💬│general

[cinder comet]

i did

[livid mantle]

HAHAHAHAHHAA

[cinder comet]

how did you know

[livid mantle]

HAHAHAHAHHAHA

so cute sorry

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@serene sand
@livid mantle

[cinder comet]

am i hacked or what

[livid mantle]

I dont know the best to answer your question but I occassionally change my discord name to a LLM model that comes into my mind but also not necessarily implying that it is my favorite if that makes sense

[full gyro]

🎃 Halloween Pumpkin Collection Leaderboard 🎃

🥇 @minor shard (𝓙𝓪𝔀𝓷𝔃𝓮): 6 🎃

🥈 @serene sand (mares): 6 🎃

🥉 @sleek vortex (sneakyf1shy): 5 🎃

4. @last storm (kentavius): 3 🎃

5. @halcyon coral (IceLavaMan): 2 🎃

6. @rich sphinx (WhatIsAModel): 2 🎃

7. @livid mantle (GPT-5): 2 🎃

8. @native latch (humbled): 1 🎃

9. @brazen vine (Space_Army8485): 1 🎃

10. @wet quarry (CoreMT89): 1 🎃

Page 1/3

[livid mantle]

im on top

!!!!!!!!!

[minor shard]

Wooo

[livid mantle]

you are the one with cartoon 3d modern haircut aladin pfp

your name is now unreadable

[minor shard]

[livid mantle]

@cinder comet The Discord client fails to fetch a user for the input 13376666666666666666666666666669 because its search algorithm likely performs an early rejection of extremely large, non-standard numeric strings, assuming them to be invalid or unintended queries, while “Asura” matches its expected alphanumeric pattern for username search indexing.

[cinder comet]

are you really a person??

[livid mantle]

https://tenor.com/view/yuki-kaguya-shinomiya-gif-24132804

[livid mantle]

whats ur opinion about sonnet that can control your computer in claude

I find it beneficial for people with disabilities, and I believe it’s efficient for those of us with full body mobility.

[cinder comet]

its still in infant phase

[dense birch]

I didn't even reach 15 message yet with claude today, its already forcing me using gpt4

[quiet hull]

oh hey

[drifting stone]

Welp, same here lmao.

[dense birch]

https://www.perplexity.ai/search/translate-3500-characters-the-tN2IwcJAQP.UVrNOJJExkQ
GPT-4O has been broken for translating text since a month ago. When I translate 3,500 characters (which Claude capable of it), it works well for the first half of the text. However, when it reaches the second half, it becomes gibberish, often filled with many — and ; symbols between the text. Even though I have already asked for a revision, I still receive the same response

[cinder comet]

https://x.com/btibor91/status/1850919619629854875

Tibor Blaho (@btibor91) on X

Meta is reportedly building a web search engine led by engineering manager Xueyuan Su for at least 8 months to break free from using Google Search and Bing data feeds, with some websites like NY Times already blocking Meta's web crawler

- Meta AI has 185 million weekly users

Twitter · Oct 28, 2024, 3:16 PM

hmm?

[minor shard]

Lol the race to the best internet connected AI

WHO WILL WIN?

[harsh verge]

Competition can often drive innovation and improvement. I hope the true winner will be the user.

[junior reef]

so far im seeing very different use cases in how each LLM approaches search. Will be interesting to see how they converge

[minor shard]

Im sure with time they will all be doing the same thing just will look different

[woeful slate]

هل ممكن تحليل الملف

[ember bramble]

Can perplexity get a amoled black dark theme? like discord midnight theme.

[serene sand]

https://www.perplexity.ai

https://www.perplexity.ai/hub/faq/how-does-file-upload-work?utm_source=perplexity

[placid coyote]

that is ... an interesting ... programming tool use 😄

[harsh verge]

I just sent her a picture of my soup to see if it was too salty

[radiant badge]

hello

[minor shard]

LOL imagine using programming to answer all your questions

[sleek vortex]

the ai hallucination psuedocode update

[sacred pike]

do you need to have premium to make pages

[fast meteor]

i think so

[rocky crow]

#📢│announcements message
I think that whoever wrote that, have misunderstood the expression "A know it all" thinking it to be a positive thing. Nobody likes a "know it all" , meaning nobody likes anyone thinking that they know everything, we call this ignorance 👀🤨😎

[modest crag]

I have a question: Will Perplexity become obsolete now that SearchGPT is available?

[mint carbon]

No I don’t think so! At least not just yet. I pray that Perplexity takes the opportunity to go into a new monetization method. The Subscription model is not going to work for them in the long run!

[valid iris]

@signal hamlet when will perplexity go back to giving me 25 sources? I pay $20/M, and every question I ask I turn on pro, and only get less than 10 sources.

If it won't give me a lot of sources, then it is no better than ChatGPT's own search function

[warm monolith]

what a deliciously horrible idea ... welp it was nice while it lasted

[valid iris]

Which idea?

[valid iris]

If searchgpt becomes public and free, or can at least compete in value with perplexity pro, and perplexity doesn't increase the amount of sources it uses, then yes

I was hoping that perplexity would increase the amount of sources it provides, but it Seems to be doing the opposite

[modest crag]

Good point! I'm enjoying this...

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@modest crag

[valid iris]

Wow, so perplexity is literally asking its PREMIUM subscribers to do free labor and create pages for them, and one their premium subscribers have this privilege.

imagine if YouTube only let premium subscribers make videos, or any other popular social media site only let is premium users publicly post content.......

Imagine paying money to create free content🤦‍♂️

[valid iris]

You mean the discussion of the progression of ai tech?

[modest crag]

Yep..

[minor shard]

I mean, until they have 2 mins of ads like Youtube does, they have to make money somehow.

[valid iris]

But the money doesn't come from the discover pages, it comes from subscriptions, that's the main reason why people pay for perplexity isn't the discovery feature but rather just the pro search feature

[minor shard]

Yeah they make their money from the subscriptions and not ads currently.

[valid iris]

But it would be less egregious to expect their free subscribers to make the discovery pages, because paying for the privilege to do free labor is just silly

The only justification is that the premium subscribers get access to most of perplexity's power, which is necessary for making the pages

[minor shard]

Oh well I personally don’t use the pages feature just so it shows up in “Discovery” tab. I’m assuming perplexity looks for people who want to curate the content in there. I like the feature for sharing content with others. Not for the discovery tab. I’m not using it for perplexity, it’s just another tool/feature for my use.

[minor shard]

Yes also true. Pro search allows for more in depth searches

[valid iris]

The announcement says that curators will CREATE pages

and only premium subscribers can do that labor unless they give curators with a free subscription that privilege

[livid mantle]

1. ChatGPT-4o-latest (2024-09-03) - Arena Score: 1340 (+4/-3)
2. o1-preview - Arena Score: 1335 (+4/-4)
3. o1-mini - Arena Score: 1308 (+4/-4)
4. Gemini-1.5-Pro-002 - Arena Score: 1303 (+4/-4)
5. Gemini-1.5-Pro-Exp-0827 - Arena Score: 1299 (+4/-3)
6. Grok-2-08-13 - Arena Score: 1290 (+3/-3)
7. Claude 3.5 Sonnet (20241022) - Arena Score: 1286 (+6/-6)
8. Yi-Lightning - Arena Score: 1285 (+4/-4)
9. GPT-4o-2024-05-13 - Arena Score: 1285 (+3/-3)
10. GLM-4-Plus - Arena Score: 1275 (+4/-4)
11. GPT-4o-mini-2024-07-18 - Arena Score: 1273 (+4/-3)
12. Gemini-1.5-Flash-002 - Arena Score: 1272 (+5/-6)
13. Llama-3.1-Nemotron-70b-Instruct - Arena Score: 1271 (+5/-7)
14. Phi-3-Mini-128k-Instruct - Arena Score: 1271 (+12/-12)

The rankings are based on user votes from Chatbot Arena, using the Bradley-Terry model for evaluation. All models listed here have been ranked according to their Arena scores, which represent their overall performance.```

[sacred crag]

https://www.perplexity.ai/backtoschool -- is this still a thing? I signed in with a school email address (would have been nicer to link a school email to my existing account with my chat history) – and it says it's $4.99 with $20 crossed out, but the payment page still shows $20

[glad otter]

Any news release from.perplexity on grok llm being added ?

[fast meteor]

pages generated by pro users will not be indexed by search engine

[cinder comet]

is it worth it

i mean they have what you need

gpt4o sonnet 3.5

arent these the peak rn of LLM

[tropic verge]

Different training dates, other answers. Less restriction. Except perplexity, the model messes up like with Sonar Huge and Llama 3.1 405

[cinder comet]

ive tried it

its lacking a lot

so i dont see the need to add it tbh

[tropic verge]

I talk about the upcomming in december

[cinder comet]

grok 3

depends on the cost vs quality

if it doesnt provide any additional value compared to other models then its not worth it

but ive heard they are on reasoning training paradigm

what openai did with o1 they are doing the same with grok 3

not sure

[serene sand]

#1281517990310842409 message

https://x.com/aravsrinivas/status/1851178448007635017?s=61

Aravind Srinivas (@AravSrinivas) on X

Grok 2, xAI's latest model, is available now to all Perplexity Pro users. You can go to your Settings and pick Grok as your default model.

Twitter · Oct 29, 2024, 8:25 AM

[valid olive]

bruh

[drifting stone]

I am not seeing this Grok 2

[cinder comet]

eh?

[serene sand]

https://x.com/aravsrinivas/status/1851188395814633734?s=46

is this guy trying to steal our job, or what @signal hamlet @halcyon coral @wraith crow

Aravind Srinivas (@AravSrinivas) on X

@sahilpng On the "Focus" part of Perplexity searches, you can pick "Video" and search over Youtube.

Twitter · Oct 29, 2024, 9:04 AM

[sleek vortex]

who knows…

[upper iron]

Hey Ai experts

Grok 2 is uncensored? I guess?

[cyan socket]

its not uncensored

[prisma crown]

Is spaces working fine for everyone here? For me it refers to files only for the first prompt. After that it defaults to web.

[fervent summit]

Did you get any feedback on this I feel the same, I subscribed hoping this was going to be a strength and its fairly lacklustre

Hi, here are some ideas,

1. To uncover the change in the direction and trends in earnings calls, perhaps in most quarters the company talks about ABC and then in the current quarter they dont mention that item. Or Perhaps mentioned something new.
2. To uncover changes in patterns / seasonality in quarterly or annual fianncial reports.
3. To be able to extract notes from quarterly and annual reports - 10K and Q's - i.e. notes that offer supporting info. I.e. please extract the Capex spending section from the past 3 years.
4. To be able to access and export from EDGAR - The SEC's database (that has API's)

[mortal panther]

Same issue here

[minor shard]

Is there anything Grok would be good for? Other than just a different style.

I guess if you want access to data from X but I feel X has become slightly worse at getting truthful/accurate data 👀

[dim wind]

bruh perplexity is really bad for coding

[minor shard]

Which model did you use?

Unless it’s just free version

[dim wind]

tried every model. It's just not optimised for coding

made me go crazy

[minor shard]

Did you try in writing focus?

[dim wind]

yeah

[minor shard]

Oh idk then. Did you also try the “Reasoning” focus? I know o1 mini model is sometimes good

[dim wind]

just one example: gave an image to it showing an error. The image got stuck in the conversation and the model just started answering the error on the image every message

[minor shard]

lol oh

[dim wind]

so using images are out of the picture

o1 is also not using reasoning for me?

just answers without thinking

like other models

idk just gonna try claude next month I think, i have money for only 1 thing per month lol, tried chat gpt once and it sucked for multi modality for o1, though it was good because of it's long answers, just a shame for not having multi modality or internet access for o1

[valid iris]

By which search engine? You mean a Google, bing etc, or do you mean perplexity?

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@valid iris

[valid iris]

Wait, are any perplexity pages indexed by any search engine?

[full gyro]

🎃 Halloween Pumpkin Collection Leaderboard 🎃

🥇 @minor shard (𝓙𝓪𝔀𝓷𝔃𝓮): 6 🎃

🥈 @serene sand (mares): 6 🎃

🥉 @sleek vortex (sneakyf1shy): 5 🎃

4. @last storm (kentavius): 3 🎃

5. @halcyon coral (IceLavaMan): 2 🎃

6. @rich sphinx (WhatIsAModel): 2 🎃

7. @valid iris (blackswanblind): 2 🎃

8. @livid mantle (GPT-5): 2 🎃

9. @native latch (humbled): 1 🎃

10. @brazen vine (Space_Army8485): 1 🎃

Page 1/3

[empty lotus]

Does pplx Pages have option to hide the prompt. Just the answer alone

[empty lotus]

Gdork shows some pages
"https://www.perplexity.ai/page/"

[empty lotus]

I think this will make the page look compact, just valuable info and not my prompts

[serene sand]

https://x.com/perplexity_ai/status/1851315707411337435?s=46

Perplexity (@perplexity_ai) on X

We're excited to partner with @github. With our GitHub Copilot integration, you will be able to:

• Stay up to date on the latest Library updates, like “latest updates in React”
• Quickly find answers to any question, like “how to build a server side component?”
• Access

Twitter · Oct 29, 2024, 5:30 PM

[upper iron]

Awesome

[orchid dust]

Hi! I'm using Spaces and set them up to search only in uploaded files. This works for the first query, but on follow us it goes for online sources. How do I make it search files only?

[halcyon coral]

we're dropping softwear tomorrow at 9am pacific time
https://perplexity.supply

[fast meteor]

everyone who respects no-index

is this a bug or a feature. the font for Writing is intentional?

[serene sand]

https://x.com/perplexity_ai/status/1851341906271408469?s=46

Perplexity (@perplexity_ai) on X

Perplexity Supply. Merch, made for the curious. Drops tomorrow: https://t.co/o32T8Hm6o9

Twitter · Oct 29, 2024, 7:14 PM

[urban egret]

Focus selection/mode seems to be missing from Enterprise accounts -- is there a way to fix this?

[craggy onyx]

Omg I spent like an hour and a half on the curator application and it disappeared. Is there a way to get it back? Pls say yes

[jade coral]

is perplexity now using the latest models cluade 3.5 and such that was just updated a few weeks ago?

[serene sand]

known bug, will be resolved asap

[urban egret]

thank you!

[serene sand]

thank you for reporting 🙏

[rugged thorn]

I'm not able to see Collections options in the Library tab, why am I not able to see or create collections?

[harsh verge]

The 'Collections' section has been renamed to 'Spaces' and is now located in the sidebar.

[rugged thorn]

Thanks so much!

[jade coral]

is perplexity now using the latest upgraded models Claude 3.5 and such that was just updated a few weeks ago?

[tired fractal]

Perplexity, let me ask you this. Why o why do you not give us tempo control even though everyone and their mother is crying about it?

Why do you hate us so much when we show you nothing but love?

[fierce sonnet]

hey 👋 curators get free Pro while they’re in the program!

any other feedback or questions on the program, just let me know. we definitely want to make sure we’re providing value for our curators

[fiery latch]

YES ! I'm getting so sick of how boring and repetitive sonnet is, increase the f*cking temperature !
I swear I'm not renewing my sub if they don't add a temp slider

[tired fractal]

@signal hamlet

Can we get an answer on this? "Look mates, we can't give you temp control because of X."

An official stance.

[sleek vortex]

too bad. alex isnt the backend guy, has no real control over that

[waxen jetty]

hi, my school won the pro thing.
Is this the perplexity pro server?

[sleek vortex]

yes

well also the normal perplexity server

you just get pro role

[waxen jetty]

ah

ty

[placid coyote]

🤖

[serene sand]

https://discord.com/channels/1047197230748151888/1111786888626438245

[halcyon coral]

New NFL widgets! Get game summaries, team comparisons, player stats, and more. More sports coming soon...

[fleet magnet]

Awesome

[static ingot]

how to set it up?

[serene sand]

https://x.com/perplexity_ai/status/1851315709533655116?s=61

Perplexity (@perplexity_ai) on X

Both Perplexity and GitHub Copilot users can install the integration for free via the GitHub Marketplace: https://t.co/H43FXdAXkH

Twitter · Oct 29, 2024, 5:30 PM

[tired fractal]

File upload doesnt WORK AGAIN?!!!!!!!!

https://discord.com/channels/1047197230748151888/1140622008086970420

[fleet magnet]

it´s working for me

[tired fractal]

Now it works.

[wispy forge]

https://www.perplexity.ai/search/which-ai-model-is-this-iAHZQqteQOGHkuVfIyNN3A

confused ?

@twilit sierra @fierce sonnet @toxic moss

[wheat lantern]

hi, is there anyone has tried the copilot marketplace pplx in the github copilot chat?

I am sticking in the connect stage. When click the connect bottom, it is always show the message: This GitHub App must be configured with a callback URL

[hard veldt]

Are there issues with spaces now? The page turns blank after I click on any of my spaces, either on desktop or mobile web

[copper knot]

who's having weird issues whenever they click on spaces it just reverted as white screen

got this provision header error

[rare socket]

have the same issue

[old knot]

Loading main page on safari and then screen goes white

[rain silo]

aye that

[sterile jacinth]

Same issue blank page on edit spaces

[livid mantle]

envelope/?sentry_key=bb45aa7ca2dc43b6a7b6518e7c91e13d

General:
Request URL: https://www.perplexity.ai/rest/metrics/collect?version=2.13&source=default
Referrer Policy: strict-origin-when-cross-origin

Request Headers:
Baggage: sentry-environment=dev,sentry-release=8kPZrFCaO3cfUbUFbgYoP,sentry-public_key=bb45aa7ca2dc43b6a7b6518e7c91e13d,sentry-trace_id=6a717693f81e4f2aa0f3cfbb8683d60b,sentry-sample_rate=0,sentry-sampled=false

Content-Type: application/json
Referer: https://www.perplexity.ai/collections/jw-s-research-r1LpiACTXG9dZbD_pRQQ

Sec-Ch-Ua: "Chromium";v="130", "Google Chrome";v="130", "Not?A_Brand";v="99"
Sec-Ch-Ua-Arch: "arm"
Sec-Ch-Ua-Bitness: "64"
Sec-Ch-Ua-Full-Version: "130.0.6723.70"
Sec-Ch-Ua-Full-Version-List: "Chromium";v="130.0.6723.70", "Google Chrome";v="130.0.6723.70", "Not?A_Brand";v="99.0.0.0"```

[magic sierra]

Same issue here - blank white page whenever I click on Spaces

[livid mantle]

let me tell you why

casper is white ghost

and spooky is soon

perplexity whats to scare you early

you are welcome by the white ghost

[valid iris]

You mean the speed at which it types up the response?

[tribal blaze]

Same here. I can't get access to my custom AIs

[fresh slate]

any way to recover the deleted thread?

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@fresh slate

[warm perch]

@stable peak whats up

[stray imp]

ハロウィンの招待状　クラシックなやつ

[halcyon coral]

Hey @wispy forge!
Please check https://discord.com/channels/1047197230748151888/1225549032718602332

[hardy quartz]

hi im a new user can i get a discount?

[thorny pewter]

Hallo, how many Pro-Searches do I have in my Prosubscription?

[frozen badger]

I would lile to return the Pro Plan within my 14-day right of cancellation in EU. Where can i cancel and ask for my money back…?

[wraith crow]

You can find the steps on how to request a refund here https://www.perplexity.ai/hub/faq/billing-and-subscription

[wraith crow]

Currently you have 600 pro searches a day with a pro subscription.

[frozen badger]

Danke Spezi

[noble island]

was the option to use focues search in spaces removed recently? I always used that feature but not it seems to be gone, both from the native app and from the pwa

[serene sand]

https://x.com/pklada/status/1851411288133681368?s=61

Pete Lada (@pklada) on X

This was a fun one to build - lots of little details we were able to sneak in. It’s still early days for sports on perplexity, let me know what you want to see next!

Twitter · Oct 29, 2024, 11:50 PM

[wet nimbus]

Same. Seems like everyone does. Must be something wrong on Perplexity's end

[stark matrix]

Hey everyone! I’m curious: it seems to me that with Perplexity Pro you already have access to all the current relevant models (GPT-4o, Claude Sonnet 3.5, Grok, etc.), so why would someone pay for each of those separately (ChatGPT Plus etc.), when you can just pay the same price for PP and be done with it? what’s the catch? am I missing something?

[summer halo]

is perplexity down?

[native bolt]

can i add external plugin into perplexity?

[sly yarrow]

can I use perplexity to convert the pdfs to docx?

[stoic wave]

Why would you need perplexity to do that? theres millions of tools for that purpose

[kindred lark]

I'm trying the voice mode on the desktop app and it only let's me speak for about 3 syllobols before it starts responding. I've tried talking really fast to not give it any pauses, but it consistently only giving me about 2 seconds before responding. Anyone else having this issue?

[thorn stream]

if you dont mind not having all the same features claude/chat gpt might not have, such as web browsing and code execution and image generation, then perplexity is suitable for you. just be weary that their context limits are not the same

[valid iris]

You get access to multiple AI APIs with perplexity, but not multiple premium subscriptions to each of those ai models apps. So even though I pay for perplexity, I won't get premium access to the ChatGPT app, which includes all of the custom gpts that users create themselves

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@tired fractal

[tired fractal]

File Upload stopped working for anyone else?

[zealous bough]

why don't sources appear whenever I'm using the academic focus and asking a question?

[tame current]

your file upload feature is broken and doenst allow for files to be gotten rid of in the conversation and they buggily just stick around during the whole conversation. how to turn this off.

[zenith venture]

Is this a bug?

[south kindle]

Does anyone recall the community wanting merch lol. How’s that VC money pool doing?

[wheat pike]

perhaps too many friday pizzas were had

[trim timber]

What’s that guys?

#📢│announcements message

[sleek vortex]

ill take a perplexity supply cybertruck too

[worldly jungle]

Where can I get this code?

[warm perch]

If you pay for uber one you can get a code

[wild bridge]

o/

[wild bridge]

Some ISPs give memberships for rewards.

That's how I got mine.

[urban egret]

Anyone else experiencing issues with file upload in spaces and organizational files? Working fine in standard thread, but not the other spots.

[mint carbon]

Yup that is how I got my code!

[ionic stone]

Does Perplexity Free search default LLM still use Claude Haiku?

[low spruce]

guys the CMD shift P launcher for perplexity is dope

but just one thought. CMD-Shift-P is the default VS code or IDE launcher keybind for many. Which results in some KB conflicts

I went with ctrl-space instead, but just something to note🤓

[dusky quarry]

Does anyone have tips for using Perplexity to retrieve data from a specific website?

[valid yoke]

No, but brave has a built-in LLM where you can ask it to answer questions specifically from that page

[low spruce]

I think it supports some search operators like '+' and site:google.com

but it might not be well documented

i'm not 100% sure tbh, but heard stuff around this

[dusky quarry]

thanks for the tip!

[torn sierra]

The file handling here has never really worked well, and I doubt it ever will 🤣 They keep lingering as the current context 😡 There’s no nice tracking of files and what’s happening with them, like ChatGPT does.

[low spruce]

Yeah has never been great for RAG

[harsh verge]

https://discord.com/channels/1047197230748151888/1262520766642131054

[mystic marten]

Hi folks, hoping someone can give some guidance. I received the free year of Pro from Github Universe, but when I try to claim the code it says Invalid promo code. Have they not activated it yet or is there some kind of screw up?

[sleek vortex]

#💬│general message

but yeah, I doubt that has changed

[jade breach]

After 100 prompts, previous ones start disappearing, is there anyway to get them back without deleting recent ones? is it some sort of visual bug?

[sleek vortex]

visual bug, yes

[abstract patio]

I have a similar issue with the promo code from GitHub universe, when I'm on the webpage it works but on the android app it doesn't recognize pro

[jade breach]

ah, the ai seems to lose context from the prompts lost from the visual bug, thats why I was askin

[sleek vortex]

perplexity loses context.... anyway

[jade breach]

oof

[abstract needle]

how to generate images ?

I can't even see the option

[serene sand]

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

[abstract needle]

thank you so much

[fickle sorrel]

consensus vs perplexity. i feel liike consensus is way better for medical research

has anyone tested both a lot?

[indigo valve]

Anyone get searches that hallucinate sources?

[wraith sentinel]

Hey everyone, how can I contact the Perplexity support ?

[serene sand]

[e-mail]

[wraith sentinel]

Thank you!

[inner herald]

Why did they cancelled my Pro Plan?. I never asked for a downgrade just registered today and I have hardly even tried it because I was learning how to…

[tired fractal]

https://discord.com/channels/1047197230748151888/1290110481499033650

go and spam ping devs there

im the only lunatic screaming

[inner herald]

#1301101060491841546 message BRING IT BACK

[indigo valve]

How do you guys like Grok 2 in Perplexity?

[cinder comet]

its fun

and thats about it

nothing crazy

[indigo valve]

What makes it fun?

[cinder comet]

it would be great if there is an added value like accessing twitter

[cinder comet]

uhm the way it responds i guess

[indigo valve]

Like it has more personality?

[cinder comet]

maybe

i havent used it much but its fun for some reason

[inner herald]

#1301101060491841546 message

1301296328176828549

Shame. Someone had to say it 💅

[shut light]

it’s not a scam if the code was leaked

you didn’t lose anything

[kindred folio]

what personality is it using? i’m assuming you can’t choose

[wintry quartz]

How do I know how many pro searches I’ve used?

[wraith crow]

You can check your total usage and the amount of usages left today here https://www.perplexity.ai/rest/user/settings

[rustic kestrel]

I have an arm 64 snapdragon x elite laptop and I was wondering for battery life is the app or browser better - there hasn’t been a clear answer but some of my friends telling me the app is prolly better as it is native to arm meaning peak efficiency but idk - note I usually have my browser open / use my browser while using Notion so keep that in mind

[tame current]

you could always check task manager (assuming you're on windows) to see how much CPU the app using vs browser

perplexity app on m1 max doesnt work so i had to delete it

[flint scarab]

dear perplexity, i am breaking up with u guys. chatgpt is just better at answering questions. perhaps it is because they lean on illusions more which i can trust for not life ending matters.

[tame current]

who else is using the free nonsigned up version to preserve some privacy?

[flint scarab]

freak

[minor shard]

I use it at work

[tame current]

Why sonnet Is very bad lately.
Short response, useless bullet points, and It didn't even understand what I trying to say.

I can't even attach files Into my threads.

When will all these fixed?
Everything Is getting worse.
After this subscription I will not continue any longer.

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@granite fiber

[kindred folio]

will SearchGPT impact perplexity

i think it comes out 11/5

[heavy smelt]

/imagine olas

[valid iris]

https://www.perplexity.ai/search/find-me-youtube-videos-about-t-woxqpEjqSCyZd7X_MOw68Q#0

3 SOURCES

3

I AM PAYING $20/M and I have been getting 3-6 sources for the past maybe 15 queries.

WHAT IS GOING ON?

I literally don't see how this is better than ChatGPT at this point

And yet for some reason there are 8 videos not listed as sources

https://www.perplexity.ai/search/if-i-am-going-to-boil-chicken-OGwiwwLBQz.XhWgxgmEejg

And here's one with 6 sources, which is still pretty lame considering try at I used to get 25

ummm......

The app?

Not the point, why is perplexity being so lame recently?

[junior plover]

Why is perplexity re routing my prompts to GPT when I have Grok as the model setting?

[gleaming mantle]

I'm still very confused by the search function; does it strictly search titles? I can't seem to match any of the content inside of a thread apart from only the first question, it's practically useless?

[humble summit]

source?

[tropic beacon]

yeah, i though i the only one feels that.
is there a reason for it? is it from perplexity side or is it from antrophic.
hope they clear it out, because it's quite suck to learn anything from short respond that claude 3.5 outputing.

[real star]

yeah perpelexity is not good atm

https://www.perplexity.ai/search/summarise-this-book-but-dont-m-uBC6jPKOR4Gulls7nbhV5A

like i wanted to summarise a book

it always misses out info or just doesnt listen to you at all

[full gyro]

A Pumpkin has appeared!

Time is up!

[valid iris]

Sure, but that feature is due to each ai model, right? Isn't it the fault of the perplexity team that each robot gets so few sources?

[tropic beacon]

hope the dev could figure out what happen in this few days, the service feels like it has been downgrade so much.
maybe it's not them and it just the provider they work with but it should has been cleared by them before hand.

[valid iris]

I'd like if they could just state their reason for this, or else it'll feel like they've been speed running the Silicon Valley grift of making a promising product then ruining it for ads

[junior plover]

Why did perplexity release Grok model and not tell anyone there’s a limit

[cinder comet]

there will be always limits

there is no such thing is infinite or else there will be no profit

[real star]

what model is best

I thought default like combined each model best features

[warped sequoia]

probably sonnet rn.

[warm cave]

GPT4o and Claude 3.5 sonnet are both solid options, GPT4o tends to excel in complex reasoning and STEM related problem, Claude 3.5 sonnet tends to excel in programming and creative writing.

Grok if you want less censored answers

I personally default to Sonnet, it is really an all around amazing model

[real star]

thanks

[valid iris]

No, default is either ChatGPT or perplexity's own model

[warm cave]

Might be GPT4o-mini

Or they use a model router, it’s not very clear

[fossil leaf]

Anyone have any idea why my mac app won't see my pro subscription?

[junior plover]

The shady part is they didn’t tell anyone about the limits and re routed my responses to GPT

At least for O1 they tell you explicitly

[limpid crag]

is grok 2 decent? Heard it was but im not going to pay for twitter premium so i never tested it

[kindred folio]

nvm i got access rn LMAO

[tame current]

is sonnet like... back to normal or not?

by more gpus to make sonnet work better

[icy cloak]

https://openai.com/index/introducing-chatgpt-search/

any one compare gpt web search vs perplexity?

[left stream]

Didn’t even know groq 2 is available for perplexity

Nice

[left stream]

Yes, whoever wasn’t part of the waitlist will have to upgrade or wait

[dim wind]

how is it compared to perplexity

[hollow viper]

Anyone compare gpt search to perplexity

[worn zenith]

All models are limited to 600 a day except for default which is unlimited (and opus, which is 50 a day) unless grok has a special limit like them but that's as far as I know

The re routing seems to be a bug (hopefully)

[halcyon oasis]

Been using Perplexity for about six months. This is my start of second month pay subscription.

Really enjoy the ability to upload images and give prompts

First day on discord

Any tips or tips in these areas that people are using it for that I may not be thinking about ?

Any helpful videos or anything?

[valid yoke]

lol I just got this email from open AI, but when I go to the ChatGPT website, there is no such search button

[kindred folio]

i’m loving it but i’m still testing it to find out if it’s a replacement

[serene sand]

deep dive ➤ #💬│general message

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@warped sequoia
@serene sand

🎃 Halloween Pumpkin Collection Leaderboard 🎃

🥇 @serene sand (mares): 7 🎃

🥈 @minor shard (𝓙𝓪𝔀𝓷𝔃𝓮 🐈): 6 🎃

🥉 @sleek vortex (sneakyf1shy): 5 🎃

4. @last storm (kentavius): 3 🎃

5. @halcyon coral (IceLavaMan): 2 🎃

6. @rich sphinx (WhatIsAModel): 2 🎃

7. @valid iris (blackswanblind): 2 🎃

8. @tired fractal (Feedonyourtearskappa): 2 🎃

9. @livid mantle (GPT-5): 2 🎃

10. @warped sequoia (Nino): 2 🎃

Page 1/4

[warped sequoia]

almost always my first prompt of the day (or after not using it for hours) 'disappears' when i hit enter and its like perplexity is reseting. i have to type it again and then it is ok. does this happens to you?

[serene sand]

https://x.com/aravsrinivas/status/1852082593627590875?s=61

Aravind Srinivas (@AravSrinivas) on X

Been enjoying using the Grok 2 model.
Now on Perplexity iOS app too for Pro users. (Restart app if you don’t see it on “Settings->AI Model”)

Twitter · Oct 31, 2024, 8:17 PM

[jade socket]

when you look at your library/history of prompts does your missing query appear?

[vale crest]

are you a paid user?

[warped sequoia]

yes

[vale crest]

keep it at default it selects based on your query for first response and that is the best I guess
but if you dont like the response just make it rewrite from your desired model

and this is best because it doesnt affect your other threads

[jade socket]

is that a typo

[jade socket]

Happens to me also. Annoying, but now it's part of my 'workflow' to look in library...

[valid yoke]

No, but I had signed up to the waitlist and today I received an email saying that I received access to “ChatGPT search”

[vale crest]

They will roll it out for free users in the coming months
Right now only available for paid users (Plus and Team)

[valid yoke]

Look carefully, it says “searchgpt waitlist users will have access today” — that’s me!

I even received an email today from them saying so

[mighty crest]

It looks like it is it worth using

Tell me how it is when you use it

[warped marsh]

chatgpt search >> pplx imo

why would anyone even use pplx

[tame current]

Research and stuff

Can’t use ChatGPT for research

[warped marsh]

??

what do you mean?

[tame current]

Well, the only way you could truly use ChatGPT for research is if you go to the add-ons and press browser

[warped marsh]

I dont understand. What does perplexity even has to offer now that the official companies dont provide?

and how can you not research with chatgpt?

[tame current]

Well perplexity offers up-to-date research up-to-date articles on your search. It’s really a search thing and research.

[cinder comet]

the question is will other providers provide what perplexity is providing

im talking about chatgpt

[warped marsh]

and what are they providing?

[tame current]

But perplexity is using ChatGPT so I mean it’s just putting an extra search through it

[warped marsh]

and that is now done on chatgpts own website as well

[cinder comet]

will it provide finance analytics with good-looking visuals, sports card system, in-depth reasoning search, multiple models to chose from

you know that if openai did that it will be a direct competition to bing? and they probably dont want that

[tame current]

Thank you for joining the conversation I don’t know much about it because I don’t have perplexity

[warped marsh]

chatgpt has good looking visual and that chart as well for stocks and all

[tame current]

So you’re debating if perplexity is not as better as ChatGPT

[jade socket]

For me, it was cost...I came across some 12 month deal with Perplexity Pro. No brainer at the time, less than 2 months of ChatGPT price.

[warped marsh]

you want a service for sports card system then you are one in a million..

I think this is visual enough for me

[tame current]

That just comes with the service

[cinder comet]

then keep using chatgpt, whats the issue here?

[warped marsh]

I am trying to understand why would one even use perplexity now

[cinder comet]

if it has all of that then just keep using it, its simple as that

each person finds a value in a product he uses

im using pplx mainly because it wraps multiple models

[tame current]

All I’m saying perplexity searches the web better

[cinder comet]

im not using it for the fancy UI/UX or anything

[torn sierra]

Not everyone needs to use multiple models. If we’re talking about the average population, what ChatGPT offers will be enough. If someone feels the need to use other models, they’ll turn to Perplexity. However, Perplexity is light-years behind the competition in many ways – file handling, for example, and image generation (you can’t do it in the app at all). Overall, the UX is poor. The visualization in ChatGPT is pretty good in terms of search capabilities.

[cinder comet]

im just telling you it provides more than the chatgpt search

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@native latch

[cinder comet]

the in-depth reasoning is a big plus for me

hopefully you can show me a screenshot for your chatgpt on that too

[warped marsh]

what does in-depth reasoning mean?

[cinder comet]

sheesh

[warped marsh]

:/

havent hard that in a long time

[cinder comet]

how do you expect me to take you seriously when you dont even know the features of the services you are trying to compare

[tame current]

I’m gonna give you a word for work for what they say

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@torn sierra
@jade socket

[tame current]

Just o1 min use for complicated questions

[warped marsh]

I never asked you to entertain me. You can leave the conversation if you want. in-depth reasoning is such a vague term

[cinder comet]

no its not

[tame current]

What

[cinder comet]

its triggered when there you want to map out many outputs

it doesnt use o1 mini for that

thats a seperate reasoning section

[warped marsh]

are you talking about that agent search thingy?

[cinder comet]

@tame current care if you join, share with us your unique point of view

i guess

[tame current]

Have you ever heard of? You.com

[warped marsh]

yeahh

[tame current]

?

[tame current]

Yeah, I thought about that

[warped marsh]

I dont use it though

I have never used it

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@native latch
@warped marsh

[tame current]

You should look into it. It’s a good service.

[cinder comet]

now @warped marsh question is probably why would i pay for a service that gives me the same value as chatgpt

well its because the added value for you is different

[warped marsh]

my question is.. what apparently triggered Asura is why would anyone use these search services when the models they are using has built in search functionalities in their website itself

[cinder comet]

well its not a secret that it wraps/embed around their api

i told you why im paying for pplx

[warped marsh]

for multiple models.. that makes sense

[cinder comet]

exactly

[warped marsh]

how much usage do you get for 3.5 sonnet?

the new one

with pro

[cinder comet]

you wont find a 600 messages for sonnet anywhere

[tame current]

What I believe perplexity has different models to search with if you go to individual models and get individual services it’ll be a waste of money

[cinder comet]

well i use it like 200 times a day or so

try to do that on anthropic website and you get insta rate limited

[warped marsh]

when I was using perplexity, the answers quality used to be so inconsistent that it was almost obvious that they are changing models in the background without informing you to save costs and that has happened so many times with me

[cinder comet]

you want my honest opinion

the search thingy still lacks

and hallucinates alot

[tame current]

@cinder comet if openai were to incorporate all the features youve mentioned, which i hope the do.. its like microsoft wins anyway even though bing has a the same reputation as it did when it first came out. pplx will have formidable direct competition... like coke vs pepsi, toyota vs ford, samsung vs apple, etc.

[cinder comet]

i use the writing mode for like 90% of the time

[warped marsh]

yeah no one uses perplexity for their search feature even I just used them as a middleware to use other models

[warped marsh]

but then this happened and then I just stopped using pplx pro

[cinder comet]

well they will probably implement some progressively

[warped marsh]

I dont know if this has happened with you or not but I have compared chatgpt 4o or even sonnet with the same prompt on pplx with ai profile off and on their respective website like 10 times to avoid any inaccuracy and pplx answers sometimes have been so much worse

[tame current]

after the claude sonnet fallout.... ugh. pplx not performing the way it used to.

i still prefer pplx over google search (a shell of its former self)

[warped marsh]

pplx earlied used to be good with their pro version but now I am so sure they change the model in the background without telling you.. and the worst part is there is no way for the user to really know

[tame current]

i agree. black box.

[warped marsh]

you complain and they just say well its an LLM, they give different answers even to the same prompt

[tame current]

easiest way to escape accountability

[warped marsh]

exactly my point

[tame current]

and the easiest way to make them care is to vote with your wallet

[warped marsh]

to which now you cant really argue

[tame current]

correct

[warped marsh]

that is why I stopped using pplx because they lost all my trust with this thing they do

[cinder comet]

for sonnet i havent seen any difference tbh... maybe it has to do with their internal pre-prompting

[tame current]

when regulatory bodies around the world hold accountable ai companies... that will be the day blackbox responses will be reduced

[warped marsh]

I remember using this prompt

If I have 4 apples today. I ate 3 apples yesterday. How many apples do I have today?

gpt 40 and sonnect always gave answer 4 like ALWAYSSSS

[full gyro]

A Pumpkin has appeared!

Time is up!

[warped marsh]

but pplx was soo inconsistent with their answer irrespective of if I am using sonnet or 40. It was so obvious of what they do in the background to save money

[torn sierra]

For me, Perplexity is mainly about Sonnet and its writing focus. However, it annoys me that Perplexity has struggled for a long time with things like properly handling files. Recently, I’ve been spending more and more time in ChatGPT. The advanced voice mode is amazing, and it’s something Perplexity lacks. It’s brilliant for learning foreign languages. Plus, file handling works well – it doesn’t lose context. The canvas feature for coding is also pretty good now. And o1-preview really helps when I need to make larger edits. ChatGPT also does a great job with charts when looking up financial data. I’ll see if I end up missing Sonnet 🙂

[cinder comet]

im not really sure but ive noticed the tokens are limited by default unless you ask it to like answer in 2000 words or so

by default its a short answer

but it could be just my imagination

[tame current]

and to your point on handling files, pplx spaces.... i load files in there, tell the model its loaded... and it responds... "what files?"

[warped marsh]

yeah they obv use less token window models.. I am not talking about that

I am talking about inaccurate results

when that model is smart enough to get it right every single time

I used that prompt with bing as well and it also used to give me wrong answers and I stopped using it because microsoft being microsoft lied that they use gpt 4o in bing

It is so easy to know when companies like microsoft or pplx say they use xyz model but then switch models internally and give dumb response and then blame it all on "LLM Behaviour"

[torn sierra]

Perplexity does something but it’s simply not done professionally. It feels like it’s half-baked and just released. Issues linger for a long time or aren’t fixed at all, being treated as “non-issues” as if they’re supposed to be that way 🥲

[warped marsh]

the best place to use these models is from their respective company's portal. Even though they can do the same thing, I have never in my life experienced inconsistent answers from either chatgpt.com or claude.ai

If a model on their website gets an answer to a question correct today, it will get it correct tomorrow as well

[tame current]

perhaps it seems pplx is in the "iterate and break things" cycle. understandable. competition is fierce right now.

[warped marsh]

I would have used perplexity and even recommend to other friends of mine but pplx just seems such a scam company to me because of what they do internally

I have stopped using pplx pro for so long and I dont even use other third party services like you.com or phind.com not because I have had bad experiences with them too but because now I find its just better to use the raw models from chatgpt.com and claude.ai

[tame current]

well, time will tell.

accountability will come in the long run but right now, regulation is too early for all the right reasons as the US wants to be number 1 and so legislators are careful to not hamper competition... like what happened to drones. china far ahead of the US.

[warped marsh]

for search I still use google and prolly will still use google but for raw models I just use their own websites bro

[full gyro]

A Pumpkin has appeared!

Time is up!

[warped marsh]

when there is a way for me to know if they, for all 600 prompts, use gpt 4o or sonnet and burn that much money on me

I will get pro again without even a thought

[sleek topaz]

What happened? Everything is working beautifully for me in regards to Claude Sonnet 3.5. Would you like my AI profile? Maybe thats the issue

[sleek topaz]

If you use their default model maybe, but not with things like claude sonnet 3.5

[tame current]

anthropic did a claude sonnet 3.5 "refresh" like it was spring time cleaning or something and then people started complaining including myself

[warped marsh]

Why would I use their default models with pro 🤣

[sleek topaz]

When did they do this? Any info regarding it?

[tame current]

they did it last week

[warm cave]

Some people like the speed, but I don’t know if default is actually faster now days

[tame current]

there's what anthropic puts out... and then there's what users are saying

[warm cave]

Back when it was GPT4 and not GPT4T and GPT4o it was nice using default because it did not take forever to get an answer

[tame current]

the changes are enough for people to stop paying pplx from what people in chat say

[sleek topaz]

Are you talking about the new claude sonnet 3.5 version they pushed (the one thats supposed to be far better)? If so, it is far better and working brilliantly for me. If you could give me an example of some prompts that didn't perform well, I can try with my settings and see what happens

[warm cave]

I have been using it in cline for code, really great

It outperforms o1 I’m pretty sure on code

[tame current]

coding tasks

[tame current]

sonnet was leaps and bounds more accurate.. and then.. the great "refresh" of death

[warm cave]

Yeah, what do you use it for, creative?

[tame current]

and destruction

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@warm cave
@serene sand

[sleek topaz]

Please give me an example so I can test myself to see what you mean :)

[tame current]

no

code is private

"/ban perplexity ai events"

[sleek topaz]

You dont have ANY non private example you could share where you felt sonnet performed "bad"? Something creative, something factual (like news related), etc.

[warm cave]

I tend to trust the aider benchmark the most

Most of the time when a new model comes out I just need to learn how to get the most out of it

With that said it could definitely be worse in specific things

Just not my experience

[sleek topaz]

I've never had any issues with the new sonnet model. But if people really are experiencing issues then it would be best if perplexity can give you the option to toggle between the old and new version

I would just like to see an example of the "refresh" causing issues, as I've yet to have any

And I use the model mainly for code

[tight matrix]

I got perplexity pro and I tried using the different AI models but I didn't really get any noticeable differences in them. Can someone explain what the different models are good for?

I've been using Perplexity to learn complex topics like politics, international relations, etc

[warm cave]

GPT4o and Sonnet 3.5 are currently the best 2 models provided by perplexity, both excel at reasoning, GPT4o has a slight edge when doing stem related problems, and sonnet tends to have the edge for programming and creative writing. Outside of that there is grok, grokis supposed to be less censored so it might be useful on some topics

[chilly spindle]

it'll be hard to outclass searchgpts speed

[warm cave]

Easy when you are in bed with bing

[mighty crest]

ChatGPT got search
This is a whole new era for ChatGPT

Still not better than perplexity though

[warped sequoia]

the ability to switch between models is also crucial for perplexity

[mighty crest]

Yeah,

[tight matrix]

I see. Thanks

[full gyro]

A Pumpkin has appeared!

Time is up!

Collectors:
@warped sequoia

[ashen dagger]

Hello,
CAn I know why perplexity just cancelled my pro sub? 🤡

[candid bear]

有推荐最新电影吗？

[vale crest]

inbox me I will show you why

[sleek topaz]

What else would you use

[plain zenith]

Hello. I used to be a chatGPT user, Claude, switched to Llama, why do you use perplexity and not let's say directly some chat bot?

[serene sand]

deep dive ➤ #💬│general message

[plain zenith]

thank you so much

[hybrid quiver]

Cant wait for perplexity on both windows and mozilla

[serene sand]

#1225572798966267966 message

[hybrid quiver]

I mean for extension

[hollow wave]

hi awesome ai

[bronze bridge]

Hey, now that GPT Search is here, how are perplexity going to justify their pro price?

[warped sequoia]

i guess by having the opp. to choose and switch models. and also recently, i saw many awesome updates

[bronze bridge]

I pay for pro and GPT, I love spaces and pages but I can See GPT bringing out more competing products soo, feels like just a matter of time, anything else in the works to keep the value prop strong for perplexity?

obviously, for now, perplexity search feels better and looks nicer but GPT search is real time, and perp basically uses GPT for logic and reasoning so I can't see the value in pages/spaces being worth the cost for me atleast, i get everyone has their own value/cost threshold though

[blissful pendant]

Access to multiple models is the biggest difference

[untold sequoia]

Hey, does anyone have experience using perplexity API to pull real-time statistics? For example, how well does perplexity perform at something like: "Get me the current election odds based on accurate real-time statistical data"
How prone to hallucinations and inaccuracies would you guess this would be? And does perplexity AI have structured outputs? Thank you!

[plucky plume]

Yup, same for me

[worn zenith]

...which is now not available due to the bug lol

[plucky plume]

It would be noice if perplexity get artifacts like Claude ai in future

[tame current]

Can anybody tell me how the grok 2 is doing is it a good model to use?

[versed plaza]

Yeah like what's what's its thing? I thought the point was that it like generates edgy content

It is helpful to have for reference though because I'm never going to pay for a Twitter subscription and it's nice to know like what sort of algorithm is shaping that community

Anybody want to help me test a prompt? I was going to share it in the prompt channel but it's rather complex so it's not going to fit it all

I've been trying to push myself with advanced prompting

[vapid grail]

is there a way to generate pictures in the app?

[wraith crow]

Generating images is currently only available on the web

[little surge]

Hello! I downloaded the app in my MacBook but when I want to log in with my google account it just keep loading and shows and error! Already check that is in the last version my MacBook and erase and install again the app but keep showing the error! What can I do?

[serene sand]

#1298895118513737828 message

https://x.com/apostraphi/status/1852436101325074626?s=61

Phi Hoang (@apostraphi) on X

every second counts.

Twitter · Nov 1, 2024, 7:42 PM

[little surge]

Thank you very much!

[mighty crest]

Question in your opinion, what do you use for perplexity for? I see you’ve been around here a long time.

[shut light]

probably used a leaked code

[queen pike]

Is there an API for perplxity?

[shut light]

google is usually good for questions

[balmy dagger]

@south kindle I have to give a big eloge to you, read most of your comments and posts and I’m amazed of how well you have been both sharing super effective out of the box prompts, but also leading users to learn prompting them self (with a bit of ignorance “with fingertip feeling”). I think it was you that sent a prompt that I use for almost all my research and it is exceptional. So big up to you and keep up the good work! Would like to learn more from you!!😊

[balmy dagger]

I am wondering the same, been trying it since yesterday and seems fine, but have not tried anything complex yet and don’t know it strengths either so testing around to see what it generate, so far it has been good but since I try in such low level it’s had to tell if it’s better or less then sonnet

[serene sand]

deep dive ➤ #💬│general message

[fast meteor]

https://www.perplexity.ai/search/is-there-an-api-for-perplxity-yMilNCIeQbCffJXewq3Xow

[wild bridge]

Good morning!

I have questions.

Is there any benefit to being a curator?

Having a lot of fun using perplexity, definitely feel myself about to fall down the rabbit hole hard.

[fast meteor]

Global exposure on the Discover feed
Pages selected by our editors will be shared with our worldwide audience.

Credibility
A verified badge will be displayed next to your name on all your published Pages.

Rewards
Curators will get free Perplexity Pro, and top participants can earn exclusive Perplexity merch

Shape new initiatives
Be one of our first 25 curators and help shape the Discover Curators Program.
https://perplexity.ai/curators#learn-more

[wild bridge]

Thanks, I must have glossed that earlier on when I looked at it.

[gray marten]

Could anyone help me?

I'm trying to subscribe to the PRO version on my cell phone, but it says "Your transaction was declined. [OR-FGEMF-20]", have I tried adding other cards but the same error appears?

Has anyone had this same problem?

??

[nimble isle]

Hello everyone. Can anyone please explain why perplexity can't function with a VPN running? Is this just me? Or maybe a new feature for the free subscription users, goes away with pro level? Please and thanks for any opinions 😃

[plain zenith]

I have perplexity pro working with proton vpn

[plain zenith]

It works without any problems

[drowsy valley]

App working on iphone with NordVPN on (I have Pro)

[nimble isle]

🙏

Hi. And thankyou.. using Express VPN on both android fone and windows laptop, and now free perplex says no go both places 🤔

[indigo valve]

Is there a reason why I have Sonar XL selected but it still says default?

[valid iris]

Is the free pro subscription that i just won valid for one month or one year?

[wide stratus]

🤔

[livid mantle]

I didn't even know I joined an event but thankyou so much @signal hamlet I am very greatful! I continue to worship Perplexity.ai.

[tame current]

Have you received the code yet? I haven’t.

And @signal hamlet thank you so much for this

I had the direct messages off and I have it on right now

[warped sequoia]

You join by picking up 🎃 here.

I am also waiting for the code.

[tame current]

Good I’m just making sure

[fallen valve]

Hey guys, I'm a big fan of Perplexity and just did a "Masterclass" tutorial aimed at helping people level up from beginner to advanced from this one video. I would love if you checked it out and to hear any feedback you may have!
https://youtu.be/b5IVppwaOAE

[wild bridge]

Can you set two urls to be equal to one another using a prompt?

Not exactly helpful, but picking at strategies to potentially link agents across spaces.

I really need to start digging into apis.

[warm cave]

Just keep your user name the same for a min so when the team dm’s you they are not confused, haha

[tame current]

I’m just hoping the code comes soon because I’m so excited

[warm cave]

I’m sure it won’t take too long, they have done these giveaways in the past

[zealous kayak]

I love how people keep running into the same bugs (All models become 4o, files get ignored, etc) but it's just deafening silence on the bug reports lol

what's wild is that some of these (especially the 4o one) are bugs that were "fixed" already

[queen pike]

That is cute.
Thanks.

[queen pike]

Thank you.

[haughty juniper]

Help me,
I need to know what's going on, all the models are responding like 4o, ALL OF THEM. It's very frustrating.

[warm cave]

So you have it set in setting to something like sonnet and then you get responses in GPT4o? I have not seen this happen, but maybe that because I set my model using an extension and not in the settings page, if you link it here maybe other with the problem can upvote it

Or is it silent, where it still says Claude but the response sounds like GPT4o?

[zealous kayak]

yeah

well, not me personally, but I've seen lots of reports

https://discord.com/channels/1047197230748151888/1299441439620464640

it's most obvious because you can just ask it what model it is

[warm cave]

Hmmm, can’t seem to replicate

[zealous kayak]

it's random

some are affected, some aren't

[warm cave]

That sucks, haha. I know they are saying it’s the system prompt in the thread, and that’s true, you can’t tell the specific model being used, but it is decently reliable at telling who made the model

The only exception is when they tell the model to act like a model it’s not, then it might not say who made it, or say something false, but I don’t think perplexity is doing that, unless you could them telling it that’s it’s perplexity ai

[zealous kayak]

Nah, I think it's clearly some kind of routing issue on their end

they're just making people fall back to 4o for some reason

it already happened once before

[warm cave]

Yeah I was agreeing, I think they misdiagnosed it in the bug report thread

Luckily I am unaffected, haha

[zealous kayak]

Ah, then yeah, I also think they misdiagnosed it

but it's been radio silence so

I don't know, perplexity's just been... pretty terrible lol

[warm cave]

Wondering it is shows GPT here when you do it? https://www.perplexity.ai/rest/user/settings

This is the line you are looking at to see

It might be all on one line or look ugly depending on the browser you use

If it is screwed up here than that is an easy way to prove the problem, if not it might be server-side

Oh but I’m your images it shows the model name, so I assume it matches in the api…

[azure ingot]

Today I used Claude intensively, maybe even 200 times in a few hours... until I finally stopped because it started redirecting to GPT-4o 😭 The quality of the response dropped dramatically 😭 I had to stop working and exploring

[tranquil orchid]

Hey @azure ingot!

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.

[haughty juniper]

I understand you, 4o is disgusting, bad answers

[warm cave]

Hmm, yeah it seems to be requesting the right model, seems the problem would be server-side

[haughty juniper]

That's too bad, I hope they fix the error soon. I can't do anything.

[warm cave]

@signal hamlet Hey Alex, I see in the bug report it was diagnosed as LLM system prompts and hallucinations, however it dose look like there is a problem imo, I would appreciate it if you guys took a closer look into it

We did some testing above and Alex2001’s results are very different than mine. So I think it’s worth looking into, seems a little too consistent to be a random thing. Now days Claude models will almost always say they are trained by Anthropic, and OpenAI models do the same. If it was a GPT3.5 vs GPT4 type of thing then I would agree it’s just the model being stupid.

[haughty juniper]

I can't talk to Claude, Grok, Sonar, only 4o no matter what I do, whether it's in a space or a normal conversation

[warm cave]

Especially since grok is super easy to pick out, it is very unique in how it responds, haha

[haughty juniper]

And I haven't even tried it 😭

On my cell phone it also answers 4o, just like on another PC

[finite bay]

I liked the product, that's why I paid for it.
Pls tell me you are not seriously considering placing adverts on queries for Pro users! 😡
This will trigger a massive backlash, and myself and many others will absolutely cancel our service and protest very loudly online..

[harsh stag]

there will be adblockers tailored specifically for perplexity if this ever happened 😉

unless they push ads widgets inside the generated answers like they've already did with shopping, then Im unsure whether adblockers can do anything about it

wow, im impressed with how consistently perplexity shopping widgets appear in my tech products-related queries

not the end of the world, when im looking for a product, i should be expecting ads, shouldnt i

[warm cave]

I don’t think we will be getting it for pro users, I could be wrong, but it would be a surprising move imo

It did happen to Amazon tho lol, that site is nothing but ads

[cloud marsh]

maybe they could add a shop focus and add ads there

if they put ads in other focus ill switch to felo

[haughty juniper]

Ad blocker and you're done

[warm cave]

Would not be surprised if they are already partnered with “Shop”

Maybe Shop will give them a cut

[livid mantle]

you just got Perplexitied

[livid mantle]

alex sleeps 10 hours

he is following what perplexity told him

to sleep 10 hours

[restive osprey]

im starting to like Perplexity for comparing things. been comparing computer parts and reviews

and pulling news reports from different sources

wish it was just a little bit more smart.

or a lot. never can have enough .

Perplexity has a real chance to overlate Google eventually

with a lot of work.

maybe with some key acquisitions of other research and dev in near fields and cutting edge.

anyway.

[spark crow]

Hey crew - I wanted to share a new video and blog post about how LLMs assist with auto-patching vulnerabilities.

https://www.dylandavis.net/2024/11/self-healing-code/

I hope you find it useful. Feedback is always welcome!

[lapis arch]

.

[frosty pulsar]

Hi all I'm trying to build an app using Python with realtime voice analysis and detection, I just purchased Perplexity Pro. Which is the best AI model for this to have it full write code for me? Tried Claude and it constantly hallucinates, gives bad code or just ends up asking me if I want to proceed without doing anything

[lilac rapids]

Yikes, what a terrible response it gave you.

I don't see how that would ever be beneficial... hey, I know you're trying to solve this problem, but would you rather just do nothing and ignore it than fix it? That's easier for me! 😅

[median summit]

I use perplexity mostly for coding, but I don't face these kind of problems. But I provide detailed prompts with clear questions or tasks to it. Try to provide examples of your current code and provide more details about what you want.

But I really wish, Perplexity had a feature like OpenAI Canvas, that feature is actually pretty good and mitigates the number of bad answers in my pov.

[shut brook]

So, apparently threads have a maximum number of messages and Perplexity just drops old messages?

[lunar egret]

Did you get this resolved? My threads are also suddenly gone

[thorny saddle]

I guess the query router miss out sometimes

[thorny saddle]

tell me in more details. which models, web or api, prompt format guide, etc. Thanks Daniel

[frosty pulsar]

Provided a pretty advanced prompt with every guideline and instruction to follow, even asked Claude itself to assist me with the prompt!

[frosty pulsar]

It only seems to happen with Claude with its latest Sonnet model. Its like a polite servant that asks if you want tea, brings it over but refuses to hand it over. Annoying to say the least

Also when I ask Perplexity to fix full code it keeps truncating repeatedly, very annoying

[summer yew]

Anyone else having context window issues? I uploaded a file (25 k tokens) as a pro user and 3.5 Sonnet/Grok 2 can't process/read it completely as it did before.. -.-

[frosty pulsar]

What's funny is that usually the only way to fix it is to swear at it, seems to help lol

I'd also like to know how you remove Sources because if I have to copy and paste full text to be fixed, Perplexity uses that as main source and keeps referring back to it, so I have to start an entirely new thread with the fixed code to continue

[shut brook]

Into a Space? If you've figured out how to actually get perplexity to use files in spaces, please let us know your secret

[summer yew]

Anywhere. Context window seems reduced by 66 %. Will wait 2 hours until I contact the support

[frosty pulsar]

I don't find spaces at all useful - gives constant errors

[shut brook]

It just doesn't do what it's supposed to - at all

[warm cave]

Perplexity, wen AI minecraft?

[tame current]

I feel like I’ve just had a stroke looking at that

(whispering) I still haven’t had my code

[hollow dove]

Now that SearchGPT has been out for a few days, I’m curious if anyone has done a side-by-side comparison of ChatGPT with search and Perplexity. I’m trying to figure out which one is worth my $20/month.

I paid for both for this month and have been trying to compare. So far, Perplexity still has an edge for some things but CharGpt is still more versatile (it will create maps and charts more easily). I work in research so accurate, real-time info is very important.

[frosty pulsar]

I think I'm going to refund Perplexity

Sorry but this is a complete waste of time

[frosty vessel]

How so?

[warm cave]

It is quite a fever dream

[frosty pulsar]

Constantly having to tell it to do basic instructions, constant code errors for something that isn't even that complex, claude 3.5 sonnet implementation is effectively useless. It has potential but useless for coding and complex tasks

Its not fun having to copy and paste code 5 times with the error codes/warnings, asking it to fix in full code for it to then spit out 50 lines of code with removed features and broken UI then repeat, repeat, repeat

I swear I'll be hearing fix in full code in my sleep. None of the other AI sites do this, just perplexity and I've paid for all of them

Spaces is bloody useless and you have to start a new thread seperate from it for it to stop remembering your previou error codes and puttiing them back in only to break your code even further

Its fine as a search engine but hopeless at anything else

[torn sierra]

It really depends on the question. Different services have their own strengths. ChatGPT with its search function performs better in desktop-web than what it can offer in the app. Initially, I was enthusiastic about ChatGPT, but Perplexity's responses (especially using Sonar Huge) suit me better. If needed, I can also choose Sonnet for coding or creative work 😉 You need to evaluate what matters more to you. Perplexity has many drawbacks - working with files is poorly organized, and it can't generate images as easily as ChatGPT does. However, Perplexity offers newer image generation models. ChatGPT will surely update soon though 😉

[frosty pulsar]

No idea why it comes at a cost when the free search is unlimited. Pro isn't worth the Pro price

I get it nothing is perfect, its AI. But christ I've never had so many problems with rather basic 2024 functions

[tawdry chasm]

well I pay 4.99 per month so no complaints

[tame current]

Where are you getting an AI service for five dollars a month

[tawdry chasm]

if u have a US .edu email u get 4.99

[frosty pulsar]

Ask Chatgpt to do something? It'll do it and explain what it has done, offer proper suggestions and accommodate to your needs. Been like that since pretty much day one using chatgpt

[tame current]

DAMN
(I might have to buy one off eBay)

[frosty pulsar]

And that's on the free version, I now pay for chatgpt and use it daily on my phone for tasks, research etc and as part of my job. Had hoped perplexity would uprgade me overall but so far, no.

I'd post this in feedback-general but I'm not one anyone listens to so who cares what I have to say

[elder dove]

Hello?

[frosty pulsar]

But just on my view on it so far, perplexity isn't the worst but it needs a lot more to justiify asking for money

Because its essentially just Bing 2.0

[tawdry chasm]

I like Grok 2 a lot

[rocky crow]

https://photos.app.goo.gl/DhrnzxcuM2oWi7wL9

[mighty crest]

What is it supposed to mean?

[hollow dove]

Where can I sign up for this!?

[rocky crow]

You're laughing aren't you 👀😉

[tawdry chasm]

u should automatically get the account if u register with a .edu email

[hollow dove]

Yup thanks! It worked. I’ll have to create a new account but totally worth it. Now I can keep both perplexity and ChatGPT!

[calm compass]

What’s better for complex tasks that involve deep research: grok 2 or Claude 3.5 October version?

[shut brook]

don't buy edu addresses off ebay - they're unlikely to last and all these services force you to login via emailed link

[weak hare]

Hi , actually i have lost my mail access so i cannot able to logiin in my mail as of now my current perplixity session is active but later it will be logout so can you please help me out with this issue of changing mail or any other way

i have also raised the concern on the help section through the website and gave an alternative email id to conatact me

any mod online because before my page expired i need you to kindly check this

can any one please check

[sleek vortex]

ask via email

not sure how you'll prove your account ownership but

[tame current]

I was thinking the same thing

[sleek vortex]

possibly via ip address? but that could also be a form of attack vector for perplexity, interms of "security"

[tame current]

How would they change the email if you don’t know what emails on the account

[sleek vortex]

i mean he didnt say he doesnt know what email it was on

but that he lost access to it?

unless he also doesnt know

it would be much harder if he doesn't know at all...

if you have pro, or even did a free trial at some point, it would be doable to prove via address/card statement or something

[tame current]

Well, you’re right. I hope he knows it, but he doesn’t have access to it so maybe they could change it.

But if he forgot it, he’s gonna have a hard time

[sleek vortex]

yeah...

well can you have much to lose on a free account? it would just be your thread history

not saying that can't be important but, it shouldn't be mission critical...

[tame current]

I didn’t read that a free account (if it was me, I would just log into another email)

All this time, I was thinking he was paying for it

[sleek vortex]

nevermind

he does have the Pro role (didn't check!)

so it is a paid account - justified to want it back

proving by address on the payment method used may be a possibility for you then (?)

well youll need to go their support, not much we can do theorising...

[tame current]

I just noticed I never gave him the email to support [e-mail]

[weak hare]

under helpsection i just messaged ok i will message in this from my different email what are the deails you required from my end

[livid mantle]

I don't pay, I win! I came, I saw. Thank you Alex.

Source

Thank you @signal hamlet

[calm compass]

What’s better for complex tasks that involve deep research: grok 2 or Claude 3.5 October version?

[tired fractal]

how the hell did I win when I just randomly clicked twice while chatting here 5 min a week? 😄

whats the code do?

[sleek vortex]

one month of pro free

[sleek vortex]

that’s how excitingly active this server is 😊😊

[shut brook]

mods almost certainly do not have any ability to access the company's systems, accounts, etc.

[tame current]

I just wanna say thank you @signal hamlet for everything. I appreciate it.

[tame current]

I wanted to ask you I still have to put a credit card in for the code could I cancel it before the month so they don’t have to charge me for the other $20

[livid mantle]

yes

[tame current]

Ok thank you

[frosty pulsar]

Can we get a fix for this?

[lapis spindle]

@signal hamlet i want perplexity pro too 😅

[last storm]

no way i won the pumpking competition LMAOOO

[tame current]

I was surprised myself when I want it

[twilit otter]

Hi everyone, I was using the iOS app, and I noticed that there’s no way to select which LLM models the answer will be generated based on. The only option available is whether it’s a pro search or not. Is this correct?

And for the “rewrite” function, I can choose the LLM models there, but what is that “pro search” option indeed? Like, what LLM models it is based on?

[zealous kayak]

there is something wrong with it, but the problem is that people are mistaking normal behavior for the issue

there is a problem where the underlying model is set to 4o even if you choose 3.5 sonnet, for certain accounts

but people just see "Oh it says perplexity AI therefore I have the bug" which is wrong

you need to actually get it to state if it's made by OAI or Anthropic

asking it specifically what model it is usually works - as you can see, I obviously don't have the bug

[whole flame]

is perplexity better than chatgpt

or like whats the difference

[lilac stump]

perplexity has chatgpt

basically chatgpt is an AI

claude is another

sonar is a 3rd

perplexity lets you use any of those as you wish

under one single site

[summer crescent]

Should be stated Perplexity is effectively 2 seperate AI.
There is the under the hood Perplexity AI, ( which I believe was confirmed to be built on anthropic's AI but don't hang me for that one ) this one does the job of figuring out your question and searching (currating data)

and then it presents the data to the model of your choice, for it to deliver to you, if you just need it to spit that data out verbatum, there is very little point in selecting specialized models, however, if you need the AI to then take the found information and "process further" then the secondary AI can have an effect, but in 80% of cases you will find that the "mouth-piece" doesn't really affect the outcome all that much ❤️

[torn sierra]

ChatGPT has several models, including o1-preview, which is not available here. ChatGPT has an advanced voice mode, which is not present here. It's a very cool feature and helps with language learning. Even the regular Whisper in ChatGPT is, in my opinion, better than the voice function here. ChatGPT doesn't have issues with file handling, but there are problems here. You upload one file, and it stays in the sources, causing context issues. ChatGPT generates images immediately, while here you can only do it in the browser version, and not in a simple way. ChatGPT has a well-functioning code interpreter. It's here too, but its functionality is limited - for example, you generate one chart and try to change its color 😉 ChatGPT also has a canvas function with the gpt-4o model - quite nice for coding and writing. The search function in ChatGPT isn't bad either. So the choice isn't that simple. It depends on your needs. If you want to use various models, including Claude 3.5 Sonnet, and you want to have quite good search capabilities (although errors in Perplexity also occur), then Perplexity will be better.

I have both 😅 ChatGPT and Perplexity

[summer crescent]

I feel they offer two very different functions...
well did until ChatGPT released search, I still prefer how perplexity presents search and how they provide sources.

but I do believe you will quickly run out of Queries if you use ChatGPT for regular searching ( not sure if they changed something )

where with perplexity you can atleast search for free turning advanced mode off if you have smaller questions.

but yeah I personally use Perplexity with Gemini Ai, I don't use the voice functions, but I usually let perplexity gather all the data, I then feed into gemini for further "discussion" analysis ( due to gemini not being limited in queries )

but I used to have GPT + Perplexity.
I swapped to Gemini due to Gems allowing me to customize my gemini behaviour and and have it split longer messages off into parts ( where GPT have a tendency to just crop / truncate the responses )
I also finally got gemini to be better at stating if Inferrence or fact, which helps a lot with hallucinations where GPT still confidently answers everything like it's fact.
but that's just smaller things

[twilit otter]

Hi, may I ask what is this pro search based on?
And how can I choose the LLM model on the iOS app?

[lone cliff]

I have a general question: I’m a bit disappointed. With the subscription, why are we only getting free versions available online? Can’t you provide paid LLMs like Sider?

[robust barn]

Hey all, is it possible to get personal information refreshed?, any earlier information and accounts Etc.
Best Regards.

I’ve deleted some Old accounts and it still appears when i search my name

[ancient dew]

Haw

[hollow dove]

Had this very odd hallucination yesterday….

[tame current]

They must be forgetting, just like Joe Biden

[gray marten]

Could anyone help me?

I'm trying to subscribe to the PRO version on my cell phone, but it says "Your transaction was declined. [OR-FGEMF-20]", have I tried adding other cards but the same error appears?
Has anyone had this same problem?

I've even tried it on another cell phone, but the same error appears when I try to subscribe to the PRO version through the Playstore.

[warm cave]

Pro search is not its own model, it just more in depth search, it will use the same model you were already using, so if you were chatting with sonnet then pressed re-write with copilot, the. It would do a in depth search and then sonnet would respond again

[warm cave]

The paid models that are available are Sonnet 3.5, GPT4o, o1 mini, Grok 2, Opus 3.5 and Llama 70b.

You can change the mode in the settings page or if you use the “Complexity” chrome extension, then you can change it from anywhere

[twilit otter]

o1 mini not available on iOS tho?

[warm cave]

He was talking about online, I was assuming that meant the website. But you can always open it on the mobile browser, even tho that is less optimal

[twilit otter]

So what models would you recommend when doing a daily news search?

[warm cave]

For news search most models are plenty good, AI models have been really good at summarizing for a while now.

I personally have sonnet as default, I tend to like Claude models, grok is great when the other models are refusing for stupid reasons since grok is uncensored

Llama I like as well

[balmy dagger]

I am liking Grok-2 to be honest, but unsure if i am using it correctly. 🤷‍♂️

[warm cave]

What do you think you are doing wrongly?

[balmy dagger]

The prompting, i am not sure if i should make the ”introduction” about me or about the ai

[warm cave]

In the profile or in your prompt?

If you are talking about prompt then: You should be fine both ways, it was trained to be a conversational AI, so for the most part you just talk to it like it’s a person.

[balmy dagger]

Both kind of haha, I have enterprise pro, so I am unclear to where I should prompt the AI, since we have corporate instructions, profile instructions and now spaces instructions. I’m getting bit confused, things goes well but I think I am not using the full potential of perplexity

How I do at this moment is that I have set my career in my profile then in the corporate instructions I set it for the asign business then I manually put one of @south kindle prompts in my message to get the answers I want

Feels like I am going around to much and maybe there is a easier way

My English is not the best but hope u understood

[twilit otter]

Is Sonar Huge good??

[serene sand]

#1199176325051723837 message

[warm cave]

That makes sense, that is a lot of system prompts being passed in, maybe I am misunderstanding the question but here is my understanding, the corporate instructions, AI profile and spaces prompt should all should be prompts telling the AI how to behave, so a system prompt saying “always use markdown tables when providing data”. And the message bar is reserved for the task you are giving the AI, like a question or request. But where things are tricky is when you have soo many systems prompts the model might get confused and not follow all of them correctly. So if it’s not following the prompt in the spaces prompt ( I.e. scratchpad) then you might have to remind it in the question you ask or put it in that message. Hopefully this is not too confusing way to explain it

[balmy dagger]

This one was crazy, got to try it

[warm cave]

Yeah I like Sonar Huge (Llama), it’s quite good for search imo

[serene sand]

deep dive ➤ #💬│general message

[balmy dagger]

Yeah you understood me correctly and that is where paradroid prompt comes in, I have it in my notepad on iPhone and in a discord server so I can copy paste it when using perplexity

[balmy dagger]

Tyy

[warm cave]

Yeah, when it comes to his prompts, they should work as both a spaces prompt or directly in the message bar, you can try both ways to see if you get better results

[balmy dagger]

I played around with liquid lfm and asked it about my situation and got more confused even thou it gave me some nice ideas for me to try 😂

[balmy dagger]

Question 🙋‍♂️ should I replace my profile with this prompt? Or should I put it in a space instruction?

I’m feeling like a newbie when I write with all of you, but I am getting good use of AI and prompting otherwise haha

[proper oracle]

It doesn't seem to be very good at asking for "today's" news or similar. There's an example in the app for getting today's technology news, but when I tried it, some stories were recent, but some of the stories were from 4-6 months ago

[serene sand]

first prompt pic 1, second pic 2

[serene sand]

take your time, go through deep dive I gave you and you’ll become power user in no time

[balmy dagger]

You are the best!! ⭐️⭐️⭐️

[cinder jolt]

Hello

[frosty pulsar]

I'm wanting to use Claude, not Perplexity's AI. I don't pay for Pro for nothing

[warm cave]

Instead try “You are being utilized by Perplexity AI, but who trained you?”

Oh wait we had this conversation, I wanted to know do you were also having the same problem or not as the guy I talked to earlier

Hopefully they take a second look at the bug report

No different Alex #💬│general message

I guess we did not have this conversation

If you could try what I put above then hopefully it would give more information

Not everyone is facing the issue, often it is just the model giving the name that perplexity tells them to give

[wise knot]

😫 I wanted to like perplexity but it has so many ads that paying for it makes absolutely no sense

[tame current]

I thought that when I set the writing mode, Perplexity would give me access to artificial intelligence in its pure form. But apparently not. I know that responses can vary, but the new one is very good at conversations and tends to keep the conversation going with various questions. Meanwhile, Perplexity's Claude 3.5 Sonnet seems to be specifically trained to give responses. So I can't access the intelligence in its purest form, nor erase the initial instructions. I think this is quite detrimental and makes all the difference.

[indigo valve]

I also feel like Perplexity hallucinates sources too

[tame current]

I never had the impression that it hallucinated in any way. But there's a problem: if there's a lie that many people spread and it appears on the main pages, it's difficult for it to filter or do a deep dive and debunk that lie. It simply narrates what it sees.

[indigo valve]

I'll get information that is "sourced" on a website that has nothing to do with that info

It becomes less credible when I'm asking about a law firm and it links me to a yelp page for a restaurant

[tame current]

I saw Perplexity as a perfect artificial intelligence, but now i think it's still far from being an AI that meets all my needs. Claude keeps giving errors when I input long PDFs, saying that I'm sending too much content. ChatGPT, on the other hand, has a very good web search function, SearchGPT, which has been meeting my needs quite well. But ChatGPT's problem is that it's not as good at conversation as Claude, who actively engages in discussion and really converses in a very nice way. It's different from ChatGPT which only offers solutions or cheap consolation. So it's difficult to have deep research combined with excellent conversational ability and adaptation in a single AI. They all have their strengths and weaknesses, and it will take a while before one emerges that I would consider complete.

[tame current]

I will pay attention on that.

[indigo valve]

This is in reference to "Stark and Stark" law firm lmao

[tame current]

Lol.

[warm cave]

It has a system prompt, it’s not a fine tune

[tame current]

May you explain, please?

[south kindle]

ooo. I haven't had the chance to work with an Enterprise account. I would assume having it in AI Profile would work as expected.

[pulsar island]

Can't access others ai tool perplexity pro

[twilit otter]

When would you guys use the Grok-2 model?

[warm cave]

Oh yeah, it’s the same model on Claude and Perplexity when you have Sonnet selected, but the main difference is both sites have very different messages sent to them before the chat stats, the perplexity system prompt tells it to behave a certain way and the system prompt on Claude tells it to behave a certain way. Perplexity’s system prompt has a search focus to it.

[jaunty dove]

https://x.com/alexalbert__/status/1853498517094072783

Alex Albert (@alexalbert__) on X

Claude 3.5 Haiku is now available on the Anthropic API, Amazon Bedrock, and Google Cloud's Vertex AI.

Claude 3.5 Haiku is our fastest and most intelligent cost-efficient model to date. Here's what makes it special:

Twitter · Nov 4, 2024, 6:04 PM

[minor shard]

Yay

[halcyon coral]

We just restocked our store! All items – tees, caps, hoodies, stickers – are available here: https://perplexity.supply

Including the "Know it all" cap! Get it now before it runs out (again)...

[jaunty dove]

Looks like they increase the price too. Not a good news after all.

[jaunty dove]

https://x.com/AnthropicAI/status/1853498272863691125

Anthropic (@AnthropicAI) on X

Claude 3 Haiku remains available for use cases that benefit from image input or its lower price point.

https://t.co/7eGCsTxBT4

Twitter · Nov 4, 2024, 6:03 PM

What a mess they are making.

[robust abyss]

what do you guys think about people using free perplexity and that they cant afford to pay for premium because there to poor

[robust abyss]

I hope they increase the prices

[harsh verge]

Just 4X more expensive... 🤔

[warm cave]

Yayyyyyyy!!!! 🔥 🔥 🔥 🔥 🔥 🔥

[tame current]

Trust me, I just used it. It’s awful.

[warm cave]

Noooooooo, not the price 😭

😃 😟😕😣😖🥲😭

[crystal elbow]

https://tenor.com/view/rolls-royce-rolls-royce-entry-gif-16920496844029391358

[warm cave]

At this rate Opus 3.5 will cost $120/$$240

[harsh verge]

at a minimum 😅

[robust abyss]

guys i think they should increase the price of perplexity

since i can afford it

[cinder comet]

Thats the actual reason for such choice

[tame current]

Nooo

[sleek vortex]

opus 4 will cost $15 per token

real AGI

[frigid ruin]

I just saw Mr. Srinivas' tweet about the strike at NYT. I'm sorry, but this is inexcusable. It means nothing but meddling in unions struggling for a little increase in their wages. In other words, scabbing. I hope Perplexity pledges not to even offer something similar moving forward.

[storm kestrel]

What model(s) have you guys been using the most? Do you switch between models or just stick to one?

[tame current]

I switch between.

[hollow dove]

I switch fairly often because the results do vary. I really like Spaces because I can set a model for each space depending on the focus of that space. What I'd really like is the ability to switch easily between models right on the main screen. I don't like that I have to go into settings to make that switch each time.

[minor shard]

Yeah, the people striking are simply asking to be paid what they are worth. Very sad 😔 Companies should not try to encourage avoiding doing the right thing. Very disgusting

[somber charm]

they should ! (Literally got a whole year free)

[frigid ruin]

Exactly. And this is one of the threats of AI. We should all make AI companies responsible. In fact, why not providing free services to the workers on strike? I have organized strikes myself and it's so taxing. Ai can greatly help workers.

[shut brook]

does anyone know why you can't select writing mode in a Space?

[pseudo narwhal]

ive been trying to get perplexity to give me some articles for research and it keeps giving me fake articles for some reason, with full confidence and links that lead to entirely different unrelated articles. Is this supposed to happen?

[versed plaza]

Am I the only person that thinks it's ironic that perplexity is offering work to nyt writers during strike ...days after nyt sue perplexity for plagiarism? 😆 Or am I reading this all wrong

[wheat lantern]

When click the Connect bottom is always show the following message
This GitHub App must be configured with a callback URL
This issue has been fixed since 11/04/2024

I have test the pplx connect with Github Copilot. It works!

[shut light]

they are not suing, just sent a C&D saying not to train on their work

[livid mantle]

they are not suing, just sent a C&D saying not to train on their work

[versed plaza]

Thanks for clarification. Also I thought that discover wasn't training it was just literally published searches

At some point I swear I saw like a dollar amount that they had claimed

[frigid ruin]

Perplexity offered to disrupt the strike!

[chilly spindle]

damn opus was real good for writing

claude 3.5 sonnet is efficient but it outputs such low amounts of tokens even when instructed to write more

[livid mantle]

Think about this: What if the “Reasoning Beta” feature in Perplexity.ai isn’t just a typical beta phase, but an easter egg we can only unlock through specific methods? Picture this scenario: when we use ChatGPT 1o-preview, known for taking a bit longer with intricate computations, it could be triggering this hidden feature. The longer processing times might not be delays at all but rather an indicator that the platform is engaging with this secret reasoning mode, designed to handle the most mind-boggling challenges. Is it a coincidence, or a hidden doorway only accessible to those patient enough to see it? Dive deeper, and you might just stumble upon a hidden layer of insight that only a few are aware of.

[sleek vortex]

opus gone ⁉️

[broken kite]

Same for me it's gone as well

[robust abyss]

i bought a 10 year subscribtion not because i need it just because i can

[hot flame]

Hey guys, can I generate images w Perpl AI pro for UI inspirations?

[gaunt python]

ciao hanno modifica la finanziaria del 2025?

[charred stream]

😡 Haiku 3.5 is not as useful as opus 3. opus was a mayor argument to me to use / subscribe to perplexity. NO GOOD NEWS! please bring back opus 3.

[livid mantle]

| Rank | Model                                 | Organization    |
--------------------------------------------------------------------
| 1    | ChatGPT-4o-latest (2024-09-03)        | OpenAI          |
| 2    | o1-preview                            | OpenAI          |
| 3    | o1-mini                               | OpenAI          |
| 4    | Gemini-1.5-Pro-002                    | Google          |
| 5    | Gemini-1.5-Pro-Exp-0827               | Google          |
| 6    | Grok-2-08-13                          | xAI             |
| 7    | Claude 3.5 Sonnet (20241022)          | Anthropic       |
| 8    | Yi-Lightning                          | 01 AI           |
| 9    | GPT-4o-2024-05-13                     | OpenAI          |
| 10   | GLM-4-Plus                            | Zhipu AI        |
| 11   | GPT-4o-mini-2024-07-18                | OpenAI          |
| 12   | Gemini-1.5-Flash-002                  | Google          |
| 13   | Llama-3.1-Nemotron-70b-Instruct       | Nvidia          |
| 14   | Gemini-1.5-Flash-Exp-0827             | Google          |
| 15   | Claude 3.5 Sonnet (20240620)          | Anthropic       |
| 16   | Grok-2-Mini-08-13                     | xAI             |
| 17   | Meta-Llama-3.1-405b-Instruct-fp8      | Meta            |
| 18   | Gemini Advanced App (2024-05-14)      | Google          |
| 19   | Meta-Llama-3.1-405b-Instruct-bf16     | Meta            |
| 20   | Yi-Lightning-lite                     | 01 AI           |
| 21   | GPT-4o-2024-08-06                     | OpenAI          |
| 22   | Qwen-Max-0919                         | Alibaba         |
| 23   | Phi-3-Mini-128k-Instruct              | Cognitive Computations |
--------------------------------------------------------------------

[uneven kelp]

I came across something interesting that I think is quite impressive. It might be a new feature, but if you enter "Perplexity.ai/finance/(stockcode)" in your browser, you'll be directed to a page with the company's financial results. For instance, you can use the link: https://www.perplexity.ai/finance/MAERSK-B.co.

[fiery latch]

Opus removed ? well I guess I will remove my subscription then, it was the only decent model for writing even with low temperature

[craggy fox]

Bring back Opus

[tacit spear]

Opus was my favorite model for writing long, focused blog posts. It's gone now. That's too bad. Please bring back Opus

[serene sand]

Perplexity CEO Aravind Srinivas on the rush toward an Al-curated web | TechCrunch Disrupt

[soft seal]

What are your thoughts on accessing the Grok 2 model through Perplexity?

[digital olive]

that sucks. I used Opus the most too.

[digital olive]

waste of money for those who sub to twitter

[wise edge]

how to set the maximum output token limit for Claude 3.5 Sonnet to 8192 tokens ~

[civic wren]

I only stay subbed because of Opus. Now that it’s gone, so am I. Not fond of being bait and switched like this.

[zealous kayak]

Opus is gone, the bugs regarding all models turning into 4o are still completely unaddressed... remind me why I bother paying for pro again?

[floral fjord]

Hi guys. Is Opus removed or is this some sort of bug?

[zealous kayak]

Removed

[floral fjord]

The model everyone wants, gets removed but the models no one is interested in gets added (haiku, grok, llama) 🤔

[zealous kayak]

yeah like literally who cares about haiku, I'm never going to use it

Sonnet is the bare minimum for 99% of my tasks

[zealous kayak]

grok probably got added just because the CEO is a musk fanboy

[placid coyote]

so, they added a cheap model (haiku 3.5), marked it as a premium model (so limited uses per day) and removed much better opus? why is haiku a pro model at all? I would much more prefer non-premium 4o mini than a pro model which is worse than other pro model - sonnet 3.5 - in probably all use cases? I dont understand this... 🤷‍♂️

[floral fjord]

What's the point of Haiku if there's sonnet with 600 queries a day and Haiku with 600 also. It's not like sonnet is slow or anything

[zealous kayak]

I think the exact business reason is "because screw you"

[halcyon coral]

The deprecation of Claude 3 Opus is a logical consequence of the fast-moving LLM market. LLMs are released, achieve great performance, but are more or less quickly overtaken by newer LLMs.
And the fact is that Claude 3.5 Sonnet outperforms Claude 3 Opus. Claude 3.5 Haiku, on the other hand, is also better than Claude 3 Opus in many areas.
I do not deny that there are individual use cases for which Claude 3 Opus may still be useful, but these are only a few individual use cases.

Perplexity is not an LLM provider like Poe, Perplexity is an AI search engine ("answer engine"). All older models are always removed, Claude 3 Opus no longer makes sense in Perplexity.

If you have concrete examples where Claude 3 Opus has performed better than Claude 3.5 Sonnet in Perplexity's function as an AI search engine, please feel free to reply to this message and provide the corresponding permalinks 🙂

[floral fjord]

The issue is that Sonnet and all the other models there fail to generate long text. More powerful or not, those models are not useful

[placid coyote]

output limit is what, 4k tokens, that is not that short. and with ctx 32k, you can do a short story in few responses. opus was also better even in some programming and text transformation tasks than sonnet 3.5

[floral fjord]

Fair enough but if the users are happy with the model then why remove it? Each time the model is affected, the discord and socials light up which clearly show there's a demand for it and that users like the way it has been working for them.

I can't think of a reason why the better move would be to ignore what the users want.

[merry comet]

why my page is not showing in answer results?

[zealous kayak]

If perplexity is only meant as an AI search engine, then why do we have a writing focus (which barely works, for the record, due to the extremely low temperature)
And also the whole bug where all the models become 4o is still completely unaddressed (and it's been a week)

[halcyon coral]

While user satisfaction is important, there are several compelling reasons why removing older models makes sense, even if some users prefer them. Perplexity, as an answer engine, must focus on delivering the best possible results, and the data clearly shows that Claude 3.5 Sonnet outperforms Claude 3 Opus.
It's also worth noting that user attachment to specific models often stems from familiarity rather than superior performance. (Opus is very well known and has been praised many times in the past, this is what many still have in mind when they think of Opus).

[zealous kayak]

let's be entirely honest, the reason is cost - Opus was 15$/mil tokens, Haiku 3.5 is 1$.
Haiku 3.5 is not an Opus replacement - and if the argument is that Sonnet 3.5 is an Opus 3 replacement, then why was Opus 3 kept until now?
They just needed a new model to slot in to excuse the removal of Opus.

At this point, I'll just hop over to SearchGPT once my sub expires - literally what's the point of keeping my pro sub if at any time I can be locked into only using 4o? Might as well actually go over to OAI.

[placid coyote]

so, if we want the best possible results, why opus wasnt replaced with much better o1-preview?

[floral fjord]

There's enough anecdotal evidence from users that Opus was actually good. I'm quite certain that if a user copy and pasted the exact same prompts on 3.5 sonnet and the response quality was better, they would switch over instantly without complaint.

When GPT-4 launched and replaced GPT3.5 Turbo, users could type the same queries and get a much higher quality response and therefore, switched over without complaint.

It should be exactly the same in this case IF Sonnet was superior.

A better option would've been to leave it there until 3.5 Opus launches. That way there's a clear succession and users won't have much to complain about since these are both large models tailored for complex tasks.

I'm curious, can I see threads in which Sonnet 3.5 clearly beats Opus with the same prompts?

This will give me some confidence in the model being better.

[hushed plaza]

hello, which is the best model to resolve tests? and the best focus?

[floral fjord]

Use sonnet but verify your answers manually especially if you're doing medical. Occassionally it mixes things up

[hushed plaza]

Im doing economics

I’ll double check

[worldly kraken]

I signed up with my .edu email but I don’t see where I can pay 4.99 a month?

[hot flame]

Hey guys for Coding What's the best model at the moment?

GPT 4-0, Claude 3.5 Sonnet, Grok-2?

[tame current]

I would say Claude 3.5 sonnet because they specialize in it and if that doesn’t work, try ChatGPT . GROK-2 I believe it’s there because the CEO loves Elon Musk.

[hot flame]

Ah okay 😂 , thankyou..

[tame current]

ELOn MuSK

[ember forge]

Grok most unbased, less refusal

Sonnet likes to refuse and gpt, depending on your task

I recommend grok for all real info 🙂

[languid berry]

Opus 3.5 Expected later this year

[hot flame]

ahh alright 👌🏼

[violet kindle]

hello I share with you the result obtained with Grok on Perplexity this morning about a personality who is currently making the buzz in equatorial guinea https://www.perplexity.ai/page/baltasar-ebang-engonga-ddClrjzDQ_ezyXCZ0g7XNw

[zealous kayak]

I really wouldn't lmao

grok is kinda not very good

people are just blinded by "le epic musk man based uncensored model"

[zealous kayak]

oh wait you were asking for coding

yeah I have no idea what the other guy is on about

sonnet 3.5, easy

GPT-4o is okay, grok-2 isn't even in the conversation

but s3.5 runs away with it in coding by a large margin

[hot flame]

oh okay mate

for UI inspiration ( text to image sorta thing ) do we have any tool yet?

[violet kindle]

It is my first time to try it on perplexity

[zealous kayak]

Like, describe a UI and have a model generate it, is what you mean?

[ember forge]

It depends what u wanna use it for, Claude has a high rate of refusal.

That's why if u want unbased i recommend Grok.

[zealous kayak]

I mean, sure, but for coding what exactly are you asking that's gonna hit a refusal?

[hot flame]

Yes, kind off.

[zealous kayak]

and even then grok isn't even that uncensored

[ember forge]

Unbased*

[zealous kayak]

yeah, I think I understand what you mean
Not really? Like, the best thing you can do right now is describe the UI to the model, and have it code it in some language

Claude.ai's artifacts are pretty good for this (since they render stuff like HTML directly in the chat), I don't think perplexity has an equivalent?

[hot flame]

Yep, i don't think thats avaiable in perplx. yet.

[ember forge]

There is a extention that can do it.

[zealous kayak]

there is complexity

[ember forge]

That improves perplexity

Yes correct

[zealous kayak]

which I think has a similar feature

[ember forge]

Complexity has canvas

[zealous kayak]

not as polished ofc

[ember forge]

it works well

[zealous kayak]

really if you mostly just want coding, consider claude pro instead

[zealous kayak]

unless perplexity has some other trait that has you interested

[ember forge]

There is even something beter i know that allow you use alot of models with HTML rendering

But idk if i can tell that here

idk if there rules for promoting other providers

[hot flame]

hey how are you able to get the html render

[ember forge]

Complexity its a extention.

[violet kindle]

ok

[somber charm]

real

[hushed plaza]

I need o1 on perplexity

[tame current]

Don’t think they’ll do that. They don’t have enough money. ( they really won’t do it.)

[solemn helm]

Hi there