💬│general

[maiden elm]

I want to use it for an app I'm building but I'm not a developer so i have nothing so far.

I mean I have something already but I haven't hooked it up already... I'm on it but it's slow cause I'm improving the code

My program sends an instruction and 3 pieces of information to the AI, then for each of three informations the AI returns a description that is inserted along the the terms, that way the user has less work to do.

[thorn moss]

Is the announcement today or tomorrow?

[sweet jasper]

Tommorow

Nov 6

[thorn moss]

and we probably expect GPT vision tomorrow or tuesday that is why we expect that day it will be available on perplexity like @neat elk said?

I am really hoping for this

[tame current]

Mezclar

[dire fjord]

this ai is really cool

[wide stratus]

According to my research, LLama isn't "open source". You have to accept a license on the meta website first, before you can download it from there. So it's not legally open soure.

[digital siren]

Gpt-4v and advanced data analysis on perplexity would see me cancel my ChatGPT plus subscription.

[zenith atlas]

AFAIK most of the open sourced stuff have some sort of licenses, but yes LLama license is not approved as open source license. But in fact most of the extra regulations will kick in only if your product have more than 700 million monthly active users so, in my opinion, for most purposes it can be considered open source

[calm marten]

This

Perplexity is a strong competitor tho

@fickle pond what is the role of Explorer?

[fluid hawk]

well, you can also download it from huggingface, which is legal according the the llama2 license

and if you download it from another user, then you don't need to request access

[halcyon coral]

If you post something that other users really like and they give your answer a star, you will get the Explorer role after a certain number of stars and your message will be immortalized in #⭐│starred 😉

[pliant ridge]

I think the perplexity web search extension is not very good and I still wanna use google for a good chunk of my searches, but you can add perplexity as a site search on chrome and use pplx as a shortcut to access it

[tame current]

Perplexity failed non linear thinking test

https://www.youtube.com/shorts/rqI3xF9vDd0

Current Chat GPT and Pi.ai passed the test:

[digital siren]

https://www.perplexity.ai/search/854bdc73-07b6-49b6-9de9-5b76a29193ef?s=m

I would love a hybrid writing and all mode. When I’m iterating on a long report, I want to give the model context without it trying to search the web. Then after it has all the context I want it to search the web and prepare the report.

[tame current]

Interesting. As you can see, I got a different answer (9 hours) in my reponse

[sweet jasper]

You are using gpt3.5, gpt4 handles this with ease

[tame current]

GPT 3.5 gave me the correct answer

#💬│general message

[sweet jasper]

Thats interesting

Maybe they fine tuned the model to answer that riddle

I have seen it b4

[quartz mountain]

Labs.perplexity.com

[tough turret]

hi all! Nice to be using ppx pro! I signed up mostly to have access to the API as I have a client that could benefit directly from the research (i.e. grounding) capabilities. It was a bummer to notice though that the grounding elements were not being returned in the API. is this the expected behaviour?

[zenith atlas]

API for now uses llama based models without access to internet AFAIK, full API support for perplexity internet search (it’s called RAG or smth) is planned but we have no date for it

[halcyon coral]

#💻│coding message

#⚡│ask-community message

[sweet jasper]

Will perplexity keep using gpt4 api or is there plans to change to gpt4.5 turbo

Or that being offered as a separate model

[tough turret]

Ok, thanks @zenith atlas and thanks for the heads up to dont expect a date. I will transition to a temporary solution with the client until we wait for this to be available. Are you aware of anything that could yeld similiar results as what ppx web deilvers today? I couldnt get anywhere near these results.

[zenith atlas]

Icelavaman sand a link above where it states that RAG is planned for late November (I wasn’t aware of that) and I can’t recommend you anything as my programming knowledge is pretty low and I don’t know how this api’s even work😅

[tough turret]

dont worry, thank you for your help anyway!

[tough turret]

Great one. Thanks for that.

[tough turret]

The solution I am designig for my client ideally involves the ability to put perplexity on top of 200 pdfs. For this to work, I am guessing I have to wait for the " Embeddings API " feature realease, correct? When this is available, It will be possible to upload these PDFs to an endpoint and reference JUST THEM on my inquiry?

[sweet jasper]

That sounds about right

[sweet jasper]

Is there currently a rag chabot implementation that reads from github repos

[tepid light]

Hi

[muted pine]

LET'S GOOOOOO

GPT 4 TURBO - 128K TOKKEN

GPT-4V

DALL-E3

API OUT NOW !!!

[vapid summit]

Will GPT 4 turbo be available on Perplexity?

[hallow linden]

I hope so. i want that supercharged pdf reading

[buoyant mist]

I really really hope so

[sweet jasper]

Gpt128k for less understanding

😭

[tropic crest]

I think GPT4 all tools supports that too

pdf reading

[hallow linden]

right, but I want that 128k context

I want it to be able to read a textbook

[thorn moss]

Is there a problem with discord? Anyone?

[hasty otter]

Hello

[ionic solstice]

For perplexity API, I want to use it on a small community search engine so people on a page can send search requests for free, will I incur any extra charges? (I am on perplexity pro plan) for doing this? I noticed its rate limited here: https://docs.perplexity.ai/docs/rate-limits it almost sounds too good to be true, in not being charged extra for doing such a thing, but wanted to be sure its not abusing the API at the same time Perplexity API that is therein, I only expect a small number of users.

[ionic solstice]

hi

[halcyon coral]

Yeah, but is almost fixed again

[hasty otter]

Who can help me? where i'm can take API Key in my profile in Perplexity?

[halcyon coral]

do you have a pro subscription?

[hasty otter]

no, i'm need buy this?

[halcyon coral]

atm the API is in beta only for Pro users.

[hasty otter]

okay

[thorn moss]

Hello bro @barren pawn any idea when will perplexity do the real deal? Any spill? Hehe

[halcyon coral]

There are no extra costs atm 😉

[ionic solstice]

If there is will I get notified

the project I am doing I dunno at the min due to high costs like this ideally want it to be non profit but its not easy with AI haha

[halcyon coral]

Yeah it will be announced 🙂

[ionic solstice]

I am trying to enter the market on AI in parts of society where its largly ignored so poorer areas and such

[ionic solstice]

https://vitus.ai so far I just list your two sites for the ease of access and accessibility, but I thought maybe its better to replace that with a search thing that links into your API

my issue is future wise, I would not have the money to keep using the API, ideally non profit and AI don't mix well due to high cost of GPU's haha

[digital siren]

Please give us 128k gpt-4 turbo 🥹

[thorn moss]

i cant waitttt

[formal parcel]

I DEMAND VISION

[wide stratus]

Interesting... So the entire license doesn't make any sense anymore... 😂 WTF Meta???

[wide stratus]

Wtf what aspect ratio is your display? Hooooly cooooow!!! 😂

[muted pine]

Perplexity or Poe drop first DALL-E3 and GPT 4 128k ? LET'S FUCKING GOOOOO

[digital siren]

iPhone 15 pro max

[digital siren]

Did Poe drop it already?

[muted pine]

not yet

[thorn moss]

Oh gotta see that in my wife this week. I ordered it. I am fine with my s23 ultra. 😂😂😂

[digital siren]

I switched from a OnePlus 10 pro to the pro max. It’s difficult to get used to lol. There’s more available ai apps on iOS though

[thorn moss]

I was iphone way back but its too limiting. Thats why i went android

Then i told my wife when your 15 pro max is delivered gimme your 13. 😂😂😂

[formal parcel]

They said no GPT-4.5

Isn't GPT 4 Turbo GPT-4.5?

[dark solstice]

question, you guys added a "prompt" option on the creation of making a collection on perplexity what is that about and how affecting is it?

[hearty marlin]

cant agree more

[digital siren]

It’s just like custom instructions

[cloud sequoia]

how do you think about OpenAI devDay。

I DEMAND VISION too

[cloud sequoia]

GPT 4 Turbo is only cheaper GPT 4 , not gpt4.5

[halcyon coral]

GPT-4 Turbo is GPT-4 Vision, faster and with cutoff in April 2023

[regal frigate]

Does Perplexity plan to integrate the 128k GPT-4?

[wide stratus]

Oooof, what has Apple done? 😂 I'm so glad I never owned an Apple device. 😆 Motorola is way to go! 😁

[wide stratus]

Wow I just discovered that that OnePlus 10 Pro's aspect ratio is even slightly worse, at 20:9 compared to the iPhone's 19.5:9 one. 😝

[cloud sequoia]

What core new capabilities were released at OpenAI DevDay? Where are the new opportunities in the future?

how do you think about this topic?

[sweet jasper]

More context window, faster gpt4

At the cost of reasoning abilities

As was the case with gpt3.5 turbo

Gpt4 in perplexity still uses the old gpt4 api rather than turbo

So for now its safe?

[pastel ferry]

I was wondering if its possible to integrate the API key and SillyTavern?

[digital siren]

OpenAI says it’s a smart model. It may just be because it’s in preview right now. I haven’t noticed a major difference.

[zenith atlas]

From the day gpt4 was released people are saying that it’s getting dumber, and for all this time I have not seen any changes in how it responds to me

[digital siren]

I think people conflate dumb with censored. GPT-4 usually gets more restrictive as OpenAI sees areas of “abuse”. Yes, it makes the model slightly less useful to be restricted, but doesn’t make it more dumb on the remaining allowed tasks.

[cloud sequoia]

just cheaper and faster for perplexity. nothing difference.

[digital siren]

Larger context and multimodal is a big deal for search and projects on perplexity

[cloud sequoia]

why , and what can we expect next?

[cloud sequoia]

why , and what can we expect next?

[digital siren]

Because you can upload images and perplexity will be able to analyze, explain it, and research it. In addition, larger context window helps to analyze long pdf documents and maintain the original context when producing a report.

[zenith atlas]

This and also the fact that gpt4 is not consistent in his responses - it can react to the same prompt in dozens different ways even without version changes. I mostly just start a new chat if I feel that it’s acting dumb and it mostly gets fixed in 1-2 extra tries

[cloud sequoia]

but chatgpt has the same ability as perplexity , which can search bing and rewrite the answer

ChatGPT`s new feature will threat the new user growth of perplexity ?

what make you still using perplexity?

[sweet jasper]

Perplexity search is way faster

Than gpt4 browsing

[zenith atlas]

bing search works with less sources than perplexity, and its finetuned with different goals.

I personally wait to test file upload with gpt4-turbo to check if chatgpt cuts the files or feeds them fully to a model, as its in higher priority for me then internet search

[dusty void]

Is there anyone who want to use my referral code? https://perplexity.ai/pro?referral_code=V1HM4CO5

[cloud sequoia]

no， we are all Paying users.

let`s continue the topic ? what make you still using perplexity?
I want more answer , I dont know why I should continue use it after the openAI day.

[thorn moss]

Just wait for the devs to update and you will see

I trust perplexity

[halcyon coral]

Better search feature (Perplexity= AI search engine) and the features will be improved soon

[cloud sequoia]

thanks , I believe too ~
there are no new feature since — 10/07/2023 .

just wait, and we can see .

[halcyon coral]

Pplx API

[dusty void]

Ohh I see is there any platform that I can share publicly?

[cloud sequoia]

sharing channel？ or maybe x.com is a good idea.

[dusty void]

Okay thank you @cloud sequoia

[wooden portal]

hello

[dusty void]

May I ask, what is API keys? Like in Perplexity Pro you can add API keys. I ask Perplexity but seems that I can't understand at all and still new in AI field platform

[dusty void]

Hi hello

[wooden portal]

can perplexity ai access the internet

[crisp roost]

It can search on the web

[crisp roost]

With an API Key, you can create your own programs using the models on labs.pplx.ai

[halcyon coral]

For developers to create own products, you can find more information on https://docs.perplexity.ai

[dusty void]

I can create my own program? Like app? Software? Or just program that we create in Perplexity that focusing on defined tasks?

[muted pine]

Just show Poe already have dall-e3 and gpt4-turbo

[thorn moss]

That was fast. I hope perplexity can do it today

[muted pine]

Yep Same !

[tame current]

I think the API can be replaced directly, and the model can be changed. I guess I am testing other ones now.

[thorn moss]

I dont know bro. Devs seems quiet

I really hope today

[tame current]

yeah

[spiral cedar]

Has there been any updates on using the API for commercial use yet ?

[wooden portal]

can perplexity ai open links directly?

[halcyon coral]

No, still in Pro beta

[thorn moss]

Any spill the beans on gpt vision bro?

Today?

[wooden portal]

what are the benefits of perplexity ai pro over chatgpt plus

[tame current]

Virtually unlimited gpt4 every day

[wooden portal]

perplexity gpt4 cant open links

[wide crow]

I like both perplexity and chatgpt premium, but I do think openai is going to just devour all other ai companies tbh

[carmine fern]

🤣

[maiden elm]

One reason that may not happen is specialization: open AI cannot develop on all fronts...

[thorn moss]

day 2 no word from dev. sad

[halcyon coral]

It has already been announced that DALLE 3 and GPT-4 Turbo will be added, if these features are available it will be announced! 🙏

[thorn moss]

you mean GPT vision also right?

but probably not today? 😦

[halcyon coral]

Yeah

[muted pine]

api are already on

[halcyon coral]

Yes, but the whole thing also has to be implemented.

[muted pine]

yep juste need to wait like 3-4 weeks

POE are already add DALL-E3

[tough turret]

Hello all, from a technical standpoint, can someone explain this discrepancy between but goes out on the Web x API ?

How can I work to get my answer recieved from the API closer to what I get on the web?

[halcyon coral]

You will definitely need to customize your system prompt, but it is very difficult to get similar results as different models are used and the API does not have internet access.

[thorn moss]

3 to 4 weeks tf. haha

Oh boy. Yeah actually i noticed perplexity really updates slow lately

[tough turret]

Is this something that can be expected soon on the API?

I undestand that making smt like this happen would take several upgrades to the API:

• make whatever is the model/architecture of ppx's web version to be available on the API? (i didnt see this in the roadmap)
• make internet search available on the API? (i didnt see this in the roadmap)
• make RAG available on the API? (I undestand that this is in the roadmap, https://docs.perplexity.ai/docs/feature-roadmap)

[muted pine]

Yep i will personaly stop my subscribe before update

[thorn moss]

geeez let us give the devs their personal life. maybe they are on team building or something. 😂😂😂 JK.

more important is, its gonna come

yeah boy!!!

[halcyon coral]

Web search is on the roadmap (LLM-RAG) 😉

[muted pine]

For sure that reason why i say 3 or 4 week for good implementation

[tough turret]

I feel this so much, everywhere, every team involved with gen ai, pressed by users, clients, friends, society hahaha... It is like an epidemic of data scientists / devs on burnout...

[thorn moss]

yeah. Reason also why they dont interact so much cause more question will come up. I understand though. I also have a job. 😂😂😂

and speaking of the big updates yesterday, it is better for them to stay quiet. 😂😂😂

[viscid star]

Hi Perplexity, would you consider adding the Phind 34B model? It is much better than CodeLlama

Phind-CodeLlama-34B-v2 achieves 73.8% pass@1 on HumanEval.

[sweet jasper]

Isn't phind proprietary

And not accessible

[sweet jasper]

It won't take long, you just have to add a different url

[thorn moss]

say wuuut

[viscid star]

No it's open sourced

https://huggingface.co/Phind/Phind-CodeLlama-34B-v2

[desert field]

Hello, I just susbscribed as I see great prospects with this application. Question: What are the best use cases for each model? (Perplexity, GPT4, Claude 2)

[desert field]

Perhaps I've answered my question here: https://www.perplexity.ai/search/What-are-the-qQU5X7ZZQ1C_WknOsh.f4Q?s=c

[muted pine]

no lol if it's only a URL he would have changed it after the announcement

what's weird is the silence of the teams, that why i say 3-4 week min

[digital siren]

I imagine they’ll test it to make sure it doesn’t degrade perplexity products

[somber pine]

Hello, @everyone
We're developing the LangChain Project.
Plz DM me who does have experience with Langchain.

[thorn moss]

They are silent because many follow up question will come but downside is a lot of questions and uneasiness will come up as well as impatience.

Actually its better to have a little announcement or a little estimate of deadline. It kindda ease.

[digital siren]

Does anyone know if google Gemini is still going to come out anytime soon? It was rumored to come out this fall and be better than gpt-4. It’s been radio silence for a long time though.

[halcyon coral]

I assume that this will be released next year...

[tame current]

https://youtu.be/L18ocun3W18?si=oFLUT7o5et7DLZ-x

[muted pine]

just post in announcements for exemple

[tame current]

correct , it is happening right now

soon end of perplexity as it's just a wrapper

[tough turret]

Hi, is there any way to have more than 4 pdfs used in a prompt/assistant in Perplexity?

[wraith crow]

thats not possible, 4 files is the maximum atm.

[tame current]

Eliminating wrappers one by one

[tough turret]

Whoever tweets that is totally disconnected from the reality of the use GenAI inside companies.
Every corporation has its own use case, basically we have been doing this as a consulting gig since earlier this year and haven't done 2 equal bots since. Every interation has its own requirements. I dont think wrappers are going anywhere.
As for Claude and Perplexity they seem totally fine to me and have a reason to exist (and will have for a while). Theres no solution better for a problem a client of mine is facing, as the one that Perplexity deliver. the way it formats the content and presents it merging documents, internet text generation, is way ahead of GPT in my opinion. Claude dives deep in longer contexts like no other contender.q

As long as perplexity keeps blending SoTA content presenting and bringing inovative ways to interface with it the way it does, it has its moat, in my opinion. OpenAI has no research improve in Generative Interfaces since the begining. They basicall have just added some quick prompts and etc. Its been a year since their launch and we are still texting to get where we want to get. Doesent make sense.

[lusty pilot]

While it's accurate that certain businesses may flounder if they don't differentiate, it's also worth noting the market sometimes allows similar products to flourish. This success often relies on marketing, clever business strategy, or the establishment of an efficient ecosystem amongst others. And to assert that perplexity is just a wrapper would be a misjudgment. It in fact searches and aggregates knowledge at a much faster rate than chatgpt with gpt-4-turbo can.

[tough turret]

I think that's a lot of ground to be conquered in the interfaces front, and PPX seem to be ahead. I myself am tinkering with some concepts of what could be a liquid interface that would adapt to the users requests. The form below that was given as the response to the users request was created dynamically.

This not only looks more logical, in my opinion, but also it actually saves a ton of back and forth in the conversation (money/time).

[dark solstice]

i have a questions

is it normal to put your name and birth date as your password?

or it is it stupid?

but as compared to putting a bunch of random letters

[digital siren]

Did you develop this? I like that interface!

[tough turret]

yep, nothing special about the code, I guess.... have a look and see how you would improve it and let me known.

[tame current]

You have your API key in the script ☹️

[tough turret]

i have removed!

havent I?

[tame current]

The only disadvantage of ChatGPT for web browsing is speed as of now, I do not see any advantage of using Perplexity instead once that’s improved

[tame current]

On top of scripts.js

[tough turret]

rvked tks

[tame current]

🤙

[tough turret]

anyway this is the prompt that generates the form

{
role: "system",
content:
you are going to determine if you need additional context from the user to answer his question. if you need additional information, you will ask for it using a HTML form. That means you will write an HTML form with all the information you need from the user to give the best possible answer. Use the following form elements when needed: text, submit, date, range, checkbox, file, radio, textarea, and select. Format the output using css and tables to be aesthetically sound
}

[dusty void]

Hi hello. I'm still can't understand AI very much and I do have one question, what is LLM and other idk what it is? What's the capability and purpose of each one of them?

[ionic rock]

Question. If i report an answer and for example flag it as not useful and even give a short reason, does anyone read this and will it be used to improve the search results for future users?

[dusty void]

I think so, maybe. Idk so much about it

[digital siren]

Perplexity offers you access to Claude and gpt-4 for one $20/mo subscription. I’d be happy to pay for a “wrapper” if it bundles several good models with a nice UI. Poe does this, but I don’t like their UI.

[digital siren]

https://www.perplexity.ai/search/1172b438-f6b5-4180-8805-95b72f89d946

[cloud sequoia]

when will perplexity open their open-platform ? especially RAG ability. I cant wait.

[halcyon coral]

Hey, you can use /dream in #🎨│dream 🙂

[dusty void]

Thank you @digital siren

[twin frost]

Each microservice instance can have an independent database instance.
Right
Wrong

[shadow stump]

guys tell me why I should use perplexity over phind

[red canyon]

I'm getting this error a bunch this morning: "Error in processing query." Anyone else?

[broken vector]

what about the servers

[broken vector]

I can't use perplexity at all

[thorn moss]

copilot is down as well

[halcyon coral]

Yeah, also FT GPT-3.5 #🔔│status

[fathom narwhal]

same issue. doesn't process queries for now, not with GPT4 nor CLAUDE 2.

[cloud sequoia]

I cant use perplexity right now

[halcyon coral]

GPT-4 and Claude should work, but not with copilot.

[halcyon coral]

#🔔│status

[twin oak]

When will it be up again?

[halcyon coral]

No eta, major outage from OpenAI

[gusty blaze]

Why can I no longer press enter to search? Whenever I press enter to search on my Windows pc it does a line break instead, so I have to click the blue search button

[fluid mauve]

Afternoon. is thee a guide or something to assist me in generating a decent image in #dream. i end up with some images which are decentish but there is always an eye somewhere wrong or a missing arm

[granite quartz]

Better value, if using gpt-4 at $20, has option to use Claude 2 also. Various modes, perplexity API, mobile app, sources not as distracting

[shadow stump]

Thanks

[granite quartz]

You're welcome

[minor charm]

hi

[turbid iris]

I see that you improved Perplexity's "copilot".... Now can it be 100% GPT-4? Or is it still a mix between GPT-3.5 and GPT-4?

[round sentinel]

Hello everyone!

I'm going to hold an online session as a student researcher in cooperation with some German universities to present perplexity.ai as a powerful research and content generation tool.
The audience will be German professors, teachers, students and other people mainly in the educational field.
I was wondering if there is someone from the perplexity team who could have a chat or call with me so that I can outline all the benefits perplexity has as a tool and explain all the features technically correct.
Dont get me wrong, I work with Perplexity every day, I know the benefits and love it, but I thought you might have some time for me!

Its all research related and nothing economic about it, if thats important to you!

[thorn moss]

Die sind alle bescheaftigt leider und selten bekommt ein Ruckmeldung Freund

Trotz direkt Nachricht

Guten Abend ubrigends

[round sentinel]

Guten Abend! schade, aber verständlich, das Team wird viel zu tun zu haben

[thorn moss]

Ja, heutzutage bekommt nicht zu ausreichend info, aber probieren vielleicht jemand hat zeit oder bock.

Fruher Team war aktiv mit eine Fragen zu antworten, aber merkt man das nicht mehr hier.

[noble stirrup]

Is it only me or did perplexity ai get nerfed lol

[thorn moss]

Not for me.

[muted pine]

@kind vault hi, When dall-e3 HD ?

[cloud sequoia]

@kind vault what bug did perplexity experience ? do you have any root cause report?

[tame current]

unsure why i cant get pro role but theres a major issue with wolfram mode as it cuts the end off of the equation

[frosty galleon]

Hey guys, the PT chat has made a bomb update now, huh, are you perplexed, will you be able to follow it? I'm hoping so, huh?

[ionic rock]

wtf guys, since yesterday in 50% there is just no answer coming or this...

[sonic vapor]

OpenAI APIs is outages currently

They outage multiple time since yesterday on all of their services.

[tame current]

never had a problem tho

not for the last couple days

[sonic vapor]

I subscribed to their server status. It has sent me multiple emails since yesterday.

https://status.openai.com/

Look like not all just some location is outstage

[sour hill]

Hi there. My use case is around asssisting neurodivergent study partuclary where synthesis is not an issue but remebering where something was read is impossible, This relies on effective sssummarisatin of text/pdf/or web documents. What I am finding with most models (particluarly ssice june) is that thewy will describe an inmdex of the ddocument rther than give you the cliff notes (tells you whats in ity rather than showing you). eg "In this document the authors decribve how to make a bunch of tea and the factors that are important in making a good cupo of tea" rather than - "the autrs suggest the best cup of tera is made bny adding a heaped teaspoon of any black tea to a tea ball and letgting it steep for 4 minutes as sany longer will prodduce a strugent buitter tea and less a weak tera for most tastes". I have tried many different forms of promt eg summarise the methods and results or show me don't tell me (which I know breaks the rule on negative prompting) , but to no avaail. I find that even if I get it answering correctly it wil very quickly revert to describing a text rather then a true summery. THi sis vexing me greatly 😮 and it is across many different products so I am gussing itsa as a resault of changes to back end api's/agents

[bronze pine]

Hi, I am a Generative AI LLM GPT-4 LangChain Prompt ML engineer

[tame current]

@tame current sorry to ping but can you help, i dont want these issues to dissapear in chat

[sour hill]

did anyone else get access to a new front end for GPT4 for a short while only to have revert to the standard one a little while later? (UPDATE - looks like its universal, chopping back and forth beytween the two and /create becoming unavailable). Which of cousre leads to the question, when the API's behind all this stabilisse how long before we see that capability here 🙂

[tribal flint]

Can anyone provide me a referral ?

[sweet jasper]

Just tell it to continue the equation

[halcyon coral]

Hey, please try "k^4-36" 🙂

[velvet anvil]

Here you go!
https://perplexity.ai/pro?referral_code=STT4NMBG

[drifting shadow]

https://perplexity.ai/pro?referral_code=PDXJNSVM You will have to create a new account in incognito.

[cloud sequoia]

replit-code-v1.5-3b error invoke.

[digital siren]

Why doesn’t anyone ever mention Ai21 labs? https://www.ai21.com/blog/introducing-j2

[cloud sequoia]

what advance do they have?

[digital siren]

They have a 178 billion parameter large language model. Supposedly it evaluates by humans higher than ChatGPT. I haven’t seen any of the ChatBot leaderboard type scores, but it’s still a massive model. I got access to it in about 1 minute and for free too.

I’m excited to find another serious contender in the llm space and it’s ice that it’s not an American big tech company.

[thorn moss]

I am more excited first on the upcoming new implementation. Zeym they are too silent.

Even @halcyon coral their representative cant say a thing. This upcoming new implementation too confidential

I think im gonna pre ejac when there will be an announcement

[halcyon coral]

I am not an official representative of the Perplexity tram, I am only server staff and take care of the users in case of questions and problems 🙂

But I can assure you that the team is working on implementing new features from OAI and there will be an announcement soon.

[thorn moss]

sorry for my assumption bro. hehe. oh my how i wish we knew how soon it is today or tomorrow or several days. I am still sticking on what Sir denis told last week that it will be early next week. fingers crossed

like what i have said, i think im gonna pre ejac once announcement is gonna take place.

[halcyon coral]

Please be patient, I don't think there's any point in adding a new feature which is then of poor quality, I'd rather wait a few days longer 😉

[thorn moss]

yeah. hehe. sorry for being impatient. i stayed until 1am awake here in germany waiting for it. haha

everyday now. but i should stop.

patience....

[zenith atlas]

even openAI are still not pushed their new version of chatgpt to all plus subscribers, and people who got new version of UI and gpt turbo are saying it have a lot of bugs and problems, so its better to get nice features in a week or two then pure frustration and dissapointment today

[thorn moss]

Its ok to release it i think but let us have the choice to switch then they can improve it for the next coming weeks. let the people tell what is the problem i think.

[tame current]

mmmmmmmm

[sweet jasper]

Does it have function calling? What is task specific apis

[feral saddle]

I am new and I want to know about the nfts and how to create them and which platform is best

[muted pine]

OpenAI totaly destroy DALL E3 on CHATGPT now we can only generate 1 image ...

Somone here try the API it's good ?

[regal frigate]

i want gpt4 128k

[thorn moss]

No ETA. just soon

[boreal pasture]

well, perplexity is dead now

[arctic pawn]

Hi, good morning everyone!

[calm marten]

yea ikr really weird shit

[upper island]

Will we get gpt 4,5 on perplexity?

[thorn moss]

They are probably thinking about GPT builder or so called GPT's. But it is really kinda weird they are so silent now. A little communication be better.

Too many changes

[thorn moss]

is there GPT 4.5? that was fast. I thought GPT 4 Turbo only. since when?

[upper island]

Same thing 4.5 is 4 turbo

[thorn moss]

oh? Good to know. Cause you are the first one to tell 4.5 and 4 turbo is the same. now i know

[tame current]

Hows the pricing

[foggy swallow]

They are not though, performance is at best marginally better, even a little worse at some preliminary benchmarks. It's not half a version jump, even Sam Altman said it's not GPT4.5. It's just faster and bigger context length.

[sweet jasper]

https://aider.chat/docs/benchmarks.html

This is still pretty big, the fact that if you keep your prompt less than <73K , gpt4 will have good to perfect recall regardless where your documents are

So 64k perfect recall vs 100k from claude which is using RoPE optimizations so larger prompts seem to get only a high level summarization

[coarse venture]

Is perplexity still better then open ai chatgpt plus?

For like web search

Will the price go down with API cost change

[tame current]

J

[lament imp]

This is probably relative to whatever you are searching. The model interprets what a web page says, but bases the web pages on what is searched. What is searched, is based on your query, which can be interpreted in several ways. So realistically, it's variable on many different factors, as well as your own personal opinion about whether or not what was searched was accurate.

Doubt the prices will go down, however.

I don't know of any monthly subscription service that has lowered their prices after it was set.

[sweet jasper]

It all depends on how they implement the retrieval models for search

Which we don't know, but perplexity search is pretty good

However follow up questions are hard to do with search because the search takes over if your prompt isnt specific enough

[tropic crest]

for web search absolutely yes

[coarse venture]

Ok thanks all

[tame current]

Hello, I am very angry and I don't understand why the developers and community managers are turning a deaf ear and remaining silent. I have been with Perplexity for a few days with an annual subscription (which I regret). What are the developers doing? It's been a week since ChatGPT 4 Turbo - 128k was available. Claude-2 is really bad.

And the worst part is that there is no announcement about progress. It's unprofessional, guys!

Give some news to the people who are paying, for heaven's sake!

[thorn moss]

Someone cant take it anymore. Silence means something indeed but i dont want to throw any conclusions.

[thorn moss]

Yeah a little communication will do. I agree.

[short gale]

Just FYI the 128k GPT-4 Turbo model is only available as a preview currently & heavily rate limited so it’s not ready for production-level traffic yet

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

Per OpenAI’s blog post they “plan to release the stable production-ready model in the coming weeks.”

Also if you’re subscribed you should also voice your concerns in the private pro channels—I think the discord invite link should be in your profile settings

But anyway yeah it’s been quieter the past few weeks, last post in #📢│announcements was October 6th

[digital siren]

GPT-4 32k is amazing and Claude 2 is also amazing. Claude definitely has its own uses.

[digital siren]

Absolutely better for web search on perplexity

[cloud sequoia]

I

I cant use "📷 Image Search
Your searches just got a whole lot more visual. When you search for people, places, or things, we will now include images to give you a better understanding of what you're looking for.

Try it! ► https://pplx.ai/photos-of-sf
Perplexity 🧵 的最新改进：

📷 图像搜索
您的搜索变得更加直观。当您搜索人物、地点或事物时，我们现在将包含图像，以便您更好地了解您要查找的内容。

试试吧！► https://pplx.ai/photos-of-sf"

Is it a bug ?

https://pplx.ai/photos-of-sf can not be access.

[tame current]

Hi, what is the difference between Perplexity Pro and Chat GPT Plus?

[chilly forge]

Create a #1140622008086970420

[hollow wadi]

What is copilot really used for?

[zenith atlas]

he can help clarify your prompt and can search for more sources then regular non copilot model

[muted pine]

Woah … Poe just drop Dall-E3

When in Perplexity …

[thorn moss]

No ETA, just soon.

[sage compass]

is there a way to customize the ai to behave like u want?

[halcyon coral]

You can use the AI Profile for this

[sage compass]

are u able to show me

[halcyon coral]

https://blog.perplexity.ai/faq/ai-profile

[sage compass]

ty

can i somehow upload png

so it can analysis the pitchure

[halcyon coral]

No, this is not (yet) possible

[waxen shard]

hola

squiusmi

hello😀

[carmine linden]

Hey @blissful brook! For AI art you can use the ⁠#🎨│dream channel and generate it with the /dream command

[trim kiln]

has the perplexity team ever shared the serving framework used for labs? it is perhaps the fastest llm inference I've used

[short gale]

It’s their pplx-api https://blog.perplexity.ai/blog/introducing-pplx-api

Blazing Fast Inference
Perplexity’s LLM API is carefully designed and optimized for fast inference. To achieve this, we built a proprietary LLM inference infrastructure around NVIDIA’s TensortRT-LLM that is served on A100 GPUs provided by AWS. Learn more in the Overview of pplx-api Infrastructure section. As a result, pplx-api is one of the fastest Llama and Mistral APIs commercially available.

[trim kiln]

ah, I can't read. thx for the link. looks like they're using trt-llm

[short gale]

Aye TRT-LLM for software & A100 GPUs for hardware

[neat elk]

we have many optimizations on top of trt llm

btw, try out our online llm (pplx-7b-online) on labs.pplx.ai and let us know what you think

[woeful ember]

Hi I’m a user and a fan of Perplexity.ai 🙂

[halcyon coral]

Hey @blissful brook,
please share your project in #😎│cool-projects.
Thank you!

[wicked blade]

When will Perplexity finally include voice and image recognition?

It's already available on platforms like Bing, but nothing on Perplexity.

😅

[thorn moss]

gonna wait for weeks or maybe a month bro. there is no definite time, even if the models are released GA

to be sure maybe a month.

[short gale]

Yeah it’s difficult to determine a reliable ETA because like paul just mentioned there’s no telling how quickly OpenAI will be able to onboard enterprise API customers even after the new models are out of preview which all we know will be “in the coming weeks”

Infrastructure problems could easily cause delays; also the holiday season is coming up which means lots of OpenAI employees away on vacation

[civic socket]

Has anyone else noticed that Perplexity has, for the last week, become extremely vague and generally "stupid" to the point of usless for most answers ?

It's absolutely definitely become a different "personality" all of a sudden, no longer being helpful for most things (I typically ask tech questions, like protocol port numbers, programming language info stuff, and so on - so it's blatantly obvious that something big has changed under-the-hood, in a bad way)

[meager sparrow]

Lmao me too man I would ask a question and perplexity would give me a vague answer and then I would ask another follow up question and it would literally repeat itself again…

They have also been so many times when I ask questions about applications to places and perplexity would literally contradict itself and tell me to click on options and things and don’t even exist on the fucking page LOL

[long summit]

Any coupon codes? :P

[cloud sequoia]

is it a bug ? https://www.perplexity.ai/search/ACML-2023-FPWDydNXR0GYHS4Umy.EBQ?s=c

[cloud sequoia]

https://perplexity.ai/pro?referral_code=SC4NT94J

haha , here you are

[signal hamlet]

Hi @hollow snow! We have the #😎│cool-projects channel to share your projects. Please post it there.

[muted pine]

hi no code for Black Friday ?

[lament imp]

Yes, this occurs occasionally.

[hollow snow]

Ah, thanks! Didn't see that in the sidebar.

[regal frigate]

when gpt4 128k i can use here

[muted pine]

1-2 months

[raven bloom]

Hi i have a question, i am building a little bot for a server, can i add the api from perplexity rather than my GPT for the bot?

[thorn moss]

I agree with this one. At first we did not expect that. 😂😂😂

[muted pine]

😂

[thorn moss]

I really thought it will really be released few days after announcement but nah. Too much with the hype thing.

[muted pine]

my opinion it's clear, the dev work for V2.0 of Perplexity with Dall-E3, Whisper and GPT-4-Vision and for sure 128K tokken for GPT4

My only fear is that developers will abandon Perplexity due to the competition from ChatGPT, except that I find ChatGPT works less well for getting information on the internet than Perplexity. Or that another app buys Perplexity, or recruits the developers to work on other projects, because they have been very quiet since the OpenAI conference...

[thorn moss]

That are my thoughts also. information from perplexity that is why i cannot stop subscribing to it.

I can really feel the difference of information of perplexity versus chatgpt for real

That sets apart perplexity from chatgpt and chatgpt has also limitations

Performance wise, perplexity for me is also faster than chatgt. Perplexity team knows i think their edge and what set them apart from chatgpt. Man they are highly intelligent people that work for something i guess they wont abandon it.

[muted pine]

For me, perplexity is bad for answering questions during a discussion; it responds well to the first question but when another question is asked, it has no memory.

Whereas CHATGPT has a memory.

But for the rest, Perplexity is better.

[thorn moss]

Well that can be improved. For me censorship from chatgpt really sucks

[muted pine]

Just wait GROK

from xAI

[thorn moss]

Perplexity is very good for information

Yeah heard it.

[muted pine]

but the bad point for PERPLEXITY it's limited by API

OpenAI release on CHATGPT before API

[thorn moss]

Sad. Let us hope after a month they will hit us hard. Zeym

[muted pine]

yep ! And let's see GROK too !

[strange kelp]

Because Perplexity is not a chatbot, it is a search engine. So, just like Google, it can suggest you some other related questions but doesn’t remember old queries. It searches new sources for every query.

Don’t expect it to be a chatbot. It is good as search engine.

[pure abyss]

looking forward to the Vision version, too！

What I hope is that no need more than 1 paid-model---Just 1, the best one is enough, don't let users to do the choice

[sweet jasper]

Are you sure?

Qcrually you might be right

Perplexity 7b online is cool , it was able to answer who was parthenopaeus compared to the other llms. Only gpt4 and gpt 3.5 managed to get it right without internet

[cloud sequoia]

https://www.perplexity.ai/search/Llama-2-Open-5sxyhhNkTE6cYRoUrg5c6Q?s=c

seems a bug？

[wraith crow]

That does indeed look like a bug. You can report the wrong answer the bottom of the thread by clicking on the flag icon. This will help the AI team to improve the output.

[tepid mural]

When can we generate images?

[wraith crow]

You can generate images in the #🎨│dream channel.

[old forum]

⏯️

[tame current]

me just finding out api keys exist for perplexity 💀

no gpt 4

guh

[tepid mural]

Ahh, true. Gotta check it out.

[sweet jasper]

Whats the context size now

[hallow linden]

Grok is going to be a total, abject failure. It won’t be as smart as GPT, it won’t be able to reference information as well as perplexity. It will also make the cringiest, dumbest humor because it’s designed to appeal to a man who doesn’t know how “that’s what she said” jokes work

[granite quartz]

Any ideas what Dennis in the announcement means by - also we made some changes to make our answers to be more concise, but not more work is ahead hear.

"not more work is ahead hear"

@neat elk

Sounds great either way!

[neat elk]

lol sorry, a typo

should be "but more work is ahead"

[hallow niche]

You seem a little too invested in Grok failing my guy

[granite quartz]

Ah great thanks for confirming 🙂

[neat elk]

thanks for spotting it 🙏

[hallow linden]

I mean if twitter gets a built in LLM, that’s great. But it’s being positioned as a competitor to GPT-4, and I think anybody who thinks that a company with the resources of X can build an LLM that will be as smart as GPT-4 based off of tweets is completely out of touch with makes a good model.

[hallow niche]

Well said but I don’t think that means Grok is guaranteed to be a failure. Say what you’d like about Elon but he makes business successful and it’s hard to predict how good a LLM will be that was announced publicly a week ago. Most people never knew openAI would blow up until they did

[zenith atlas]

LLM based on twitter dataset sounds equall to a LLM based on 4chan's /pol/

[hallow niche]

I think it’s more about access to real time data than access to people’s tweets lol

[sweet jasper]

@neat elk is it possible to know the increased sizes in the announcement 🤔

[tough turret]

Hello, just a snooze as I have a couple of use cases on hold for this. How's the roadmap for having internet grounded search on the API? Is it still on the end of November?

[sweet jasper]

I saw pplx 7b and 70b on api I think with search

Wait no its default without search, only labs seems to have it rn

[dire sapphire]

Is the announcement for the api or just perplexity ?

[granite quartz]

Does the increased context apply to file uploads also?

Also wondering what the context size is/was for search?

[digital siren]

Instead of hating on Elon and Grok, wait to try it and see what you think then. More models are great for us.

[tough turret]

can I apply to run the Lab version? It would be really cool for my use case....

[sweet jasper]

Yeah, does search apply also to writing modes?

Or is it tied solely to the focus all

[signal hamlet]

Hey @sweet jasper! Writing mode enables you to have a conversation and perform tasks with the help of AI assistant without accessing the internet. You can read more about the search focus here: https://blog.perplexity.ai/faq/what-is-search-focus

[serene tide]

What can I do with Perplexity that I cannot do using ChatGPT-4 Browsing?

[tough turret]

Thats a deep question. Are you looking for a response in the perspective of and end user or an application?

I think they have totally different outputs. ChatGPT doesent for for the web unless you (1) explicitly tell it to do (2) phrase your question in a way that he will decide that it is better to answer using the internet.

As for APIs, none of them have native support for internet search. I manage to particulary solve that by integrating intent recognition and SerpAPI into my bots.

[serene tide]

I don't think its that deep. It seems there is nothing Perplexity can do that ChatGPT-4 with its new browsing cannot. Sure you may need to ask it to use the internet but so what?

and perplexity uses GPT-4's API so I'm trying to understand why I'd want to go outside my usual ChatGPT-4 to use GPT-4 lol

[gusty jolt]

i have both ftw, quickly use pplx to research while gpt-4 slowly generates a long winded response

[serene tide]

so speed is the difference? i can just put my Custom INstruction to be more concise?

[gusty jolt]

how many pages does chatgpt browse actually hit? last i checked it was mostly concealed. pplx can hit dozens

[serene tide]

it does like 3-8

[gusty jolt]

and last i checked it still doesnt have any citations

[serene tide]

so it hits more pages?

it does

[gusty jolt]

oh it does now?

[serene tide]

when i use it yes

so the only difference is number of pages it reads? that's just a prompt difference no?

when i use bing to search instead of google i want to kill myself because its much worse. I dont get that feeling with perplexity - its aboutt the same as my ordinary work flow with ChatGPT4

[gusty jolt]

i hit my chatgpt 50 query limit all the time. so maybe (having pplx) is more beneficial for power users

[serene tide]

i abuse ChatGPT-4 and i never ran into any ratelimiting lol

maybe im glitched

[gusty jolt]

if you see no benefit then dont invest.

[serene tide]

my objective with this question is to figure out if im wrong about my view

a strong example would convince me

[gusty jolt]

everyones personal circumstances and use cases are different, there is no single truth, there is only the truth you make for yourself

[serene tide]

i disagree. Google is better than Bing for search FULL STOP lol

and Apple is better than NOkia no matter what lol

but idk - let me know if perplexity does something that no one else can do

[gusty jolt]

i stopped using google search long ago. its no longer a search engine, its an ad engine

[ancient grail]

I agree that the ads across the top are annoying, you sometimes have to scroll down a few hits to find what you're really looking for

[gusty jolt]

even the results you're looking for are just another layer of ads

[ancient grail]

I mean SEO is going to happen for any search engine

You can always game the algorithm

[gusty jolt]

we dont have to game any systems now that we have a proper answer engine ie perplexity.ai

[ancient grail]

You don't think ppl will start to break AIs too?

[gusty jolt]

maybe. but until then, its a solid choice

google used to be awesome. long long ago

[ancient grail]

I do use perplexity as my main search engine now, but still need google to look things up quickly

Like for example a store or map location or when I basically just want it to autocomplete a url for me

[gusty jolt]

well yeah if you want help buying something, google knows what you want and can help you get it

[carmine linden]

Hey @tiny sonnet! You can use the bot in the #🔎│ask channel. Tag @ PerplexityAI and ask your questions.

[ancient grail]

this is what i do too

it's also nice to just get two interpretations of the question

[gusty jolt]

yeah and you get a chance to curate what you copy&paste back into the chat

instead of losing your context with irrelevant junk

[ancient grail]

wait when do you lose your context?

[gusty jolt]

i mean like, if you ask chatgpt to research stuff, in doing so it can find irrelevant info, which requires rolling the older content out of the context window

keeping the context window on point and lean is counter to throwing out a wide net for research

[hallow linden]

There's doing something no one else can do and doing something better than anyone else can do. On paper, chatGPT has web search. But perplexity is simply better in terms of speed and number of sources returned, which can be 20 with copilot. Not to mention that with perplexity pro, you can perform filtered searches, like through reddit, youtube, and academic papers.

Perplexity is optimized to be an AI-augmented search engine first, but is also a pretty capable chatbot, especially in writing mode. ChatGPT is a chatbot first and is a decent but slow AI-augmented search engine. Your value varies based on your use case

[tame current]

gpt4 mode is like more of a chat bot than anything, reminds me of autogpt

[chilly forge]

Its changed to 100/3hrs

[gusty jolt]

nice

[granite ocean]

Hi

[gusty jolt]

i just got locked out of gpt4 for 13.5 hours.... 'To give every Plus user a chance to try the model, we’re currently dynamically adjusting usage caps for GPT-4 as we learn more about demand and system performance.'

i didnt even get 50 prompts let alone 100

ohwait is just GPTs

[hallow linden]

That’s another plus for perplexity. OpenAI seems to prioritize API access over ChatGPT plus access during outages

[chilly forge]

Thats sad !

[sweet jasper]

You cant subscribe to chatgpt plus for now lmao

[sacred trellis]

so many people from devday 😵‍💫

[safe violet]

[sweet jasper]

How to join the pro discord? I currently have a sub

[zenith atlas]

There is a link for pro users in settings on perplexity, you need to leave the server and join with that link

[sweet jasper]

It didnt work but i will try again

It guides me through the tutorial once more without adding me to the channel

[wraith crow]

#👑│pro-lounge should be available for you now

[fiery dust]

Hello, im part of the replit core partnership that got announced yesterday. I haven't gotten an email to redeem my free months of perplexity pro.

[signal hamlet]

Hey @fiery dust! Those emails should be sent by repl.it. Please contact their support, if you haven't received the updates.

[fiery dust]

Ah that makes sense! Thanks for the quick response, Alex!

[tough turret]

Hi, i am only reposting this because I felt that my last question was not answered:

• how's the roadmap for adding up to 200 pdfs to a private embedding using Perplexity API? Where can I keep informed without having to ask here?

• how's the roadmap for internet grounded search via API?

• when can we expect to have web quality results using the API?

Thanks!

[blissful raptor]

"how's the roadmap for adding up to 200 pdfs to a private embedding using Perplexity API? Where can I keep informed without having to ask here?"
Not on the roadmap yet.
"how's the roadmap for internet grounded search via API?"
coming pretty soon. meanwhile, play with the pplx-7b-online and pplx-70b-online on labs.perplexity.ai
"when can we expect to have web quality results using the API?"
not sure if I understand the question. do you mean when will API match the product in quality?

[topaz chasm]

Hello, I have a pro account, but when I ask a question, the system is Hello, I have a pro account, but when I ask a question, the system tells me using gpt 3.5 why?

[zenith atlas]

you can change the model in the settings

[topaz chasm]

I do but same

[signal hamlet]

Hey @topaz chasm! You are using the GPT-4 model, as set in your settings. You can also see a small label with the model name under each answer.

[signal hamlet]

Hey @meager sparrow! Could you file a bug in the #1140622008086970420 with some examples. That would be helpful, thank you!

[topaz chasm]

Seems like not gpt 4, Because the application gives the answer gpt 3 to the answer of which model are you using

[signal hamlet]

https://discord.com/channels/1047197230748151888/1173623903596326912

[stone sedge]

v

[viscid star]

Wow so I just noticed a pplx-70b-online model

This is everything I could have wanted 😄

Was this announced/did I miss something?

[gusty jolt]

via the API or just in labs?

[viscid star]

labs

[gusty jolt]

i dont see the citations wah

[cloud sequoia]

coming pretty soon. meanwhile, play with the pplx-7b-online and pplx-70b-online on labs.perplexity.ai
--- so cool！

when will pplx-7b-online open From API ?

[hollow obsidian]

Lol....
I know from long long ago AI Roleplay in early 2023, that the AI fooled me with "give me a moment, I will create this" or whatever and I waited and waited xD Until my brain realized AI can create things in seconds.

So... the AI did fool me and have no access? 🙂

yepp.... fooling me xD

I did edit my reply to "please give me now a comparing" and it worked.... this is weird O_o it was too much in "roleplay" mode 😄

[tame current]

perplexity be broken

wont let me login

:\

ive tried like 4 dif emails and it never sends

[zinc tree]

me too. even if perplexity pro.

[stone sedge]

It says I'm not even pro anymore lol

[distant viper]

So the page is broken?

[gaunt rose]

perplexity down?

[distant viper]

I tried to log in but it all failed

[distant viper]

Maybe. I'm having it read books for me, which kind of sucks.

[prisma marten]

Never seen this before. Is the service down?

[signal hamlet]

Should be back up again 🙏
#🔔│status message

[dreamy plaza]

It’s still down for me

[frank rivet]

It might take a little bit more time to go back to normal, you can also refresh by doing either of these:

1. Clear browsing data, including cookies and cached images and files
2. Restart browser

[prisma marten]

It’s back up for me thank you!

[dreamy plaza]

It’s working for me now, thanks

[turbid hound]

Hey loving perplexity pro at the moment!

How do I know if perplexity is reading off a file and not hallucinating under the "Writing" mode. I've uploaded a book as a pdf and it is able to answer me questions. However, when I ask for specific quotes from the book, it tells me I'm sorry, but I couldn't find any notable quotes from the file .

Now I feel sus of its answers. How does it work? Is it answering based on the data is trained or is it actually coming from the file? Is there a prompt that acts as a litmus test to determine if it is actually reading off the file or just giving answers from the model itself?

[distant viper]

In resources if I add a file and I add the web resource again it disappears, is this normal? Is there any way to add two resources at the same time?

I added the file as a translation rule, but missing web resources many proper nouns and ambiguous words it translates incorrectly.

[torn zealot]

If I pay for the yearly subscription, is the 600 GPT4 daily quota locked in? Or is it still subjected to change.

[boreal pumice]

Hi everyone! I'm a librarian and I am learning about AI tools to help our students in navigating and vise usage. I saw that perplexity is using academic sources, which is great, but a lot of academic databases such as Science Direct, Proquest, etc. are locked behind paywall. So, how can we rely on the answer, if most of the sources are locked?

[velvet anvil]

Not sure there is an easy way unless most authors start publishing their papers in easily accessible ways or if perplexity integrates with the big publishers.
I shudder to think how much IEEE, Elsevier etc would charge them for this.

You can also reach out to authors and ask for their paper. They are allowed to share in this manner and then maybe you can use the file in your queries as an attachment?

[boreal pumice]

Yes, me and my colleagues brainstormed a bit about the issue, and we came to conclusion that the query will not be precise. Which means that this tool can't be used fully in academic research....

[velvet anvil]

I think summary of all papers is still openly available for indexing. So it might be able to point to the relevant papers but if the content is behind a paywall, then deeper queries won't work.

[boreal pumice]

Yes, this is the case. Because I'm looking for a way to ease and guide our students through AI swarms.

[velvet anvil]

Good luck! If you find something that works for research consistently, would be interested to know.

[flat agate]

Hi, I studied librarianship, information sciences and archives 23 years back on Spain but never worked as librarian, I worked on the "dark side" near 10 years developing a library management software... I was thinking about your question, I thought about google scholar but I thing It would be not valid too... I'm very very rusty on librarianship but I would love to know if you found any possible solution 🙂

[meager sparrow]

Cool your point is????

[meager sparrow]

23 years is a long time

I can’t imagine studying anything for that long

I will get bored asf

How do people study things for so long without getting bored??

[turbid hound]

It seems like it’s summarizing concepts from the abstract. So it could, at most, point you to resources that may contain what you want. You’ll still have to dig the papers deeper. However, I found that most replies are pretty generic and they don’t have the details I want. I say this even with perplexity pro copilot, and GPT4

[flat agate]

Sorry, english is not my first language, I meant I finished my bachelor's degree 23 years back! I studied librarianship out of passion but unfortunely life made me a software developer. I find a great research field using AI on libreries so I'm curious about possible solutions 🙂

[vital goblet]

You might try elicit... AI based scientific literature search - shows abstracts of the abstracts and variables used\researched. U will find it here: https://elicit.com/

[turbid hound]

Yea I did but not very in depth. their credit scheme makes the pricing and predictability of costs hard. I like the per use type of billing like perplexity

[waxen narwhal]

does anyone actually use discover?

[narrow silo]

@neat elk how to train my dragon?

I mean AI

[vagrant river]

This is $20 per month, and the difference in functionality with ChatGPT is too far apart.

[viral lantern]

Hey guys 👋
Excuse me if this has been asked before or is not the right place to ask...

So me and a couple colleagues working in the health sphere have been thinking about joining our digitalizated repositories of information together to make a searchable and chattable bot to organize some of our acquired knowledge and make it easily accessible. I've been thinking about this ever since chatgpt first came out.
However, so far i have not seen any publicly available technologies where this is possible on a large scale ( like several Gigabytes of files atleast if not more)
Except for training a LLM yourself running it locally on your computer, which would probably not be a viable option.

Any thoughts? Would be very much appreciated 👍

Cheers guys and i wish you a great afternoon from Germany 🙂

[sweet jasper]

Hey man, thats actually one usecase that rag fits well

You will need a vector database like chromadb, and to split your documents into manageable chunks(langchain document splitters work fine here). To have the llm work then, you just have the user query the vector database, then inject the vector database result ALONG with the user original prompt into an LLM.

This might seem overcomplicated but it's actually quite simple, with gpt128k you get up to 64k context with perfect recall, which amounts to alot of words, about 50 pages of context.

Vector databases are not as intensive as llms and is how perplexity and openai have implemented external knowledge outside the model initial training.

In case of perplexity , its getting search results injected directly into the prompt but the concept of vector databases work better for your use case.

[boreal pumice]

Yes, pretty much this. However, my goal is to collect and educate students about AI as useful, yet, not-so-perfect tool

Oh, and one more thing: there are 3 things that students will look in AI tools: picking and creating bibliography, looking for relevant papers and write the assignments for them LMAO.

[flat agate]

I still have nightmares with the bibliography section of my assignments and the different styles of citation, on that time handmade on word 😱

[halcyon coral]

Perplexity is not a platform to "train" anything, but you have the possibility to customize your AI profile.

[frank rivet]

These mental maps are a great way of visualizing the systems and processes for the technology behind Perplexity on a general level; thanks for sharing. Did you make them yourself?

[viral lantern]

Thank you so much for the direction. This is huge. Well you seem way more informed than me, do you think the current state of Rag's would be sufficient for our little project? It's not supposed to be that expensive, i suppose our group of colleagues has no rush and could wait for the technology to catch up .

One thing I wasn't sure about:
How would the Rag go about understanding tables of data? I guess one could screenshot tables and maybe get a visual gpt to convert it into raw text perhaps? Main point is would it be able to use all the data within PDFs and maybe certain websites or chatlogs.

do u think the tech is mature enough yet?
Seems like the gpt's may be a kind of stepping stone to what I've been imagining for our little project.

[sweet jasper]

Sadly no, but it's a pretty good high level overview, often times its easy to get lost in architectural components

[sweet jasper]

RAG will be cheaper than finetuning an existing, as for the pdf tables, you can probably parse a pdf and do some Cleanup on the documents via automation. Look into OCR, You could have vision into part of architecture, although gpt4 rn is quite limited due to devday.

Gpt4V is very impressive, it is able to understand nuance within it, so I think on the vision regard its mature, in context of Rag, your mileage will vary on what model you use.

Your main issue for accuracy is going to be how you handle context windows (how much the llm can remember), your documents will consume a good chunk of it.

If you if you are injecting 1 to many documents in a prompt, local llms will be very limited as they typically only support up to 4k tokens, unless you are using them to clean up prompts then doing the big lifting with bigger models like claude and gpt4, which support up to 100k and 128k tokens of context length respectively.

@viral lantern

The hardest part in your case might be how you manage those pdfs 😛

You can also save images in a vector database by converting the image into base64 as metadata (this complicates the pipeline, so might not be the best approach)

[solid creek]

is it worth buying perplexity pro over a GPT-4 Subscription?

[stark lichen]

if you ask me. heck no. gpt-4 sub is powerful

this is just for fun

[solid creek]

oh ok

[gusty jolt]

depends on your use case. if you want 'unlimited' api calls for a subscription cost (during the pplx-api beta rollout), and dont need gpt4, then perplexity is the better deal

[stark lichen]

what this is nice for is to quickly lookup something simple and give you up to date info.

[gusty jolt]

yeah the search UI is also better than chatgpt web browsing

is also nice having all search chats separate to general chatgpt chats

[stark lichen]

yeah that is THE only thing this thing does so it better be better then chatgpt or what is the point 👨‍🏫

[tame current]

https://youtu.be/UHKoF6sl7Wg?si=3EWZS_S2LrC6EcMV

[tough turret]

"when can we expect to have web quality results using the API?"
not sure if I understand the question. do you mean when will API match the product in quality?

Exactely @blissful raptor

[stark lichen]

afaik there is no api this makes preplexity a lot less useful

all we get is a plugin that gives us some half baked parsed content that is sometimes useful

[viral lantern]

Very interesting. I saw this github page that already let's u "scrape" websites only using their urls and converting those into knowledge files for the custom "gpt"s by openai.
I'm not too familiar with the file size difficulties, there's probably still a couple months/years until you can make gpt's that allow a file size of say 50gb...
Overall that definitely sounds preferable to running LLms locally, since then me and my colleagues would have to worry about hardware requirements. And probably still have to pay for api usage and whatnot...

What do u estimate? Will something like gpts be available for an affordable cost at the file sizes I'm thinking of?

[tough turret]

Thanks Aravind, i used on the lab.perplexity and it looks pretty great. It doesnt like it is grounding thought.

[tough turret]

While I wait for the Internet grounded search on the API,
I am considering using Tavily https://app.tavily.com/chat Have you guys tried it?

Any caveats?

[turbid pebble]

Hi all, I recently wrote a blog post about Perplexity, which you can check out at: https://open.substack.com/pub/newroad/p/perplexity-is-a-next-level-generative?r=1uszy4&utm_campaign=post&utm_medium=web

Please feel free to share it across your socials! Cheers

[stark lichen]

nice article 🙂 the co-pilot is unique to preplexity and sometimes it works like magic

[wraith crow]

Great article, you can also share it in the #😎│cool-projects channel

[short gale]

https://blog.perplexity.ai/blog/introducing-pplx-api It’s currently in closed public beta for pro subscribers

[stark lichen]

ohj man this is what i wanted

[neat elk]

you can also try all the available models in labs.pplx.ai

we added two online models there this week

[snow harness]

I wanted yo ask if anyone knows about if there is a pricing discrepancy specifically in Canada. I can get the pro for 20 usd online but it's 26.99 in the Canadian app store... and both don't include exchange rates for CAD. Or this is what it appears to be ... ??

[turbid hound]

not sure if you are a developer, but you can get someone (or you) to explore Llama Index. They have made great strides in creating the tools needed to this with relative ease. Their founder Jerry Liu is extremely active on twitter as well. Every tweet from his practically a banger

[frank rivet]

This is because of the currency exchange. The app store will display the exchanged total, whereas Stripe will not. It will only show the total in USD.

[thorn moss]

will the rewrite features ported to mobile soon or not yet planned? 😦

[short gale]

How do the online models differ from the chat ones?

Just wondering the use cases for each

[ionic rock]

In the "AI profile" I dont understand where the real system promt goes.

I want to add this: "Do not repeat information already provided in the same thread. Do not include common-sense warnings in your response."

[snow harness]

Thank you!

[lusty basin]

Your AI profile is a part of the system prompt so you should add those instructions there. You can’t modify the rest of the prompt

[ionic rock]

So "Intro" ?

[lusty basin]

That would work

[ionic rock]

lol bad prompt dont use it. it makes GPT even repeat itself within one answer 😂

[lusty basin]

Does the API support the new pplx online models?

[turbid hound]

i found that online is more politically correct, while the alpha is more uncensored. Not sure about performance wise

[merry aspen]

I have a code from Replit to a 4month of Perplexity. But I can't find where to enter it. My previous subscription is cancelled. Thank you

[sweet jasper]

I wasn't familiar with Llama index, it seems wonderful based on the short glimpse. Specially with the bazillion loader option s, thanks for the share @turbid hound

[sweet jasper]

Custom gpts once fully released might fit your use case, loading 50gb files directly seems very far away unless theres a major breakthrought and even nowadays its extremely unoptimised if you have to query that data without breaking it into chunks, remember, compute scales quadratically for context length, so to for gpt128k it takes 977 times more compute than a 4096k model does.

A 50gb file is going to have at least 2.5 billion tokens worth if a token is 20 bytes average, presuming it's only words would require 381 million times times more compute than gpt128k

@viral lantern

And you might not have perfect recall if you fit that all to memory with current architecture, you would need at least double the size of the llm, so 5 billion token context history

It would be like the scenario in hitchhikers guide to galaxy waiting for thr answer from the supercomputer

[cloud sequoia]

bug ？

pplx have bug？

[coarse river]

Im experiencing the same thing

[halcyon coral]

Please have a look at #🔔│status message 🙏

[turbid hound]

based on how proactive they are in the space, I am not surprised if they can become huggingface level cult favourite.

[short gale]

I think people not understanding these concepts is a lot of why OpenAI has been having such a headache with custom gpts & service downtime/degradation—the number of users trying to upload hundreds of documents most of which are full of dense/ unstructured text is huge. Expectations of the current capabilities of custom gpts are wildly outsized

[viral lantern]

Wait what?
So when it considers web searches or it's current use of knowledge databases, it's only at the scale of a few mb ?
I'm not too set on using a large model like gpt128k, although of course, ideally a better model would be superior to a more simplistic one.

So what's the current limit for knowledge databases to be considered by a decent llm, do you know?

[cloud sequoia]

Does the API support the new pplx online models released？

[sweet jasper]

Yeah, it's quite funny too, as there's no way the model will recall with accuracy the data that was inserted at the beginning, so the knowledge base will be filled with holes, ig there should be a limit for how many docs they allow

[sweet jasper]

To put it simple, perplexity with copilot is doing a very nuanced trick, it fetches 21 websites AND summarises it, a full html page typically is a few mbs, but it has a lot of tags that you need to remove that don't add anything to the context. The summarisation IS the trick that allows it to feed so many sources, because you don't blow up the context window super fast by putting the entire html in the prompt

Notice the copilot last step

For 5 resources and less it doesn't summarise it I believe

Gpt4.5 is cheaper than gpt 4 and has 128k context so for your use case it's even better

When it rolls out to perplexity, its probable the amount of sources fetched will increase as well

Assume that you can get perfect recall for your documents based on if your prompt fits 50% context window

That should be your benchmark

So gpt 4.5 supports 128k context window, your documents+user query should not be more than 64k.

https://media.discordapp.net/attachments/1047649527299055688/1172395780615712830/image0.jpg?ex=65696428&is=6556ef28&hm=3f0194d7e8d739d203a61685952145b980d86794c1cfaaabae8193db32ed3db9&

https://www.reddit.com/r/LocalLLaMA/comments/16cbimi/yet_another_rag_system_implementation_details_and/

Also another great read

[vagrant river]

In the experimental models of labs, none of them can answer this question.【Put 50g of sodium hydroxide into 500g of pure water. Assuming the initial temperature of the water is 0 degrees, how much will the temperature rise after adding the sodium hydroxide?】

[velvet anvil]

TIL that I have some things in common with open source LLMs

[velvet anvil]

On a serious note, you can probably raise this in #🦙│feedback-playground

[sweet jasper]

Is it 26 degrees

?

[jaunty grail]

Can I ask someone here?

AII = Search across the entire internet
Wolfram|Alpha = Computational knowledge engine
Academic = Search in published academic papers
Writing = Generate text or chat without searching the web

There are 3 options here, for those that are suitable for such as lab reports, papers, journals and others.

what's good? if you can please dong recommendations how

[vagrant river]

👍

[solemn onyx]

hi

i just purchased the perplexity pro

i want to do stuff with the api, but idk is the copilot thing supported there?

or maybe it's one of the other models?

[tame current]

brother i am waiting for an answer to the same question

[halcyon coral]

Hey, no Perplexity and the Perplexity API are two different things, you can find more information on how they work and the supported models here:
https://docs.perplexity.ai and
https://blog.perplexity.ai/blog/introducing-pplx-api

[tame current]

Not super intuitive. branding them identically with complete different use cases. Not ideal. You click get started on this landing page, theres really no easily accessible information that there is no API for perplexity.ai, and it lists out a bunch of model names, which for the average person, is a bunch of gibberish because theres no reference in to what any of those open source llms even do

well ux was certainly a final thought here

[meager sparrow]

What perplexity is doing is impressive

Ngl

The update is pretty cool

[sweet jasper]

Regarding the llm descriptions, it is a very valid point

Maybe we see some improvements on the side of the docs

[tame current]

Yeah, just as free feedback, for whatever its worth, think about your broadening user journey. I found Perplexity in a discussion Reddit, having never heard of it, thought it might be a cool solution to check out (automated research). I tested it in the UI, thinking there would be a reasonable way to use the same features via API, but its not the case. And when you google it, you sort of get this false sense of hope that seems accurate but isnt given the completely separate use cases. Also, getting into this you're starting to branch out from your hyper niched ai dev type enthusiast audience to your general science crowds, entrepreneurs, etc that arent going to know what a pplx-api is and would otherwise might believe that its just whatever the name is called for perplexity.ai API. It's just super confusing man and feels like a bait and switch from a new user standpoint, especially when there is a big bold heading on the sales page for API access, but only in fine print and only as an expert user would you really know the difference between "open source llm models" and just trying to get API access to what your doing on the UI in terms of running copilot which, again, is a completely separate audience from people trying to use open source llms. Like, why would I, as a person trying to automate research, want to buy an interface for open source language models (surprise, I dont). Anyway. I digress, take it for what its worth.

And when you click learn more underneath it, it dumps you into a FAQ of just unrelated junk and certainly doesnt help me understand that there is no API for the service, or even what open source language models are, and in fact, even a technical user but not neccessary dev type might even confuse this assuming that whatever tech powers the UI is just open source

[tame current]

The tool is actually awesome overall, but, what your selling, look at this page, i feel like im trying to buy the powerful ai research assistant, the value proposition, the API, but im not at all. It's...well. Could probably use a second look from a marketing team.

[gusty jolt]

if you want better search results, then dont use google

[strange kelp]

Academic is perfect for authentic resources when you are researching. However, you must check sources link if you want to cite them. I am doing Masters in Pediatric Dentistry and I use Academic Focus for research to prepare seminars and presentations.
Sometimes, when I open a PubMed link given in the sources, it says the page is no longer available, even when GPT-4 has clearly cite that particular source number in the output. (So, I cannot add them in citations)

[cloud sequoia]

how do you think about OpenAI

sam is out.

[cloud sequoia]

why Sam Altman been fired?

[formal kayak]

Discord question is not responding

@kind vault discord questions are not responding

[cloud sequoia]

#🔎│ask message

[cloud sequoia]

bug occur to discord bot.

[formal kayak]

When it will be fixed?

[formal kayak]

@frank rivet Any ETA ? When discord bot will up?

[young dew]

@fallow jungle hi

[fallow jungle]

Hi

[young dew]

You speak portuguese?

[formal kayak]

@neat elk when bot will up.. have to do some research 🥲 web version is not so handy.. please bring it up asap

[halcyon coral]

The bot has been taken out of service...

[formal kayak]

Have to search alternatives 🥴

[thorn moss]

what with these bots btw. I have never used them

[restive crest]

I see the same problem where the iOS app does not return any search results, for about a month now

[grim trellis]

Sorry deleted my message - it would not load ever so I deleted the app and reinstalled. The search is working for me again

[grim trellis]

I had issue with magic link logging in but after attempting 3 times it ended up working

[restive crest]

Reinstall resolved the issue, thank you!

[formal kayak]

I just asked a question and it will be saved for pro long time 🥲, now they discontinued it's hurts. Rest of bots are 💩

[formal kayak]

It is possible to enable the bot again or its permanently closed forever 🥲?

[tame current]

Wsg peeps

[grand hearth]

Yah we loved the bot in my other discord.

[tame current]

[formal kayak]

Pro subscription has discord bot? I'm planning to subscribe

[tame current]

What’s the bot in question?

[jaunty delta]

Can anyone share any Slack or Discord groups for San Francisco entrepreneurs that you're a part of?

[tame current]

Claude 2 my beloved

It just reads so much easier than whatever GPT-4 spits out

[mild robin]

Does perplexity have a discord bot version if not how would I make one?

[halcyon coral]

No, the bots are no longer supported

[gaunt moat]

Will there be no weekly events anymore?

[halcyon coral]

Yes, the #ask challenge will also be canceled

[halcyon coral]

@quasi grove If you have a problem using Perplexity, feel free to ask a question in #⚡│ask-community , if you have found a bug, please use #1140622008086970420 . Thank you very much!

[quasi grove]

Hm? I haven't found anythign worth noting..?

[formal kayak]

I badly need a discord bot 🥹

Team should ask users before discounting the Discord Bot 🥲. My request is at least enabled for pro users

[sweet jasper]

You can use perplexity API

At end of november they will incirporate online models, its already in labs

So you just have to use curl or write your own script

[halcyon coral]

@tough epoch
Please share your video in #😎│cool-projects!

[waxen narwhal]

please remember to respect the wishes of perplexity

[formal kayak]

It's beta... Soon they will disable that as well ? 🥴

[halcyon coral]

No, it will leave the beta in the future.

[pliant ridge]

i always thought pplx was 3.5 turbo but I guess they're serving their own model

[zenith atlas]

its fine-tuned 3.5

[pliant ridge]

oh nice, do they have any insights on how they did the finetune anywhere?

[zenith atlas]

🤷‍♂️

[pliant ridge]

:(

also how is pplx pro with claude and gpt-4

is the current 4 mode using 4-turbo now?

[formal kayak]

Which means api will be available without discontinue?

[zenith atlas]

no, turbo is still in preview and cant be used at full scale by other companies

[pliant ridge]

makes sense, i've always found base model works more than fine, and I prefer the faster generations than copilot

[formal kayak]

How perplexity will response on API... Like weblangchain?

[zenith atlas]

bots were not used that much, so thats the reason it was discontinued, API is actively developing so there is no reason for it to be discontinued in near future

[formal kayak]

Having discord bot at the corner of the channels.. will impact any performance 🥲

Having a bot is very useful to search my research topics ..... Kind of lazy... At least enable Direct message Bot. I believe many of them are using.... Ask users suggestions

I need only chat bot... not even image bot 🥴

[tame current]

Cope

Don’t be over reliant I guess

[tame current]

expired cool

[pliant ridge]

Don't have the pro role even though I paid for pro and rejoined using link – plz help

[tame current]

I'm totally new to AI!

My first time using AI!

No idea which AI model to choose

Do I just ask questions? Or do I also tell AI how to answer?

Also why do all models provide more or less same answers? Even after clicking rewrite, it writes the same text

[strange kelp]

Perplexity is not similar to ChatGPT. ChatGPT is good for conversation, Perplexity is good for search.

Use it just like you use Google. However, on Google you have to peep in to the links once search results are shown, but Perplexity will give you a direct answer.

just play around and ask some questions. Make sure to change your prompt to explore its capabilities.

[tame current]

aha ok, thank you!

people get pro roles?

mental

i don't think there is a pro role

[wraith crow]

Should be good now

[wraith crow]

You can get a pro role when you are subscribed

[cloud sequoia]

https://www.perplexity.ai/search/what-is-Venn-LtEnmjuKQle.ZC_05iFIkA?s=c

always wrong answer . @teal ridge @tame current

[cloud sequoia]

recently( these two weeks) ， there are more and more wrong answer.

[tame current]

i see

how do i get?

oh i got

thank you

WAIT how do you know i have the pro subscription 😭

[tame current]

same issue too

it just talked about japan out of nowhere

it ain't that bad cuz 600 prompts a day (it's lit)

[lilac yarrow]

so perplexity is like google meets ai - interesting! what advantage does it have over chatgpt then? wouldn’t chatgpt be simpler? no need for links, i just get direct answers.

[formal kayak]

anyone have Perplexity API codes in python

This Discord link is not working

[strange kelp]

Yes. You can just play around, you will will love it.

[silver gorge]

Is there any way that I can export my threads? Including the links, follow up questions, and such?

[upbeat turret]

I think the most important advantage that you can verify any information used by Perplexity by yourself

[tame current]

feels so much more solid than web browsing mode on chatgpt-4

multiple sources, reads a little easier

[sweet jasper]

Are you using custom instructions

You should probably disable them

[formal kayak]

Same question for me, can some one answer

[formal kayak]

@blissful raptor

[halcyon coral]

Already answered https://docs.perplexity.ai/discuss/652faa9bc5f287004a913a5c

[tame current]

Ok so 1 day with Perplexity and I don't find it useful

First of all: "Rewrite" button does nothing

Second: It's always the same 8 sources for the all questions about the same subject

Regular ChatGPT gives different answers all the time, I can ask it to rewrite how I want, ask it to use more sentences or less, change writing style, make lists, add more items to the list, expand explanations, etc

Nothing like this possible with Perplexity, it just gives me summary from the same 8 sources all the time, and when I copy the text, I have to do even more work to edit out the refrence numbers 1, 2, 3, that I don't care about

[blissful raptor]

the API will soon allow for web search. aka, the pplx-7b-online and pplx-70b-online models.

[blissful raptor]

rewrite button isn't meant to sample a "different" answer with the same prompt. it's meant to rewrite your answer with a "different" model. "it doesn't do anything" == you claiming all models are "same", which is really a prompt-dependent claim.

[thorn moss]

feels really good when perplexity ceo answers customers questions, it makes a big different than we having no clue to our questions. thanks for this.

[thorn moss]

i mean also from all the staff

[formal kayak]

Try to search this with these models on labs.perplexity.ai "она хочет испечь торт из трех ярусов" and search with the Web Perplexity UI. Web perplexity is Intelligent and able to predict what exactly I'm looking for...

[thorn moss]

are you trying to say that it is more intelligent than gpt 4? i dont know actually what set apart this 70b or 7b online does from gpt 4

[formal kayak]

Combo of Web + GPT = Perplexity Web UI ... Those GPT doesn't fetch real-time data from web.. I'm not sure how perplexity web UI is working are they scrapping web or simply calling Google or some other search engines API?

[thorn moss]

but it says online right so it is connected to web i suppose

[formal kayak]

Nope it's not fetch real-time as seen in perplexity Web UI

[thorn moss]

ah ok. weird cause sir Denis told us that it uses internet so i misunderstood it

[formal kayak]

API still in beta...mmmm if we want to use API beta the have to use Google search API's combo to fetch real-time... Something bla bla bla..... Anyway my fav Discord bot was discontinued looking for alternative ways 🥲

[formal kayak]

It's not something I'm looking for in the API... Those models not interacting with search engines to verify the data as seen in perplexity Web UI results.... Try with search "она хочет испечь торт из трех ярусов" on models and Perplexity Web UI..

[wanton zinc]

Can we possiblly have access to more a relaxed less evangical conversative model -- i'm a creatvie writer and i would love to use your service but your chat bot is very lets say relgious about certian themes which is somewhat getting in the way with my work

I'd be willing to pay extra for a relaxed model - agian i just want better results for creative writign

[sweet jasper]

Broski use claude or perplexity labs

Which has less restrictive models

[wanton zinc]

I'm trying to use it and its being very conservaitve

can you guys just relax the model a little bit - i just need it to be relaxed enough so i can write my first draft with your model before I send it to Sudo Write

[sweet jasper]

Part and parcel of big corps having many guardrails in place for llms

Its not perplexity fault, rather the underlying models

Like gpt4, claude etc

You might have less restrictions with open llms in perplexity labs which is low effory

[wanton zinc]

I dont understand if we are willing to pay for a more felxible model why cant we have it?

this is literally 1984 distopic

which model should i use?

[sweet jasper]

What type of content are you attempting to write?

[wanton zinc]

Smut

[sweet jasper]

Yeah thats going to be hard, no pun intended

[wanton zinc]

what model should i use?

[sweet jasper]

You can tey mistral

Try

[wanton zinc]

This makes me mad

So you're turning down a custoemr ebcause your product is for the relgious zealots?

[sweet jasper]

? Im not part of perplexity team

Mistral is open sourced

And is in perplexity labs

I can write porn with it

@wanton zinc

So again, perplexity only hosts the models, it does not fine tune them except for copilot. The smarter models have guard rails because the end user is businesses, not individuals

[wanton zinc]

is there any way i can get around the smarter models?

[wanton zinc]

whats the link for Broski?

[thorn moss]

get this guy a help my God. 😂😂😂

[sweet jasper]

https://labs.perplexity.ai/

[halcyon coral]

The well-known models such as GPT-4 or Claude are designed to be very secure; it should be almost impossible to bypass these fuses.

[sweet jasper]

You can try jailbreaking if you conceive a prompt for it, i am not aware of recent jailbreaks as they quickly get patched out if they are shared

[wanton zinc]

when will more liberal models be released?

dont comapnies know people will buy their products even more if they just relaxed their censorship polciy?

[sweet jasper]

Alot of liberal models have been released

WizardLM comes to mind

You can pay to use them in a VM or run locally.

If you have a good enough gpu for it

Again your usecase wont be supported by OpenAI and other big Ai developers

As the target market is different

Or you can buy huggingface subscription

[tame current]

jesus

Lazy to write smut?

get this guy some therapy

[tame current]

it isnt 1984 lil bro

[thorn moss]

that is what i wanted to say actually but i dont want it to make it rough for him 😂😂😂 thanks for saying

[tame current]

Theres nothing really bad when theyre talking about smut 😭

[slow kite]

any Black Friday offer for PRO?

[austere bridge]

I literally just signed up for Pro and the 2 months free comes out.....

[celest stone]

I signed up like a week before everything broke down.

I'm just rooting for perplexity at this point. dropping big promo code when you have huge user influx is big, because it increases retention a ton

now a bunch of people who weren't sure about trying Pro get to have a nice discount. and will get to see perplexity's true power and get completely addicted.

[pallid trout]

Where should i put the proml code?

[abstract thicket]

Any offer for existing paid users?

[bronze mason]

why is payment US only

[night bluff]

Sorry but, how can i use the 2 months free when i just get the perplexity pro ?

[signal hamlet]

Hi @bronze mason! It's not restricted to the U.S.. What was the error you got?

[bronze mason]

There's none, but most banks are american, cashapp isn't available in europe and not that many people (including me) have a credit card outside of america

So as much as I wish to subscribe I pretty much can't

And paypal is down for some reason

[austere bridge]

Following.. I just signed up for Pro this afternoon..

[signal hamlet]

Credit cards outside of the U.S. should work fine, including the European ones.

[bronze mason]

Yeah but I don't have one, only a debit card which uses IBAN

[feral raven]

I guess you don't. Only for new or returning subscribers. Out of luck if you signed up already

[night bluff]

its not fair..

[zenith atlas]

You can try writing to support email, maybe they’ll help if you subscribed very close to announcement, but can’t guarantee

[round dragon]

Not rewarding your actual paying members that ACTUALLY monetarily supported the service is a terrible move. Acting as if you're saying thank you and then screwing the people you should really be thankful for is ridiculous

[austere bridge]

I mean.. I signed up at 1:46pm... announcement just came out..

[wintry laurel]

I'm having the same issue too, I can't sign up. All my cards are being declined, even my debit card. (I promise I'm not broke 😂 ), I'm also in the USA

[bronze mason]

maybe stripe is having server issues

[torn zealot]

No problems for me

[pale sonnet]

Did anyone use the code and get the premium access?

Is it working, I have not tried it yet but just wanted to ask your thoughts

[feral raven]

It's a bummer, but companies do this a lot. Sign up a new contract and you get better terms

[bronze mason]

okay so pro benefits suddenly dissapeared?

pretty sure that aint the case

[celest stone]

thank you people who pay for this, it allows us to make a bold move and offer 2 months free for new sign ups at a time when people are coming in droves from chatgpt exile. thousands are going to maintain their subscription and have chosen us because of this discount over other possible competitors. this means our overall revenue is increasing overtime, allowing us to deliver more, better, sooner, faster.

"no but I wanted the free months"

[zenith atlas]

This promos are mostly made to attract new users or return ones who left, normal practise, dunno why you react so harsh

[fast meteor]

40usd off is only for yearly subscription I think. as a monthly subscriber you can upgrade to it with promo code thanks23

[drowsy crescent]

so, 20 + 20 $ which is 2 months discount for existing uesr right?

[zenith atlas]

I wasn’t replying to you…

[torn zealot]

GPT exile to.. a mostly GPT wrappers?

[celest stone]

Perplexity defaults to their own model which is a llama-2-70b finetune

GPT-4 and Claude are optional in the model picker.

Perplexity works better on their model

it can actually code, and retrieve info really well from websites

[frank rivet]

Please send an email to [e-mail] for more information on why your charge might be declined

[celest stone]

Besides Perplexity's own in-house models both offline and online with RAG, pplx-api serves opensource models on a super fast inference stack

but yeah, it's a GPT wrapper

[tame current]

can you only get 2 months free if you do the yearly plan? it seems like youd only get a month if you used the coupon on the monthly plan

[drowsy crescent]

@tame current it will apply 40 $ off for yearly and montly plan. so i believe it will be like 180 $ yearly. if it's monthly then 2 months off

[devout geyser]

Hi, I appreciate the thanks and indeed it has been a nice journey with Perplexity, however I was about to ask what about for those members who have paid already, possibly subscribed quite some time ago such as me? I'm guessing existing subscribers get no kind of thanks and only newcomers?

[bronze mason]

applies for monthly aswell

[bronze mason]

ever heard of subscriber benefits?

[devout geyser]

I'm already a subscriber 🙂

I was just curious, I know most companies these days do this as common practice to entice new customers to join while existing customers are often forgotten about (or worse, may sometimes get annual price increases)

[tame current]

yeah but its a 40 dollar coupon for a 20 dollar subscription and it seems like using it only gets you one month free

[bronze mason]

you get 2 months free

[tame current]

you get 40 off for 2 months

[bronze mason]

then you'll have the pay after that period

[drowsy crescent]

i got 2 months free

[tame current]

so the coupon carries over?

[bronze mason]

yes

[tame current]

i hope so

[bronze mason]

it does

you can see it at checkout

[tame current]

when i went to checkout it just said 40 off but the payment is only 20

didnt say anything about carrying over and i think the 2 months off is just referring to the coupon description since a month is 20

[digital siren]

Does anyone think something may come of this? No one seems to be talking about the fact that a majority of the staff threatened to quit today without Sam. https://www.wired.com/story/openai-staff-walk-protest-sam-altman/

[bronze mason]

one things for sure

i aint getting gpt premium or whatever it's called

[tame current]

damn

just got pro a few days ago

one must cope as hard as one can

But then think about it

its just 20 dollars

it aint alot in the grand scheme of things

[idle yarrow]

The free 2 months don't count for those that already have pro?

[tame current]

nope

[idle yarrow]

What a great reward for me helping fund the project 🙂

[tame current]

its just 20 dollars

Minimum wage pays that in 2 hours

for like a month of assistance

[compact bough]

So, its not free 2months. Its 3 months for the price of one?

[tame current]

Smething like that

[compact bough]

Lame and false advertisement

[tame current]

how

[bronze mason]

you can always cancel before you have to pay

[tame current]

thats 40 dollar savings

2 months is 60 days

[bronze mason]

so it's still 2 free months

just cancel after the 2nd month if you don't like it

[tame current]

USD conversion to cad means im paying 30 dollars a month

its like a dollar a day

horror

[compact bough]

Ohhh, if this is the case its cool
Cuz u can get and cancel same day. And still keep sub for 2 months

[tame current]

True

[feral hazel]

https://pplx.ai/thanks
I received an announcement. Is it a scam with a phishing link? Sounds good but not clicking yet.
Why is the URL not https://www.perplexity.ai/?

[eternal sequoia]

you can cancel same day and it won't cancel pro?

[compact bough]

Yea

But it seems to give 1mo free rarher than 2

[signal hamlet]

Hey @feral hazel! The link is ok. Pplx are the short links we are using, you can see it in previous announcements and across our socials.

[thorn moss]

if we are currently subscribed we cannot avail it?

[heady jay]

Heya @signal hamlet @kind vault - when’s the lifetime pro subscription for one-time fee going to drop? 1st customer right here… let’s gooooo! 🚀🚀🚀

[tough turret]

Hello all, I am here to get my weekly update on Internet grounded and PDF grounded inference on the API. Is that moving forward? Best regards to all the team and keep the awesome work.

[tame current]

how is that even feasible

[compact bough]

That would be an absurd loss of money for them. Each gpt4 call at the moment is 7 cents. Assuming they make excaptly one gpt4 call per search (they prob make more) that potentially $21 you can spend on a single day if you used up all 300 searches per day

[tame current]

API charges always exist

Nothing is truly free

especially live services that change so often

[heady jay]

Well, a one year anniversary 24-48 lifetime subscription run would be nice… just saying.

[tame current]

its always nice

but lets snap back to reality

get a job, get the bag, spend the bag, life continues

[heady jay]

Get a job? Do you know me?

[tame current]

If youre able to get an anni of free perplexity then good job for you

[tame current]

no i dont know you and i cant care enough to