luka doncic

[agile jay]

Yep, because we are better at everything.

[rocky radish]

lol

[agile jay]

We know the fundamentals

And the secret techniques, like passing the ball to your teammates.

I assume the Dutch team is great.

After all, they are all giants in the netherlands.

[warm cave]

@signal hamlet You watching it with everyone else?

[agile jay]

Nah, he is there in person.

They are about to confirm GPT 5

[south kindle]

hes probably part of it lol.

[warm cave]

But its a premere, so prerecorded right>

[warm cave]

i know, but he can watch it after, lol

[jaunty dove]

Just started

[warm cave]

music gen related

[patent parcel]

Germans randomly won the WC last year, weird stuff

[agile jay]

They recording it right now.

In the perplexity yellow rooms.

[jaunty dove]

I see 404. It’s a glitch in the matrix.

[warm cave]

Live is live, but on youtube premere is uploded ahead of time afaik

@south kindle Ay i see you

[agile jay]

I'm there too

On the Screen

30s

[warm cave]

which one lol

[agile jay]

10s

[warm cave]

a movie?????

oh lol

ad

hmmmm....

[jaunty dove]

The hell is this?

[lament atlas]

lol

[agile jay]

Was that it?

[valid yoke]

Same question

[warm cave]

lol, we all just watched an ad

[calm swan]

what da heck

[agile jay]

Rip, time to unsub

[calm swan]

what does know-it-all supposed to mean

i though we are going to get new feature or sum

[torn ravine]

That’s it😂

[jaunty dove]

No, we just got trolled.

Into watching an ad.

[warm cave]

Its like a superbowl ad but not during the super bowl

so just an ad

[agile jay]

Nah, they just need somewhere to spend all that Opus money.

[dense stone]

Lmao…what was that. I mean it was cute, but umm

[calm swan]

lolllll

[agile jay]

They spent 550 of our opus messages a day, to afford that ad.

[calm swan]

damnn

[warm cave]

lol, for real

[jaunty dove]

The rick rolling idea would’ve been better.

[agile jay]

And they didn't really show off perplexity at all in the ad.

[valid yoke]

That wasn’t the whole thing was it?

[agile jay]

Replace it with siri, etc and the exact same video...

[jaunty dove]

Yep, it was, we got forking trolled.

[valid yoke]

I wouldn’t have even known about it if it wasn’t for you

So I’ll blame you

[jaunty dove]

lol.

[valid yoke]

It’s your faults, both you, Somer, and Omdude

[warm cave]

@agile jay How many Opus do you think they are spending for youtube to promote the video

[agile jay]

All the Opuses they have stored in their basement,

[valid yoke]

What if the announcement ends up being that the entire ad was made by an AI?

[karmic heart]

hello.

Sorry this is a non-product related question.
but if I wanted to apply for a job position for Perplexity, how would I approach this? who can I contact?

Is there some other option other than linkedin?

@neat elk would you be able to answer me such a question please?

[patent parcel]

Just do your own startup, setup a mailbox in the virgin islands

If there's a goldrush, sell tools

[valid yoke]

The entire company is run by AI machines

No humans. They don’t do human.

[south kindle]

Curious what the production budget was for that video. I mean they don't have a generic fun ad with people in it yet. had that come out during some hype time like the superbowl or something, it would've been more impactful. i'm assuming we'll see this ad else where going forward.

[karmic heart]

well someone programs the machines, no?

[valid yoke]

No, other machines program the other machines

[warm cave]

I agree with code on why: #⚡│ask-community message

You can use Hailu with DuckDuck go

@elder tree #👑│pro-lounge message

oh ,wait one sec

its free

Its super fast to use if you setup duck as a quick search option, then just type d chat and click chat

[tranquil orchid]

Hey @warm cave!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[tranquil orchid]

Hey @warm cave!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[warm cave]

why do the bot be like dat

No problem

[south aspen]

So what was the announcement

A trailer for a future announcement?

[warm cave]

An ad

[valid yoke]

They announced that ppXX has achieved AGI

[south aspen]

Ad for what though

Ad for themselves?

[warm cave]

A fancy ad that has cinematic movie like quality

[south aspen]

Huh. Strange move but neat I guess

[south kindle]

a proper "fun" ad for the platform.

how they went about it just gave everyone here blue balls though.

[warm cave]

Like a Superbowl ad

Not 100% about what it does, juts kinda fun with a mention of pplx

Yeah, everything starts to think we are getting a massive update to the platform or something

[south kindle]

usually companies don't post "ads" on their youtube, with a hype "premiere" though.

[warm cave]

Lol, for real. If its an ad let me watch it, don't give me a 6h countdown followed by a 15min countdown 😭

[south kindle]

Heres some ads to watch while you wait for your exciting new ad to watch.

https://tenor.com/view/xzibit-meme-inception-gif-13033570

is the pimp my ride meme still relevant today?

[livid aurora]

This ad was trash …

[south kindle]

based on not having new features? or because as a fun ad it was still "bad"?

[warm cave]

? I don't know how imdian plays in to all of this. I think the real reason is they add models when there is hype around them to funnel users to the site and then remove them when there is no new people finding the site using them.

You know they only give it for marketing, its not out of the kindness of their heart

Free is only givent in the hope to get more paid users

[harsh stag]

#📢│announcements message is this a teaser for the new pro search?

[warm cave]

Idk, seems like a regular ad to me

[jaunty breach]

Can PayPal support Pro subscriptions?

[warm cave]

It looks like stripe has card, cash app pay and US bank account. So I think not

[wide mesa]

Is perplexity good at making code like ChatGPT?

[south kindle]

WTF man? racist much? go be offesive literally anywhere else.

[warm cave]

Depends, they both have GPT4o which is very good at code, and with perplexity you have we search that is really good so it can be helpful working with code that requires up to date knowledge.

Down sides: there is no continue button, so if the code is too long then it will be in two outputs

What other pros and cons… trying to think…

[warm cave]

Do you use a chromium browser?

[wide mesa]

Ye

[wide mesa]

Tbf ChatGPT does that too sometimes so it's not a big deal

[warm cave]

Ok, there are a few annoying things about the platform that I forgot to mention because there is an extension that fixes them. But its chrome only.

[wide mesa]

Hm what is it?

[warm cave]

Yeah true, Maybe you should just try it for a month, see if you like it, its hard to know until you lived with the pros and cons, if you get a referral code from any user its $10 for the first month, if you don't like it, switch bsck

[wide mesa]

I already have perplexity pro for two months now

[warm cave]

Its complexity, its in beta right now

[warm cave]

Oh nice

I can give you a link if you want

[south kindle]

look at him work.

[wide mesa]

Never really encountered any major issues besides the one time the bots decided they'll refuse to write anything for privacy reasons but it fixed itself

[wide mesa]

First off, what does it fix?

[south kindle]

we've been helping get users on board for the past 2 or so days. we've gotten better at explaining it and getting people excited lol.

[warm cave]

https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w

I made a page lol

[warm cave]

Like not being able to turn web search off mid chat

[wide mesa]

Hm that does sound good but I'm also wary of random browser extensions a bit

You also literally just asked Perplexity to write about the plugin instead of explaining it yourself. That really doesn't sell me lmao

[warm cave]

I did it for fun

I will share the actual announcement

I just like pages

[wide mesa]

You mean laziness?

[south kindle]

Well, given the "output" accurately matches what the actual features are, i think it worked out. I would assume he provided all the details about it in a thread and then converted it to a page. ha.

[warm cave]

Lol, I gave each section very specific details

[wide mesa]

I'll check it out when I'm home

[warm cave]

I feel offended and hurt 🤕 🤣

[warm cave]

#⭐│starred message

@wide mesa Also if you want to know what I hate about perplexity, look in feature requests, I have been very vocal lol

Uhh, I am sure the staff hate me at this point 😆

[quartz totem]

Can anyone explain me the restrictions of https://labs.perplexity.ai ?
How many messages can I send there (I assume there will be a limit?)

Can I assume that the messages on labs are sent without a system prompt, while those sent normally via perplexity have a system prompt?

[wide mesa]

Nah they hate everyone no worries

[warm cave]

Then I guess I’m not alone 😅

[warm cave]

I think you can send as many as you want, just dont seed them as fast as a bot would, lol, but they may use the data for training and such, not sure

about the system prompt, let me see if they use one

I cant see a system prompt, only thing i can see is they give the online models the date and time (And sources of course)

[quartz totem]

How did you check that? Did you ask the model to repeat all previous prompts?

Thank you very much for your answers!

[warm cave]

Yeah, its not 100% reliable, but it normally works

[broken tartan]

w

[honest grove]

Why can't I copy text into the process now?

already copy, but still like that.

[shell yoke]

It attaches the text

So you still have to type something in there e

[honest grove]

i know. i already copy. but still like that. i already copy many times.

[feral hazel]

What's the difference between perplexity pages and other AI chatbots which can generate public links for public to read? How are Pages superior?

[south kindle]

Pages is more similar to a dedicated web-page and/or Wiki page. much more focused on good presentation with little effort needed.

we could already generate public links for threads. (the functionality other LLM/chatbots offer). Pages is a formatted nice looking article. looking at any of the Pages already created should showcase this. example posted here: #💗│sharing message
The thread that Page was created from:
#💗│sharing message

[feral hazel]

indeed. The presentation looks nicer than the public links shared on other chatbots. Looks like a wiki page indeed

[loud parrot]

Hey @signal hamlet why can't Perlexity validate information using the web. It gets details incorrect.

[south kindle]

share a thread with outdated or incorrect info so they have something to work with.

or what your asking. specific things that are very new may have not been indexed yet.

[loud parrot]

@south kindle a very simple one for you. Let's ask perplexity to tell me who won the 4.30 at Hamilton yesterday.

[tame current]

what should be the correct answer?

[south kindle]

I'm running it through <scratchpad-think>. i'll share the results.

[tame current]

the wordware one ?

[south kindle]

a collection: https://www.perplexity.ai/search/The-user-below-4WdPVuwYQFiVc1un.nXF3w

what is the right answer so I can verify?

heres a problem, PPLX doesn't know what sport this referes to

i don't even know. I thought it was football or horse racing?

So long story short, you need to give more context to the model to get better results.

[placid coyote]

sometimes, the "pro search" is a bit too eager to search 😆

[south kindle]

congrats. you got to be part of the a/b testing for the new agetic pro search. it does multiple steps and research for each. some of us had it for a day and lost it. others havent had it yet.

[placid coyote]

didn't know it's not available yet to all users. it's usually pretty good, but there are some issues. one is what I posted (not searching enough or too much, it starts to happen more after few prompts in a same thread I think), another is what looks like it's writing two answers at a same time - front-end swaps back and forth between two very similar answers being written eavery half a second or so. it gets very distracting and a user has to wait until it finishes, stabilizes - picks one answer

[south kindle]

Well, given it wasn't even "announced" anywhere, they just rolled it out and haven't commented on it in the discord here, I can't even say for sure if that's actually whats happening (a/b testing). That's just what we suspect. We also found a bug where if you have a short initial prompt and use Pro Search, it will just do the "wrapping up" automatically and then amend part of the Pro Search system prompt into the models reply. showing "external system analsys" and 3 noes.

[south kindle]

Just don't be shocked when you go to use search later today or tomorrow it's back to the old/current Pro search. ha

[placid coyote]

yeah, I remember that one with 3 noes 😅 but it still managed to give a good answer, only the "formatting" was a bit strange. honestly, I thought pplx had agentic search for some time (at least under the hood) and this is just another iteration, newer version

for research, it seems to do better than what is available to all users now (what I remember was available before). on programming, well, it tends to ask a lot of questions, some of which it was supposed to do (like it asks the user "refactor this method what user asked us to refactor", sometimes several times in a row). so I guess it heavily depends on user's use-case (or predominant use-cases)

a bit downgrade is, pplx limited in length what a user can write/paste into this short questions from the system (during response generation, agents working). so now I often can't paste those 10 lines which the agent/system is asking for, so I have no choice than skip responding and hope RAG manages to find correct parts (method or function) on its own (from upoloaded files or previous responses/answers)

[quartz totem]

Could you elaborate what <scratchpad-think> is?
I am a new pro subscriber of perplexity and would like to learn as much as possible about it to make it most worth the money 🙂
Is these some "hidden XML-language" I have to learn to get the best results?

[south kindle]

it's just a structured prompt/framework I have added to a Collection. #1223058316662538331 message

[quartz totem]

Thank you, I will have a look at it.

[south kindle]

The version I have linked will fit into a collection if you wanted to test it that way. otherwise just copy/paste the prompt followed by your question for the model. I also just have a notepad I keep open with default prompts like that, so if I want to use it elsewhere I can fairly easily copy/paste it again.

[south kindle]

I just read that when you copy/paste a long string into those fields they are added as "files" now and you can just click on "skip".

that exists for both versions of Pro Search now.

[south kindle]

here's an example of it's use with copy/pasting the title of a news article. (one of the main ways I use it).

https://www.perplexity.ai/search/Yemens-Houthi-rebels-Y7pSGgxTQc2rUaTIwb_AGA

[quartz totem]

Thanks, I am just exploring the term "scratchpad thinking" (using Perplexity :)), sounds interesting

[south kindle]

not sure if searching will find much since it's not published anywhere lol. it's just a "systems-thinking" framework

[umbral mulch]

guys can I use 2 diff microsoft giftcards to buy smth that costs 20$

[south kindle]

it was built using antropic documentation though which mentioned them using a "scratchpad" for model reasoning. so I just made a dedicated prompt for that purpose.

[quartz totem]

Theeeeen GPT4o is pretty good at making things up, it generated some stuff that I approve as being potentially helpful 😄 (And I even missed to turn on pro search)

https://www.perplexity.ai/search/What-is-scratchpad-vjKtl.d9QdqFaBjwAkv4ig

[south kindle]

yours is a slightly different format, but same idea.

it made up it's own.

[south kindle]

https://www.perplexity.ai/search/scratchpad-9OmgWxf5QvaTxNDHUCUMQw > this is the thread where I started coming up with it.

[brave elm]

Hi! I purchased $3 dollars worth of credits for the API and now I can't remove my card / payment information. What is up with that? How can I resolve this?

[south kindle]

this isn't bad at all actually. works nice just in a different way

@quartz totem #1225549032718602332 message > this explains the models halucinations. likely didn't have enough information and just filled it what it "thought" it was.

[muted frigate]

hello

[south kindle]

last bit of spam regarding this. Had my own <scratchpad-think> review the alternate framework: https://www.perplexity.ai/search/fully-review-the-wW9gClrfRAWhD17DkDLsAQ#0

[quartz totem]

LLMs are a fascinating thing 😅

[jovial mantle]

Did they change the homepage?

[full hinge]

Know-it-all?

[cloud elk]

Why did they remove Claude 3 haiku from pplx labs?

[brittle flame]

Hellow everybody!
Do you know where Claude-3 has gone?

[cloud elk]

Cost issues?

[brittle flame]

where is Claude-3 ?

[elfin ferry]

oh yeah they used to have haiku there

[south kindle]

look back to last night. there was an entire conversation about it.

[loud parrot]

Yes it's horse racing.

[brittle flame]

Could you show the beginning of the discussion?

[south kindle]

Trying to find it

the person that started the conversation deleted their messages for some reason.

#⚡│ask-community message
#💬│general message
<where it's been mentioned> if you search for "haiku" you'll find all of them.

No word from PPLX on it yet. but we suspect it was "too" popular and the others were not getting used. also theres' some mention of the default PPLX model being Haiku now. or at least for writing mode. don't quote me on that, just based on search and the conversations over the last 24 hours. ha.

[south kindle]

Who was the actual winner? and also, do we know if any news articles were posted online for that specific day/race? that's what Perplexity is relying on to perform anything meaningful, otherwise it's looking at the wrong races, or making stuff up (something all LLMs struggle with, not just PPLX)

Further, is 4.30 a time? because the sites I'm finding have it listed in 24 hour format, so 16:30.

[austere kestrel]

it does seem to get it right (partially - both web and API get the trainer/jockey wrong)

but this is 2 days later

[loud parrot]

https://www.sportinglife.com/racing/results/2024-06-06/hamilton/801296/sodexo-live-handicap

[austere kestrel]

if the race had just been run, i'm guessing it wouldn't have the results indexed (maube if it was at Royal Ascot or the Kentucky Derby.. but not some random weekday race at Hamiliton ha)

[austere kestrel]

yeah Hey Lyla right

[loud parrot]

Correct

[loud parrot]

It ran yesterday

[austere kestrel]

yeah and it has the results now (at the least the winning horse)

oh i see - it was still 'yesterday' when you originally asked

ig in either case, it seems it wasn't indexed/accessible by perlexlity then, but it is now - but i dunno ha

i've always thought sport, weather and finance/stocks would be the hardest things for them index

people need like actual real time results

if they're asking for a sports score or something

[warm cave]

maybe an anti-hallucination prompt would help...

[austere kestrel]

it's good for major sport, companies (and news), but more niche things still (understandably) aren't as quickly indexe

[south kindle]

based on this source from "the sun", we can see Hey Lyla won. (i wrote this and got distracted by the litter of kittens in my house. posting it anyway, but seems we have this figured out. the news was "too" new to be indexed.

[loud parrot]

Correct

Now who won the 4.30 at Hamilton

Ask perplexity

[south kindle]

https://www.perplexity.ai/search/tell-me-who-DzuCADD0TF6wCFjRjyvCsw#0

i've "been' asking PPLX lol. this is a newer thread.

[south kindle]

that's a hallucination. we know it can, but the model you are 'speaking" to doesn't actually "know" a separate model is called for search and the results fed back into the thread.

[warm cave]

This is one of my anti-h prompts

lets see, i did not look to see what is right lol

[south kindle]

https://www.perplexity.ai/search/Horse-Race-results-cU1gbCanT9iVXZXim3YF7A#0 > I sorta cheated by using my scratchpad-think framework. it forces the model to reason thru the process, typically resulting in vastly more complete and cohesive replies.

[loud parrot]

That is the correct result. I guess you guys are on the paid version made that is why

[south kindle]

hold on, let me swap to another account that doesn't have Pro.

[warm cave]

oh, shoot, i cheated, i used pro search

Without pro

hmm, maybe its haiku, i will try to narrow it down

haiku may be inconsistent

[south kindle]

Rewording search prompt with non pro account/no pro search compared to the original prompt

[loud parrot]

Weird how it works for you guys but not me.

ChatGPT gets it wrong also

[south kindle]

literally at this point its not a matter of PPLX being "wrong", it just didn't have the latest information indexed yet, which it is now. so this isn't the "best" test truly.

[warm cave]

I am using a anti hallucination prompt, but non pro still got it wrong 1 out of the 2 times

[austere kestrel]

was this from just now? dpes seem odd if so

[south kindle]

Do you have anything populated in your AI profile by chance? out of curiosity

[brittle flame]

Thanks a lot, friend! What alternative to Perplexity Claude 3 would you recommend?

[south kindle]

Also how long is this thread you are working with? we've been starting entirely new threads/searches for it to clear any jank context.

[austere kestrel]

incognito browser - gets it fine 🤷‍♂️

[south kindle]

Someone more experienced with Labs will hopefuly chime in.

[warm cave]

Geting good results now with the NotebookLM system prompt

[loud parrot]

Nope new to the Group. Just implementing AI to collect some cash !!

[warm cave]

#1241988865489899520 message

[loud parrot]

Problem is if it's can't even tell me the result I wonder it's value.

[south kindle]

AI Profile is within PPLX settings. it'll amend anything in there to the thread/PPLX will review the profile at the start of every conversation. In the example of DailyFocus's NotebookLM prompt above, this could be loaded into your AI Profile to have it perform these steps / follow the framework automatically.

[warm cave]

Setup the NotebookLM system prompt in your profile, should help

I will make it shorter tho, i think its too long

[south kindle]

LOL.

It is. just convert it to JSON

[south kindle]

"it" doesn't actually "know" anything beyond what it was trained on and it's knowledge cutt off. For anything beyond that, search is needed. If search can't find the results, or they haven't been "indexed", both AI systems with search AND Google search will struggle to find what you need. It's a limitation of all LLMs, including ChatGPT. The issue here, is even though the models and various app have it drilled down to tell the user if they couldn't find anything, they sometimes will just make it up based on what makes the most sense.

personally for that situation, i entirely rely on custom prompts/frameworks to provide the reasoning process before the final answer. #1223058316662538331 message

I would highly suggest creating a collection with a custom AI prompt, and then just open new threads directly there to force that set of rules. example is here:
https://www.perplexity.ai/collections/scratchpadthink-wBPEohuUQH6tz5qMlH4F7g

[blissful pendant]

I usually use Sonar for newsy stuff. Works fine. The brevity is because of my AI profile pre-prompt too.

[warm cave]

I think we are woking with a non-pro account tho, sonar is very good

[blissful pendant]

Dang, how did I wonder into the general chat... oops lol

[warm cave]

just trying to get haiku to not make stuff up lol

New profile prompt is very concise

@loud parrot Here is an example of the search not geting good results, but the system prompt keeping the AI from making stuff up

not saying its 100% its still the just haiku

and haiku can be an idiot

[sleek vortex]

i ought to test that query on my tool 😆

not home yet though

[warm cave]

Yes, perfect test

Let’s uncovr the answer to this

[sleek vortex]

recent issue i’ve had with perplexity is the first response time is too slow

[south kindle]

I get a lot of joy out of these "thought-provoking" questions. I asked scratchpad about Indexing and the issue it can cause with LLMs and very new information.

[loud parrot]

So guys all you need to do now is tell me who wins the next race?

[unborn minnow]

Guys what's the meaning of this pinched fingers emoji

[jovial mantle]

Capische?

[south kindle]

Are you actually going to make bets on answers from an LLM? I would advise building something at least that looks at horse race data historically. you'll probably need a code interpreter.

[loud parrot]

You would be surprised what you can do if you provide the algorithm. I have a 30 year database.

[south kindle]

I guess my Starship Test 4 Page is popular. its been reposted twice since I made it up last night. thanks i guess?

[south kindle]

I know before the AI boom, people were doing sports betting on Baseball games, looking at every bit of data for years with specialized functions in Excel/Spreadsheets that would factor the probability. Point being, it can be done, even without AI, but AI would likely make it a bit easier to work with, having the AI doing the heavy lifting/data analsys.

[loud parrot]

Horse racing is more precise. Every horse has a pedigree profile to it and other factors that can be associated.

[south kindle]

My main fear currently. if I proceed helping, I may have some new fasicination on my hands that could cost me my savings lolol.

[blissful pendant]

Chef's kiss/Perfect

[loud parrot]

Based on the comprehensive system you provided, I'll apply it to the race at Thirsk on Friday, June 7, 2024.

Race Details:

• Race: FOLLOW @vickersracing ON X HANDICAP (5) (D.I)
• Course: Thirsk
• Distance: 6f
• Going: Good to Firm (Good in places)
• Runners: 10
• Prize fund: £4,187

Horse Analysis:

1. Angel Of England (9-11) - Callum Rodriguez (Edward Bethell)
   • Recent form: 1 win in last 3 starts
   • Pace figures: 82 (Competitive)
   • Trainer statistics: Edward Bethell has a 17% win rate at Thirsk
2. Biographer (9-9) - Joanna Mason (Michael & David Easterby)
   • Recent form: 1 win in last 5 starts
   • Pace figures: 47 (Plodding)
   • Trainer statistics: Michael & David Easterby have a 14% win rate at Thirsk
3. Ingleby Ivy (9-9) - Connor Murtagh (Adrian Nicholls)
   • Recent form: 1 win in last 6 starts
   • Pace figures: 41 (Negative)
   • Trainer statistics: Adrian Nicholls has a 12% win rate at Thirsk

[unborn minnow]

What if there are some Italians Among Us

[loud parrot]

Prediction:

Based on the comprehensive system, the top three predictions are:

1. A Girl Named Ivy (IRE) - 70% probability of winning
2. Alfa Kellenic - 15% probability of winning
3. Keldeo - 10% probability of winning

Value Bets:

• A Girl Named Ivy (IRE) - 8/1 (current odds)
• Alfa Kellenic - 7/2 (current odds)

https://www.attheraces.com/form/horse/A-Girl-Named-Ivy/IRE/3530199

What is interesting is this approach is from a member in my group using his blessed phone only. Works better than the desktop versions.

Pace Figure Patterns:

• COMP: Angel Of England, Keldeo, A Girl Named Ivy (IRE), Alfa Kellenic
• PLUN: Biographer, Ingleby Ivy, Baileys Polka Dot (FR), Without Flaw (IRE), Soul Singer
• NEG: Ingleby Ivy, Baileys Polka Dot (FR), Without Flaw (IRE), Soul Singer

All done with a LLM, CRAZY

[south kindle]

Explain that part more about someone using their phone over desktop. what are they accessing? PPLX? or something else?

You just mean the IOS/android APP over the desktop one? interesting.

[loud parrot]

Yes IOS. I am still surprised how the LLM behaves differently.

Using meta AI Llama 3

[south kindle]

you can also go to the web page directly on Safari, and access the desktop version. just for sake of having the option to change settings you can't within the iOS app.

[south kindle]

PPLX? 0r something else?

[south kindle]

sorry, I didn't sleep last night. but, are you saying both PPLX and Meta AI function differently on desktop compared to mobile app?

i know PPLX does, but they do strive for device parity in terms of use and functionality. ideally the platform works the same regardless of how you manage to get access to it. keyword "ideally".

Our failed attempts poisoned the highlight for Bing search. Ha. Oops.

[loud parrot]

Yep, incorrect result.

[warm cave]

Well copilot is just bad lol

Oh what

It used pplx as a source 😂

[loud parrot]

I would have thought that would be a pretty basic thing

[deft wagon]

lmaooo

[loud parrot]

For those interested https://docs.google.com/document/d/1ZxWtnHl5QAZxHtNu6-_2QC19pFjnNBIf4CbEgMOxcHs/edit?usp=sharing

Just exploring I have a database with a NN for my serious analysis.

Surely it can't be this easy

600%

Right I guess to try on PPX

Very interesting

Looks like Perplexity has a ways to go.

Ok guys sharing another one and they will depart as I don't want to get accused of spamming or anything else.

https://docs.google.com/document/d/1gJGTZxmstXAg5JdcGbdSdfDfbN7sKO8rEeEJyBS-HiM/edit?usp=sharing

You can DM me if interested on my experiments.

[livid aurora]

🤔

[shut plaza]

what browser is that

[agile jay]

Looks like the Samsung one.

Or Opera

[livid aurora]

https://monica.im/home

[shut plaza]

ah

[agile jay]

Oh, I've only seen the web version of that.

Why does it look like a browser?

[livid aurora]

It's like for searching the web.

with ai

But you can also chat without internet sources

[shut plaza]

Monica's pricing is kinda garbage imo

I much prefer Perplexity or You.com

[livid aurora]

Yeah I subscribed once for a month but I think I will cancel

[shut plaza]

Monica is like what, 200 premium queries per month?

Perplexity is 300 or 600 equivalent queries per day?

I don't remember

[livid aurora]

For 40 Dollars you have unlimited

[shut plaza]

You.com is $20 for unlimited

[livid aurora]

True

I had you.com before but it was buggy as hell so I switched to Perplexity

[shut plaza]

Perplexity is nice

I just wish focuses were one click buttons under the search bar

Instead of under a menu

[livid aurora]

Yeah but it would be cool if you can do more things with it

[shut plaza]

I also would like image generation to be a focus

[sleek vortex]

if only i had the budget to offer this pricing 😭

they can only do this because of PTUs and huge VC

[minor shard]

Yeah same. There are better models now and they need to be updated

[minor shard]

Stable diffusion 3 comes out June 12th

[livid aurora]

yea

but only medium

[shut plaza]

Now if only Midjourney had an API

Midjourney is by far my favourite image generator

[minor shard]

Midjourney profits by being closed

[shut plaza]

yeah

[livid aurora]

I want to use it locally

[shut plaza]

I would totally pay for both Midjourney and Perplexity if Midjourney wasn't $45 CAD/month for the Standard plan

[livid aurora]

true

I want to see more model in Perplexity

[shut plaza]

would be nice

I'm fine with just using GPT-4o personally though

[minor shard]

Midjourney is wayyy too expensive

[shut plaza]

agreed

[livid aurora]

ya

[livid aurora]

I'm using Claude 3 also

[sleek vortex]

like what

[shut plaza]

Gemini would be nice I guess

[sleek vortex]

hm

[shut plaza]

Mixtral?

[livid aurora]

https://discord.com/channels/1047197230748151888/1228083292956393472

Mixtral was gone suddendly

[sleek vortex]

well im guessing most people didnt use it

hence poof

[loud parrot]

Looks like a possibility for sure. https://docs.google.com/document/d/1rj-BAeTmAc02hSATc_wuRwuO5o5ID5gMi8G_ugdAPcU/edit?usp=sharing

[runic jackal]

"Convert to Page" is one of the most useful things in AI for me since the launch of ChatGPT, so thank you for it! Is there currently any way to create a page via an API or template?

Use case is: I've started using it intensively for analyzing vendors or comparing OSS products. They often have similar structures, or I know beforehand all the sections I will want to add. I would love to give a template name and a parameter and have it go off and build the page accordingly.

[loud parrot]

And 20/1 in the last omg https://docs.google.com/document/d/14gcrycsKEHY3uMNkeEYttCMW3u7nm1_HkaaLTOSrR6Y/edit?usp=sharing

The power of AI

The Monte Carlo simulation results will show the win probabilities for each horse. This combines the form summary, pace figure patterns, and dosage index to provide a robust prediction of the race outcome.

Final Recommendations Based on Monte Carlo Simulation
Johnjay - Highest probability of winning based on the simulation.
Local Bay - Second highest probability, indicating strong chances.
Rivas Rob Roy - Consistent performer with good win probability.
Ravenglass (IRE) - Decent chances based on simulation results.

10,000 simulations done in seconds

[trim timber]

What is this all about #📢│announcements message ?

[feral marlin]

the unlimited in question

[south kindle]

Just a fun ad they paid to have made. it's not new features or anything. like a superbowl ad

here is the direct page related to the latest post in #📰│server-news https://config.figma.com/agenda/session?session_id=8cda6eacbfe4&lang=en

[pallid bough]

are they fixing follow-up questions reverting back to using other sources?

[modest token]

Perplexity just told me "Maybe you should work on something simpler"... ouch

And now it told me to consider hiring a consultant... alright, who fed it the sassy juice tonight?

[agile jay]

Lol, which model?

[modest token]

the free one

[agile jay]

Probably why...

[agile jay]

The burn, though...

[modest token]

Oh it gets better...

Based on the information provided and the challenges you've outlined, generating a managed .NET wrapper for the Blackmagic ATEM SDK appears to be a significant undertaking that may not be feasible given your current circumstances. Here are some recommendations:

Focus on Learning Fundamentals
Given your friend's advice and your goal of pursuing software development as a career, it may be more beneficial to prioritize learning the fundamental languages and technologies she plans to teach you, such as HTML, CSS, JavaScript, C#, and SQL. These skills will provide a solid foundation for your career and open up more opportunities in the future.

Explore Simpler Projects
While the Blackmagic ATEM SDK integration is an ambitious project, it may be wise to start with simpler projects that align with your current skill level and the languages you're learning. This will help you gain practical experience and build confidence before tackling more complex challenges.

Manage Expectations
Integrating with complex, low-level SDKs like the ATEM SDK requires a significant investment of time and resources, even for experienced developers. Given your current constraints, it's essential to manage your expectations and timelines realistically. Attempting to push forward with this project may lead to frustration and burnout.

Revisit the Project Later
While the Blackmagic ATEM SDK integration may not be feasible at the moment, you can revisit this project later in your career once you've gained more experience and resources. As your skills and knowledge grow, you may find better solutions or alternative approaches to tackle this challenge.

[sleek vortex]

“later in your career”- wow.

[lusty mantle]

is there a way to set custom instructions

[sleek vortex]

ai profile

[lusty mantle]

found it thanks!

[wide mesa]

@warm cave Where that plug in

[fading moth]

https://discord.com/channels/1047197230748151888/1244758414794428516 Don't forget to post your Opus 50 art here! :3

[hallow tulip]

Can someone explain this to me, I've seen people recommend writing mode for regular llm style usage similar to ChatGPT, does this mode have search enabled ? or will it be a purely based off the knowledge cut off date of the model in question ?

[south kindle]

There are 2 ways to do this. Make sure to check out Collections. You can inject a system prompt it will follow. This is nice for alternate system prompts or projects.

[south kindle]

Furthering that point, you can talk to collections directly from the main search UI or main page, instead of having to navigate to the collection specifically and then start, however, it requires a third-party extension that ads a bunch of other quality of life features that are very useful for the platform.

#⭐│starred message

It’s in closed beta currently so let me know if you want to get access and I can help.

[south kindle]

The Collection Selector

[mortal bramble]

https://tenor.com/bgRSf.gif

[south kindle]

If you start a thread with writing mode focus, PPLX won't do the extra step where it converts the user reply into a search and then processes it. Once a thread is in Writing mode, search will be disabled, and PPLX will only pull from the selected models training data/knowledge cutoff.

Writing mode is specifically for generating Text and Code.

technically there is a way to "toggle" web search on and off, even in designated Writing threads, but that's a 3rd party extension in closed beta (you can request access if you end up using PPLX more).

[warm cave]

What plug-in, complexity or the code interpreter I made?

[wide mesa]

Complexity

[warm cave]

I almost forgot 😮

[south kindle]

It's not part of Complexity yet. that will take some time I'd assume. but Feline does want to add it eventually.

[honest grove]

Sorry, something went wrong.

Please try again later.

[warm cave]

DM @harsh stag he can get you beta access

[south kindle]

I think he meant the video you posted of the Pyodide user script for code interpreter.

Or I’m confused.

[warm cave]

Yeah idk, he said complexity but maybe there is confusion

Post you best Opus 50 images here: https://discord.com/channels/1047197230748151888/1244758414794428516

There are some really cool ones there

Wait I can market it better 😝

Only 1% or people can find “Opus 50” in this image, click to find out if you are in the 1% #1244758414794428516 message

[south kindle]

Take that idea further. create an "eye spy" but instead it's locate the X, in this case, the hipster.

||bottom left beanie / glasses guy||

[hallow tulip]

@south kindle What extension is this ?

[south kindle]

#⭐│starred message

[livid aurora]

lol

[glad crypt]

how can i see the queries ive made to the pplx api?

[tiny kestrel]

This is very nice!!! Can I test it as well? 🤔

[granite grail]

Can a browser plugin even do that?

[tame current]

python, probably not directly

javascript should be possible

[lusty mantle]

woa i didnt know any of this thanks!

[wide stratus]

What just happened here? 🤔🤨🤨 (2nd half of the text)

[lyric swan]

These CloudFlare checks for users that are logged on and paying you money are irritating. I want to use perplexity instead of google and one of the google's features that you don't have yet is speed.
And this anti bot check doesn't help speed. And it makes no sense for users who are already logged on and are on a premium plan and are not abusing your system in any way.

[mystic nacelle]

google isnt that bad

[south kindle]

@warm cave is the mastermind behind getting that working.

[fading moth]

[warm cave]

Thank you 🙏, ita limited compared to what you have on chatGPT, but I have only worked on it for a short time so maybe I can get more

[tame current]

Just got Perplexity Pro! Giving it a try!

[winged moon]

When did Labs remove all the non open source models?

[warm cave]

I think a couple of days ago

We call you for support Jake for snake farm

Oh, where did he go?

[twin falcon]

Im here

[warm cave]

Hello Jake, haha

[twin falcon]

I just have a question regarding the beta access, applied more than a month ago now and no reply from any team members through my email or account

I even resubmitted my request 2 weeks ago in case something went awry, still no response

Id really like access to the citations feature for my current tool, im having to prompt engineer at the moment to get around it

[warm cave]

Is it beta for the API? I too have not gotten anything and it’s been a while

[twin falcon]

Whats with the little "Web Beta Tester" in your discord profile then?

[warm cave]

I asked I one answered me, haha: #🧪│api-general message

[warm cave]

I have the beta “pages” feature

If that’s still in beta

[twin falcon]

Hm. Interesting

Maybe if I fill out the form a 3rd time itll work

So are the developers not very active in here?

I mean even no reply from the email I sent is weird, didnt only use the form

[warm cave]

Mostly mods and the CEO

You can DM him if you want, it seems to be a problem

[twin falcon]

Which one would he be?

[warm cave]

Alex

[twin falcon]

Easy enough its in the announcements

[eternal saddle]

How do I make perplexity search only scientific / academic research? I don't want geekforgeeks in my results

[warm cave]

Academic is the mode you select, but I have heard a few people saying that they are problems with it and it’s giving sources that are not academic

I think it’s due to the agentic a/b testing

[eternal saddle]

Thanks, I forgot there's a focus feature

[eternal saddle]

What do you mean

[warm cave]

They have a new search behavior that some users have and others don’t, it has multistep searches like agent search

[tidal sail]

I recently noticed that pro search uses gpt4o. There output is somehow similar to the ones in gpt4o

[warm cave]

what do you mean? like its talks like GPT4o instead of the claude model you have selected?

[tidal sail]

I'm on free version, so it talks like gpt4o

[half venture]

Opus still 50 ?

[south kindle]

There’s a form link somewhere here from @halcyon coral . Let me search for it

[south kindle]

A separate model handles Pro Search, so this makes sense. Either the actual output is from the Pro Search model or the results from the pro search influence the writing for the output.

[warm cave]

Hmm, interesting, haiku talking like GPT4o

[warm cave]

Always

😕

😞

😢

😭

[elder magnet]

Is there a way for math problems to be shown correctly in perplexity? I was looking at getting pro but that is a big negative I see with you all.

Example of what it looks like

[shut plaza]

GPT-4 Turbo and GPT-4o automatically displays math in LaTeX

if you get Pro, you get access to GPT-4o/Turbo

[elder magnet]

Oh so it would work if I subscribed you mean?

[shut plaza]

yeah

[warm cave]

You are using haiku

Claude models do not do Latex by default

But you can instruct them in a collation or in the profile prompt

[elder magnet]

Oh ok

[elder magnet]

I wonder why they don't? Is there any benefit of them not using it?

[warm cave]

Idk, I guess the training data was human readable plain text math

While GPT4 was trained on latex?

But it’s pretty easy to get Claude to use latex

Except I have not tried with haiku, so I should not say it’s easy when I have not tried, haha

[elder magnet]

Yes. I told it to use latex then it did. Weird that it doesn't do it by default.

Also for the API $5 you get for being pro, does that reset every month or do you get $5 additional each month? Like if I only used $2 would I lose the other $3?

[warm cave]

Claude Pro website dose not support latex at all which is weird

[warm cave]

You get 5$ that expires a month later and is replaced by a new 5$

[elder magnet]

Ok so I would lose any remaining that wasn't used then correct?

[warm cave]

Yes, So use it all up in msty, haha

[elder magnet]

Ok lol

What does this mean, "Note that the search subsystem of the Online LLMs do not attend to the system prompt."

[warm cave]

Where do you see that?

I assume it means that the search ai dose not see your system prompt

So if you have a collection prompt that says only search Reddit, the search ai won’t see it, but the answer ai will

[elder magnet]

Oh I was reading that on the perplexity website. https://docs.perplexity.ai/docs/model-cards

[warm cave]

Really you just have the llama models:

[fading moth]

Hrmm

[elder magnet]

Ho do I get access to the Pro channel? I am a pro user now

[warm cave]

leave the discord, and join with the link under you settings page on perplexity. this will work unless the discord bot is broken again, if so you will have to ping a mod

[livid mantle]

leave the discord

[south kindle]

#📘│faq message

[warm cave]

Recently, chatGPT has felt like it has a very short context window. I will tell it to use a particular format of latex, and it will make a few things with that format and then revert, or I will be asking it questions about some text, and it seems to forget it very fast. I think this is just GPT4o being tunnel vision, I am switching to GPT4T to see if it is any better. But I don't remember being this frustrated when it was GPT4T

Or just use Opus, but then I will run out too fast, idk

[fading moth]

Hmm

4o has been a mixed bag with me since the start

[kind lichen]

Is the drawing function not available now?

[modest token]

Finally! After more back and forth with Perplexity, I finally managed to get my cross-platform DLL file made... THANK GOODNESS! Now I can get back to writing my code with the updated DLL.

[neat quiver]

You can say that again - https://www.forbes.com/sites/sarahemerson/2024/06/07/buzzy-ai-search-engine-perplexity-is-directly-ripping-off-content-from-news-outlets/

Here I was thinking the most obvious copyright complaint - about reproducing long sections of unmodified copyrighted inputs 1:1 at the output - would be the easiest to avoid, but it seems not quite.

Sad to see this, as I've been using Perplexity exclusively (no GPT, no others), because it focuses on one of the few things an LLM can do very well (and relatively energy-efficiently, I hope?): be a better search engine for text resources. A lot of the other things, including the insane dream of a neuralnet-based AGI, I think are just a waste of effort in a wrong architectural direction and a drain on our energy systems that humanity doesn't need right now.

[fluid mauve]

When I read this article, there were three takeaways.

1. The journalistic Publishing industry, full of paywals and monitization schemes is 'crying' once again that 'they stole our work'
2. Perplexity stated that they would need to improve the size of icons and attribution. they admitted it had "rough edges" but " they "...agree that contributing sourcers should be highlighted more prominently"
3. CNBC, Bloomberg, and Forbes were asked for a comment but none was given which seems kind of 'funny' as teh article is written by forbes.

The only other thing that stood out was that Perplexity had copied and modified a Forbes illustration. Though not an original artwork, it is distinct and unique.

I dont really condone copyright infringment, but I am also not happy about publivcations that cry infingment, are asked to comment, and then choose not to. Smells a bit fishjy to me.

I couldnt even get to the end of the article before i get hot with the 'payup to continue reading'

[sturdy compass]

If im buying the pro version, do i get 300 messages with gpt4-o and all other models? Or is it only perplexity own model?

[sleek vortex]

600 all models (gpt4o and others) in 24 hours, but opus is 50

[sturdy compass]

Thanks brother, so they getting resettet every 24hr?

[sleek vortex]

No like when you use it once, that usage comes back after 24hrs

To prevent abuse from somebody from doing like 600 at 23:55 then 600 more at 00:00

so if you use 50 at 6pm those 50 come back at 6pm next day

[sturdy compass]

Ahhhhh makes much sense, thanks for the help. Sounds like a very good deal.

[sleek vortex]

Yeah
the way they get it so cheap is that now that they have tons of users they can just pay for like a rate of throughput (say like allocated x tokens per second to pplx) of the model on cloud services like aws, azure instead of paying per token

like thats really expensive (like starts at 30k per month) but much cheaper than per request if youve got like 500k users

[sturdy compass]

Yes, thats true. You getting a lot more here i guess. A lot of models and a lot of messages. Compared to chatgpt.

Does all model using search engine?

[sleek vortex]

The way it works is

if you have pro search off, it feeds keywords from your query into a search then adds it into the model's context
if pro search on, it asks a smaller model (sonnet?) to turn your query into searches, it does those in steps, then adds the final results to the main models context for the response

though you can also turn it off with "writing" focus

[sturdy compass]

Daaaaamn thats Nice. Thanks man, I appreciate it.

[uneven perch]

Hey, my Perplexity Page was deindexed, seeing a noindex tag on it. Anyone else seeing this? Is this a new policy they rolled out for Pages?

[cloud sigil]

anyone know for chatgpt website, does it use 3.5 or 4 for basic version? like im using it for free

[tame current]

gpt4o, but has limits and will switch to 3.5 after limit runs out

[cloud sigil]

do u know the limit?

[tame current]

around 10 messages per 4 hours

[cloud sigil]

so its basically gpt4o for free now?

instead of 3.5

but u got restraints

[tame current]

yes

it is much better than 3.5

[cloud sigil]

can u also recommend other ai's to help with stuff like school?

[tame current]

and you also get code interpreter, web search, image upload

[cloud sigil]

idk what those are

ik ppx, chatpgpt, bing ai, and you.com

which ones are reliable tho?

[tame current]

code interpreter can run code, so for example if you need to solve a math problem, it'll use python code to get the answer

[tame current]

chatgpt and perplexity are best

imo

[cloud sigil]

you.com does use gpt4o no?

idk about bing. a bit sketch

[tame current]

but free version has a limit of 5 per 24 hours

[cloud sigil]

o ic

[tame current]

bing isn't that good, and slow

[cloud sigil]

ya its slow

i heard about claude ai but it says not available in my region

[tame current]

there are a couple of perplexity clones too like morphic.sh

claude is okay I guess, nothing special free version is sonnet which is not particularly good

[cloud sigil]

never heard of it

[tame current]

try it then

[cloud sigil]

k thx

i'd prob keep sticking nto my gpt, ppx and sometimes you.com

ig i can switch to you.com when im out of tries for gpt lol

[tame current]

you can also try https://sdk.vercel.ai/ won't save conversations but it's free and has a higher limit than chatgpt

[cloud sigil]

wow thanks so much

also i don't know a lot about ai rn but what would u say regarding the accuracy of gpt4o rn?

and its accuracy on like school related topics

[tame current]

basically it's the best model available right now BUT they still haven't solved the hallucination problem

it will still hallucinate on obscure topics

[cloud sigil]

also do u know about turbolearn ai?

[tame current]

but generally it's great

[cloud sigil]

ic ic

[cloud sigil]

@tame current last question before i go 🙂

[tame current]

nope

free tier seems very limited...

[cloud sigil]

ya its just smth cuz i don't wanna read textbooks in my class 😭

any particular recording ai's that are becoming more popular? like summarizing stuff

[tame current]

don't know of any, sorry

there's https://aistudio.google.com/ which allows uploading video but idk

never tried it

[cloud sigil]

i'll take a look thx btw

[loud parrot]

Why can't a basic search be used Step 1. Locate the races for today
• Type of races (3-year-old+ Flat Handicap races).
• Source https://www.attheraces.com/racecards

Comes back with nonsense.

[turbid iris]

Hi, I have a question: which artificial intelligence model does Perplexity use when it answers me? I use the free version.

[jovial mantle]

GPT-3.5 I think

[tame current]

claude 3 haiku

source: #👉│feedback-general message

[turbid iris]

Thank you!

[bright rock]

best api for coding?

[warm cave]

What kind of api? Here are some of the options with prices if you are talking about LLM’s

Llama and Haiku stand out as good budget models. Power wise GPT4o likey takes the win

[livid aurora]

Where is Claude 3

Guys how can I create a business plan with Perplexity

[warm cave]

Third image

[livid aurora]

ups

[fossil sentinel]

This pdf is cutting it close 😂

[warm cave]

where you get so much text, haha

Or you may be using video, in that case its easy

[fossil sentinel]

University Physics textbook by Young and Freedman

Hefty thing

[warm cave]

well lucky you, that was on the edge of being too big, haha, let us know how Gemini preforms

[south kindle]

I’d be curious to come up with a few test questions to verify if it can locate specific information from it.

@dapper flicker as a “tip” when sharing threads, it’s good practice to properly name the Thread. By default it just uses the first query. This helps people know what you’ve actually shared in #💗│sharing message

[warm cave]

Like needle in the haystack?

[fading moth]

Hrmm

[cloud elk]

Can someone enlighten me as to why Claude 3 Haiku was removed from Perplexity Labs?

[warm cave]

They don’t give us reasons for decisions they make, i assume the labs was a marketing move to get more people to know about perplexity, but now they are cutting down the models since non of them are hyped up.

If you want free haiku you can use duck duck go chat

[cloud elk]

To be honest i don't really agree with the marketing thing you said. But then again, this is only speculation.

[warm cave]

WHy do you think they added them?

[cloud elk]

Well i was aware of that but the model seems quite heavily censored over there.