💬│general

[drowsy hound]

Really

It also helps that it’s built in to my opera gx browser

Like on the side bar

Idk if you can do that with perplexity

Is this basically like extremely similar to bings copilot

Because that’s currently become my main chat bot for research

[warm cave]

Nice, I just started using it and am a fan

[foggy whale]

I did that before actually. was using ChatGPT Pro and Claude Pro. You get better privacy for ChatGPT through Perplexity for the same outlay. I've found image generation in perplexity to be a little frustrating and last night threw down $20 to get back into the straight Dall-E for another month but at this point my relationship with ChatGPT Pro is real transactional

[drowsy hound]

I don’t really care too much about image generation tbh personally

[foggy whale]

yeah i'm the same. its' ethically fraught and a secondary need

[drowsy hound]

I’m trying debate practice with it and even the free version is very in depth wow

What program does this run off of for the free version?

All of my friends are super anti generative ai but I’ve been using it for awhile and can see this having some serious potential for some people as an aid for them or a useful tool

[round sun]

where is the best place to ask playground questions?

[minor shard]

#🦙│feedback-playground but I think you already found it 🙂 or just this channel or #⚡│ask-community

[willow minnow]

If looking for daily markets newsletter, highly recommend Ticker Tea. Includes news recap, upcoming earnings & eco releases on a daily basis.

https://www.tickertea.com/subscribe

[minor shard]

Is this related to AI?

seems sus lol

[reef cave]

Hii

[foggy swan]

My subscription disappeared after I got my free year of perplexity pro

[minor shard]

You are not able to change the model or anything in your account?

[foggy swan]

Nope

I was able too when I first activated my subscription now I get a pop up for the subscription

[minor shard]

I would try signing out and back in, try a different browser and if all else fails make sure to reach out to [e-mail]. You can also submit #1140622008086970420 or #👉│feedback-general and hopefully an admin or support will get back to u soon

[foggy swan]

Already submitted an email

[minor shard]

Yeah unfortunately may just need to wait a little cause it’s late and I’m sure they will reach out first thing in AM

[foggy swan]

Fingers crossed cause it's a $200 subscription

[minor shard]

Yeah there have been a few other mentions of this kind of thing so they will help out pretty quickly from what I have seen

[foggy swan]

But it worked the first day, and now nothing so I'm worried

[minor shard]

I wouldn’t worry. They will definitely get it working again for you.

[foggy swan]

Hopefully

[south kindle]

The team is really solid at fixing stuff for users quickly. They'll make it right one way or another.

@minor shard they actually just credited my account for the egg hunt rewards, since we can't input discount codes. so my 1 month is currently working for Pro.

well not "just". they did it yesterday. but example of them fixing stuff for users quickly.

[foggy swan]

Well I even have an email saying my account was pro and thanking me for subscribing, just now with my code I was given for the year worked originally but not active anymore

Broke rn so I can't dump 200 for them to refund me

[south kindle]

Regarding prompting for the newly added 8x22 model:
https://www.reddit.com/r/LocalLLaMA/comments/1c0tdsb/comment/kyzsho1/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

[half venture]

Google coming in clutch

Infinite context

Currently only 1b

But once it scales

It gives basically infinite context

Openai is cooked

😭

[keen solstice]

Hi, does anyone know if Perplexity updated GPT4 for the new api or if not when they will do it ?

[scarlet schooner]

infinite context?

wow

[torn sierra]

The topic of image generation has been discussed here many times. They say that Perplexity is a search engine, not a chatbot. However, I believe this is not true because the two are interconnected. When working on something, you delve into the topic, refine the text, and sometimes you need to generate images.

[tame current]

no, it's still there

[verbal fable]

does this mean that a well trained GPT model could write a novel in minutes? lol

[fast meteor]

recycled ideas from rnn applied to the transformer

[granite temple]

Hi all, I hope it's okay to address this here but I'm looking for ways to use Perplexity for my client website. I would need to know how the API is usable and flexible on an external eshop website. Maybe I would need to get in touch with the team ?
Thank you all in advance and sorry for the message that isn't very precise.

[devout bronze]

in the UI it shows the trial but on Stripe it wants to bill the entire thing

• do note this is not on my account (i’m pro) but this is my friends account and i was double checking the stripe cause he says it’s trying to bill him the $200 without a trial, too

[devout bronze]

you can create an API key in the perplexity.ai settings, then browse the documentation for their models at https://docs.perplexity.ai/

if your use case requires web search for up-to-date information, you’d want to use the models that end in -online otherwise you can use any other model

[fast meteor]

#💬│general message
👀 👀

[granite temple]

thank you very much !

[foggy swan]

Let's see how they can fix my membership hopefully I get a response today via here or email

[devout bronze]

ah that explains it

we were using domains from our work so that makes sense

is gmail whitelisted?

[orchid magnet]

I’m bored what are some fun ai tools

[livid mantle]

"Where Knowledge Meets Perplexity" could have been a better book cover Mr. Alex Romanov.

[tame current]

probably

[fast meteor]

https://www.youtube.com/watch?v=0O2yTG3n1Vc
still no hands on reviews on rabbit r1.

[valid yoke]

Yep the universal consensus is: don’t buy it

[minor shard]

Yeah as I look back now, they did say 3 weeks for shipping after the 31st for the Rabbit R1. Almost timed perfectly with their party they are throwing. I just want to see an average person using it...

[tame current]

does rabbit r1 require internet?

[ashen maple]

Yes. It's not really much of an advantage compared to just having an app on your phone that uses the camera etc. ¯_(ツ)_/¯

[fast meteor]

probably, the cpu MediaTek MT6765 has very weak raw performance, worse than Qualcomm Snapdragon 460 (2017)

[livid mantle]

I have 400 searches left, has anyone gotten this far? I am sure many has. So what's the deal?

[fast meteor]

0 left yesterday, testing the prompt

[livid mantle]

Not going to lie but when i reached 0, after I tried to submit my input, the webpage just becomes the profile picture of Alex Romanov.

[foggy swan]

How long does it take support to usually respond to emails ?

[livid mantle]

I literally could not doing anything else.

[foggy whale]

that's a review that doesn't suck. thanks for sharing that

but also not surprising for a gen1 thing that is so tiny and offloading everything to the cloud for now

[valid yoke]

Don’t all LLMs

[foggy whale]

practically speaking yes unless you've got it running on your own box like you can do with MacWhisper

[tame current]

technically it is possible to run LLMs locally

though , I've never tried it

btw, did something change in the free version?

something feels.... different

maybe they changed the model ?

[fast meteor]

try it now!https://github.com/ollama/ollama

[tame current]

what are the minimum system requirements?

or well, recommended

[fast meteor]

Note: You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
By default, Ollama uses 4-bit quantization

[tame current]

hm, I'll try it thanks

[minor shard]

Did you see Alex's response to your Bug report

[sick geode]

Hi, are we able to summarize articles using models like Opus?

[agile jay]

The extension

[sick geode]

Whenever I try to do similar things, almost all models start hallucinating

Thinking the article is not existing, currently in the future, or having knowledge cutoffs exactly one year prior.

An example here
https://www.perplexity.ai/search/Summarize-this-article-TmAizRB7QEKBxecCdrvEFA

[agile jay]

Expand the sidebar, and then click the download thing and select extension.

[sick geode]

Excuse me but can you elaborate? You meant chrome extension right? But I actually needed it on my mobile so I had no choice but using the app 😦

[agile jay]

With the extension, you can summarize or ask questions about the webpage you are on.

[agile jay]

Use a different browser app.

Quite a few of them support extensions

[sick geode]

Is there any iOS option u can think of?

I’ll just ask perplexity 😂

[agile jay]

We are currently sueing apple, so they currently might not allow browser extensions on iOS.

[ebon acorn]

/imagine

[agile jay]

This isn't the midjourney server...

[fast meteor]

not really, they (pplx) only implemented a server side fetch, no client side fetch

[agile jay]

Oh, let me modify the extension to extract the text from the webpage then.

[fast meteor]

https://discord.com/channels/1047197230748151888/1205077911233626132

wallabag is a self-hosted web page clipper, its plugin provides the option whether to fetch the webpage locally. here is the implementation (MIT license) for the reference: https://github.com/wallabag/wallabagger/blob/bc9bae830c2f51403b1679efdfab9a497365f05d/wallabagger/js/options.js#L109

the extension feels like the second class citizen, it was last updated on 2023 oct

[agile jay]

It is, and it only works for chromium browsers. I had to repackage it for it to work on Firefox.

[near imp]

Is there a reason you don't allow users to select a model in the Prompt window? I would love to easily switch from "Pro Search" to Opus or Sonar, but I always need to do a Query first (I'm set to Claude 3 Opus) and then REWRITE. it's wasteful and takes extra time.

[fast meteor]

(dont really like the idea of creating a fork, pplx should maintain its own code well, but i have waited long enough)

also site only gives you the illusion of fetching the url on your behalf. It retrieves the internal db instead

[fast meteor]

this link redirected to cnbc

[agile jay]

It's a Twitter redirect url

[vagrant frigate]

How good is Perplexity to learn Korean?

[near imp]

That question is not sound.

because that would depend on what model you use.

[fast meteor]

yes, even the wget can handle it, not to mention playwright. my point is the site function is flawed

[agile jay]

If I was gonna use a model to learn korean, it would probably be the ChatGPT App with the dialogue mode. You can try it with the free mode.

[agile jay]

Yep, because they use an index, not a live crawl.

Otherwise searches would take a lot longer.

You can mimic what they do though, by injecting the webpage the same way they do for their indexer.

[livid mantle]

I am waiting for Gemini Advanced 1.5 Ultra.

[agile jay]

It will probably still have the terrible Google filter though

[livid mantle]

Oh ueah for sure

[fast meteor]

https://www.perplexity.ai/search/oj-simpson-death-SJUeBVjqSXef1sBmoA2GVw

[tame current]

Peepo_Wave

[agile jay]

🌊

[trail sparrow]

hello here

[trail sparrow]

I represent Creative Agency Here, and we're actively looking to enhance our services through innovative technological solutions. We're particularly interested in the capabilities of Perplexity's API and the potential for integration within our agency's ecosystem.

Our goal is to streamline our creative processes, elevate our project delivery, and provide our clients with an even higher standard of service. I'm keen to delve into a detailed discussion about the possibilities of an API integration and how it could benefit both our operations.

Could we discuss the available options for integration or perhaps arrange a meeting to further explore a potential partnership?

Thank you for your assistance. I look forward to the opportunity of our agencies collaborating.

Best wishes,
Ilya Zaigralov
Owner at Creative Agency Here
Contact Information - wa.me/[phone] or Discord, also t.me/IlyaZaigralovHere (Russian community Here)

[warm cave]

Wait, You.com has Unlimited of all of these models including Gemini 1.0 and 1.5.

I wonder what the context limits are

[halcyon coral]

32k afaik 🤔

[warm cave]

hmm, that is interesting...

Gemini 1.5 from what i know really only excels in context length and creativity, tho Claude has good creativity so gemini really seems only good for the massive context

But it is still nice to have the option even with limited context, tho larger would be nice

[minor shard]

I saw some messages from one of the admins saying their Opus model was close to 200k

[agile jay]

pretty sure unlimited means 500/day

[warm cave]

Yeah, I would guess so

I wish companys would just list the context window, too much confusion

[lost lava]

hi here, why do i get 401 from perplexity api?

[agile jay]

401 means unauthorized

[halcyon coral]

Please check the authorization header

[agile jay]

Does the API use a bearer token?

[halcyon coral]

"Bearer XYZ"

[warm cave]

also I am really liking raycast, and I have only played with it a little:

Its nice to be able to use it with both perplexity search and perplexity API

The store has a massive library of apps

[lost lava]

ok i figure out the issue, I've updated my payment, and auto top up was inactive then

[torn sierra]

They officially said it’s 200k for Opus. Have no idea about Gemini 1.5 pro

[torn sierra]

#958041026415718414 message

[lavish sequoia]

This will be extremely useful.

[trim wind]

Hey everyone, just saw this new evaluation startup. Curious what you all think https://www.bloomberg.com/news/newsletters/2024-04-11/this-startup-is-trying-to-test-how-well-ai-models-actually-work

[formal parcel]

Upvote plz:
https://discord.com/channels/1047197230748151888/1228107753596387358

[zealous kayak]

I know opus on pplx pro is limited to something like 30k context, but what about sonnet? Does that get the full context?

[torn sierra]

Same. Unfortunately, also around 30k. I tried to do something with large files. It only handled about 30k of content. So at least in the case of files, it truncates their content.

[foggy whale]

I just threw down my $20 to give the full boat a shot to see if I need the fancier models. Also haven't gotten far yet, but I like this dedicated chat window. Doesn't appear to support dragging files into it yet. i really was planning on getting more sleep tonight

[minor shard]

maybe u can get more sleep next week 🙂

[unkempt grove]

@livid mantle oque é energia solar

oque é energia solar

[wise edge]

is pplx gpt-4 reference to "GPT-4-Turbo-2024-04-09" now ?

[minor shard]

Based on my testing it has been updated to the new GPT Turbo! https://www.perplexity.ai/search/Who-is-the-sFCdW3o1S0S6KdjiiZ0e1g#0

Since it has info up to December 2023

[warm crater]

isnt it accessing the internet anyways? so that doesnt seem to be a reliable way to test that

[minor shard]

Writing mode does not use internet.

[warm crater]

ohh, i didnt explore it enough

but the chat answers like that in writing mode. just tested

[minor shard]

Yeah, it is just using the model.

[warm crater]

do you know if this new update is now "state of the art" and have surpassed claude opus?

[minor shard]

No I do not sorry. I think it will depend on your use case

[warm crater]

i saw some comparisons but nothing conclusive like the benchmarks anthropic has on their website

[tame current]

doesnt 0125 have a cut off of up to dec 2023 too

[minor shard]

I thought it was April 2023

[minor shard]

yeah no way to really tell

I know previously it only had data up to April 2023

so not sure if its the better model

I wish OpenAI would be more clear lol

[chilly nova]

what do you guys think of this? https://www.reddit.com/r/perplexity_ai/comments/1bl8kc2/perplexity_limits_the_claude_3_opus_context/

[chilly nova]

damn, time to ditch Perplexity next month then

and it's even cheaper... how di i miss it

what do you guys use to convert folders into an embedding?

i was using this but it's annoying af and super slow https://github.com/JeremiahPetersen/RepoToText

you need to spin up the container every time

i just want a tool where i can drag and drop a folder ideally and it merges all files inside it into a txt

[devout bronze]

same way gh copilot does it

just take file contents and embed them

put it in RAG

[chilly nova]

i don't have copilot. what does it do?

[devout bronze]

takes files and each file gets an embedding using OpenAI api

well, their own private one but i revved it xd

but then they send that to the LLM

[chilly nova]

i just need something like a zipper, but instead of creating a .zip, create a .txt

[devout bronze]

write a small py script to take the files and put them in a txt?

[chilly nova]

and it would be good if it had a parameter with tokens number

and it errors if your input exceeds the number and tells you what's the token size

i don't get why all these LLM providers cannot do this for you...

gemini pro studio allows you to upload separate files, very inefficient

https://github.com/donoceidon/repo2txt

[devout bronze]

cause that's not what they do

they build the LLMs, you build the tooling

[chilly nova]

doesn't make sense. embedding is part of the LLM

that's why google allows you to add it in AI Studio

but they are dumb and allow you to upload all files separately in Drive

[tame current]

@halcyon coral ?

or is it just my imagination

[chilly nova]

i hate that perplexity ai sometimes bugs and keeps referring the files

even when you clear them, he keeps referring them

and you need to start a new thread

and also it attaches stuff automatically even if you don't want to

after a certain amount of text

so you always have to remember to clear the embeddings

i wish it was better, like a space where you can upload umbeddings that persist throughout the thread, and then those that you provide via chat that are one-time only

[torn sierra]

Just got new gpt-4 turbo in ChatGPT. Well... I chose Perplexity to try Claude 3 Opus but I cant make much of it since you limit the context window. ChatGPT is far more versatile. Setting aside the context window for now, given the economic limitations, but there are still a bunch of things here that I’m just not into. However, I’ve got to say, the search function is pretty awesome. If ChatGPT or anything else falls short for me, I’ll probably stick with using Perplexity for my searches.

[chilly nova]

how do you use Perplexity GPT4 in ChatGPT? you need OpenAI sub for that

GPT4-Turbo context also sucks tho

better pay for Anthropic at that point

[foggy tapir]

I have been using chatgpt plus for 6-7 months but still I can't see Turbo in my models. How did you get it?

[chilly nova]

If only Gemini 1.5 Pro was decent

the 1M token context is the best right now

[chilly nova]

When you use ChatGPT from the chat, Turbo is automatically used when you exceed certain token threshold or use capabilities that exist only in Turbo like read images/files.

[torn sierra]

It’s been just released on ChatGPT. There was an official announcement on X. I checked, and it’s there. It has knowledge up until the end of 2023. It also clearly performs better with tasks that the previous model struggled with. I’ve just tested it.

[foggy tapir]

Okay. But this is my observation. I tried doing a code review of a python script (400 lines). It takes 500 characters of the file for review in chatgpt 4.

But when I use perplexity with gpt 4 turbo, it reviews all of my code.

[torn sierra]

I m talking about ChatGPT app and the models that are used there. I got a Team Plan there. And yeah, context window is still 32k in gpt-4 turbo unless you have enterprise sub

[celest stump]

Hi, I work for a fairly small company, and we are slowly starting to implement Perplexity in our work. Please tell me if it is possible to organize a team subscription for us (a small number of people, for example - 4-5 people to begin with)

[torn sierra]

I played around a lot with ChatGPT and its new gpt-4 turbo model. Perplexity generates higher quality texts compared to what I get from ChatGPT. At least that's what I've noticed. I will experiment some more with different models.

[fast meteor]

For corporate, can I share my account with others within my organization?

Each individual employee has to have their own Perplexity Pro account. A single payment method is able to be used for all employees.

We will be happy to help you set up multiple accounts as we continue to work on expanding our subscription services, please email [e-mail] for assistance.

[tulip terrace]

Hi, Anyone out here using perplexity pro?

[minor shard]

Yes

[crystal elbow]

anyone else not seeing how many pro searches are left on their account?

[tulip terrace]

Can you please DM, i am not able to DM you

[minor shard]

I sent you a friend request

[foggy tapir]

Where do we check it ?

[crystal elbow]

normally hover over "pro" in the chat

[stable fossil]

is new gpt4 turbo 2024-04-09 model available in perplexity pro?

Cut off date says it's April 2024

[minor shard]

It appears to have correct information up to December 2023

[glass spear]

dec 2023

[stable fossil]

Well. Upon couple of refresh. I guess I got the right answer

[minor shard]

Yeah don’t ask that. Find something that happened around October/November and ask that

[meager sparrow]

Waw so that means it is unreliable and inaccurate this year?

[stable fossil]

I feel opus is much better than gpt4. What do you think? What is you default model?

[fast meteor]

not necessarily, the llm is grounded with search results

[meager sparrow]

I mean yea but that’s because it is still running under Perplexity (the Ai-powered search engine) technically your not using the pure version of GPT-4 because it’s answer results are still bound by perplexities capabilities to some degree. Unless you go to “chatGPT” and use GPT-4 there.

[fast meteor]

that is the main purpose why pplx exists

[meager sparrow]

And when you go to You.com which is just like perplexity you get a slightly different deviation of GPT-4s capabilities

[tame current]

or using the writing focus

I'm always surprised by how many people don't know what that is

[stable fossil]

Ye I use writing mode 99% of the time.

[compact stirrup]

hi, I am new 🙂

what models do you use in the settings as defaults?

[foggy whale]

id encourage you to play around, i think easier ways to swap models are on the horizon. For me i set the defaults to Opus and Playground, but almost as much if i'm just doing a search i'll flop it to the perplexity models. Different tools for different tasks, all wrangled by perplexity being the management layer

and of course don't forget to make a choice about that training setting. Frankly, that's something i turn off and most folks I hang with do. Nice thing is that it's an option

[compact stirrup]

i just paid for pro to get access to chatgpt 4 - technically that's chatgpt 4 turbo, I am learning - because the free version of perplexity.ai still had bad answers for my programming questions

[compact stirrup]

gotta check out what that's about

[foggy whale]

I had ChatGPT Pro for tasks that weren't confidential because even if you pay they're still hoovering your data unless you're on a team plan. Going through perplexity you get the extra privacy layer there.

[compact stirrup]

and I would assume the turbo is less of an enhanced and more of an efficiency thing

[compact stirrup]

is that the "ai data retention" option? why do you turn it off?

[foggy whale]

Oh the whole issue of whether any AI based tool is taking your data and using it for their future use instead of reteaining it on your behalf is if nothing else a huge concern for businesses and people. I don't say that to fearmonger, it's just something that people need to think about when they sign up for stuff.

[arctic sluice]

GPT-4 Turbo says it is an AI made by Perplexity. Ehm???

[foggy whale]

and that level of concern is a sliding scale, which is something lost on a lot of people I talk to right now. for example, depending on what i'm doing or discussing, I want the most private thing possible. But if I need to figure out what the fastest way to thaw shrimp is while I'm making dinner, ChatGPT via it's voice interface is both totally acceptable from a privacy standpoint in my world and hell yeah because that voice interface is perfectly implemented

[halcyon coral]

https://discord.com/channels/1047197230748151888/1198438847877480629

[orchid magnet]

Opus is better

[minor shard]

I use the Sonarr model as it is fast. I rewrite to other models when I do not like the answer.

[orchid magnet]

Sonar is lightning fast

But it’s ass

[compact stirrup]

i feel like if I need fast, I use google search. I want to be sure to get the most sophisticated answer and develop some sort of intuition what I can expect from LLMs

i am trying pro to see if the latest models are indeed a notable improvement above my google search techniques

in terms of quality that is

the free version of perplexity was a disappointment in that regard. I prefer good ol' google to an LLM that is mostly correct

[minor shard]

yeah thats understandable. Thats why when people ask "Which model is best" I always says its up to ur use case. Cause everyone has different goals

[wide umbra]

Do any of the Perplexity models support functions / tools?

[compact stirrup]

asking chatgpt 4 turbo for specific prompts didn't give me any. seems like a lost opportunity ... some of the more interesting prompts to learn about LLMs require reading original sources and watching youtube? so 90s

[fluid mauve]

and playing around, lots of playing around. Utilising the API and Python or any other coding language opens a whole host off opertunities

[compact stirrup]

how is that? what do I get from using the API?

[fluid mauve]

function calling in python. roll your logic there

[compact stirrup]

i discovered that some of the simpler things can make a huge difference. e.g. trying to have a psychotherapeutic conversation with an LLM is possible. That is using an LLM as sort of an active listener who is mostly validating whatever you tell it. However, if you don't include in your prompt that you want just that, you get this encyclopedic default which can get really annoying

[fluid mauve]

I chat with Pi a lot. It makes much morse sense than my wife. Even says i should go fishing more. Mind you im not planning on ditching the wife.

[tepid portal]

Is it possible to consider more than 20 sources? For example 44 or 100?

Or I need pay premium?

[foggy whale]

Yeah you're getting it the way I came to this. Once i got into things that weren't ChatGPT where i could feed my own data to it, or really engage in a conversation that wasn't some kind of cherade the results are very different. Unless I'm trying to get repeatable results with consistent and narrow output (like say, categorizing a file) trying to over-nerd things is surprisingly unhelpful. Overall no matter what you're using it for, more context is always better though so just taking a traditional search in Google and popping it into Perplexity isn't necessarily a 1:1 comparison either

Like take Claude. Sure there's some low level documentation but the more i worked with it the more i realzed that the documentatino is in the description. Take it literally.

[compact stirrup]

in order to enjoy a chat with an LLM, I think I am missing one that actually has character and can challenge my thoughts somehow. Before I find an LLM that can poke fun at what I write, they don't serve as conversational partner

(maybe I just haven't found the prompt to get that, yet)

[foggy whale]

Have you tried flipping on your mic and using transcription instead of typing yet?

[compact stirrup]

not yet

[foggy whale]

you should! With claude certainly and GPT to a degree it does a way better job of matching your overall vocabulary and tone when you just talk

[sterile falcon]

Hey everyone! Do you happen to know if using GPT 4 Turbo or Claude 3 Opus in writing mode generates the same responses as if I was using them directly through GPT plus/Claude pro?

[austere kestrel]

I don't think so - haven't seen more than 20 myself anyway

Can't quite work it out. Think you can get it to make more than 20 queries as part of the retrieval process, but it still won't take more than 20 individual srouces from those queries

[tepid portal]

Yeah, that's why I ask, I just wondered

[tepid portal]

Damm, okay, that would be enough for me.

[dim ginkgo]

Can the recently upgraded GPT 4 Turbo beat Claude 3 Opus?

[hallow niche]

That would be a negative. It’s vision capabilities were upgraded but all the other things remain the same

[dim ginkgo]

I saw OpenAI saying GPT 4 is upgraded in terms of Logical Reasoning, coding, Vision etc.?

[hallow niche]

If that’s the case I missed that. My bad for the misinformation

[tame current]

it used to use that many sources by default

but then they reduced it

☹️

[silk umbra]

Does anyone also notice that the chat UI today doesn't show remaining credits? Just yesterday it shows how many credits I have left when hovering on Pro switch button.

[tame current]

and I don't think there's any other tool that will do it either

[tame current]

possibly, it's ranked higher on lmsys leaderboard

[tame current]

someone else said this too

[tame current]

^

possible that they changed something

[silk umbra]

Thanks, I tend to control my credit for important tasks but now without knowing how many I have left I cannot do that. Don' t know if they increase their limit or just decided to hide that info

[tame current]

yeah, I guess you'll just need to wait for them to confirm it

[vapid onyx]

Happy Friday!!

[hollow flame]

So the new GPT-4 Turbo by default is much less verbose

[livid mantle]

ya man im going to work

[hollow flame]

Giving 1-3 sentences only by default

[livid mantle]

but enjoy it

ill be at the cashier in mcdonalds to give you all the food u will munch on throughout your friday

[compact stirrup]

question: i can select an "image generation model" in the settings. but how do i use perplexity to generate images?

so far i only got "I'm sorry, but as a text-based AI, I'm not able to generate images. "

sure ... let me "help" you with your essay ...

[serene sand]

[vapid onyx]

Call out and have perplexity do the heavy lifting

[minor shard]

[foggy tapir]

As stated earlier by someone, image generation is not possible in Perplexity

[minor shard]

It is possible. Not sure where you heard it wasn't possible: #1194794305362071552 message

[minor shard]

Try this sir: #1194794305362071552 message

[meager sparrow]

[foggy tapir]

I asked the same question a few days ago, and someone replied to me with a link which started with "perplexity is more of an AI search engine.."

Thanks, I am able to generate images using the prompt

[tepid portal]

Maybe time got feature request? 😅

[minor shard]

No problem! Glad I could help 🙂

[foggy tapir]

Well, peeps doing code reviews using Perplexity, is there a certain prompt that you use?

[young dew]

drink water

[twilit halo]

raycast + perplexity is SICK

[mental finch]

In B2C production implementations, is there any way to reduce fix 5$/1000 request price? With 500 token per each api call I spend 0.0053 so it's not profitable in scenarios with over 3, 4 5 million questions per month.

Thanks in advance!

[sinful pine]

@wraith crow

[wraith crow]

please don’t advertise selling codes in the server thanks

[sinful pine]

thanks 🙂

[topaz eagle]

Hey guys! would love to connect with the team somehow - have sent a few emails to the api@ email

[cursive jacinth]

Hi sir, is Gemini Ultra or Gemini 1.5 still on plan sir or should we forget it sir? you.com has it now sir. @halcyon coral

[wicked vine]

Did you ever get it working? Having the same issue

[hallow linden]

Sounds like you need a new $26,500 a month revenue stream

[zealous kayak]

I keep seeing people harp on about you.com
to the point where I'm starting to think it's some weird astroturf

I can't find any actual numbers on what their usage caps (or context window) are

granted, pplx doesn't publish the context window number either but still

[agile jay]

Is this for the perplexity API?

[zealous kayak]

very, very annoying

[zealous kayak]

apparently 'no limit on context length'

I really do not believe this.

only way I can see this being true is if they're literally just deciding to burn cash for a week to get users, then lock it down

[agile jay]

Like he said, depends on the model they are paying for...

If they are using the lowest version, then technically they aren't limiting it...

They are just using the version with the smallest context...

[zealous kayak]

that... doesn't really exist for a lot of models

opus and gpt-4-turbo for example do not have a smaller context window version

[agile jay]

GPT4Turbo definitely does.

[zealous kayak]

I...no, it does not.

[agile jay]

GPTPlus users only have access to the 32K version.

[zealous kayak]

No, that's not how that works.

it's the site that limits you to only sending 32k tokens to it

the underlying model is still the 128k version

the site just doesn't send anything that goes over 32k tokens

it's like placing a filter in the middle

either way, another user chimed in on their discord

in short - just seems like your standard rugpull trying to pull in users by lying about their platform's capabilities

[fluid mauve]

You mean marketing? right? 😆

[agile jay]

Or not, since input is cheaper than output.

They probably limit the output then.

Also they most likely have an enterprise deal for lower costs.

[zealous kayak]

it's only cheaper on a token by token basis

you have far more input tokens than output tokens for a given request, assuming it's a long convo

[agile jay]

Yep, still 3x cheaper

[zealous kayak]

so overall, the price of a request is mostly down to how much input you have

like, limiting the output tokens and not limiting the input tokens is going to change the final price by very little

[agile jay]

That's if they aren't using RAG etc.

[zealous kayak]

well, yes, which is what I suspect

[agile jay]

They could also be using the 129K model but using RAG and other techniques.

[zealous kayak]

meaning that it's not really the full context window - they're doing a similar thing to pplx, limiting how much data they send and using RAG

[zealous kayak]

I mean, okay, fair
though usually marketing people are smart enough to avoid outright lying

[agile jay]

If it's just RAG, it would technically not be lying.

[zealous kayak]

Eh...I would say it is? You're still limiting the context window

but I guess you could probably weasel your way out of it

[fluid mauve]

Only sometimes. and if you believe it (like marketing people do) is it really a lie in their view? I know it is dubious in our view

[zealous kayak]

yeah, I guess? Either way, I find it really strange how mentions of the platform ramped up in a pretty strange way all at the same time in a few discords I'm in

¯_(ツ)_/¯

I smell an astroturf personally

[agile jay]

I think someone said something about uploading a book. Unless the book is in a txt or md file, it's a lot more than the number of words in the book.

Since stuff like epub use xhtml etc.

[agile jay]

Can't people just try them all and decide for themselves?

[fluid mauve]

Believe me, I dont think your 'spideysense' is that far off.

[zealous kayak]

I mean, if you have unlimited money, lol

[agile jay]

In my case, I just pay for all of them.

[fluid mauve]

I am pretty similar, and then forget I have them.

[agile jay]

There were a few people spamming others about it, but I believe they were banned.

[agile jay]

Yep, by far I use perplexity the most.

Would like it if they refined a few things though.

[fluid mauve]

Same. , its why i just hang out here. I mean I have ollama and a bunch ofg models running locally here but even then I havent 'played' with them much

[zealous kayak]

I would love to use perplexity, but the short context just doesn't cut it for me

[agile jay]

Like image gen, and the random stutter when clicking stuff, because for some reason they are changing the overflow value of the html tag when they're clicked...

[zealous kayak]

I get why they're doing it for opus but man, I wish we could at least get 64k context for sonnet or something

[agile jay]

I'll move to my server when open-source finally catches up.

[fluid mauve]

I ran out of contect on my Jetbrains IDE today. id just uploaded about 1000 lines of html and css. told me i was something like 3k over token limit

[agile jay]

Oh, is sonnet also capped?

Thought it was just opus and gpt4T

Oh, I don't use AI for writing code currently. The failure rate is too high.

Much faster to just manually write it.

[fluid mauve]

I am so bored of reading 17 pages of Terms and conditions of service. its bloody 10:50Pm here in the UK ON A FRIDAY NIGHT. I should have gone to the pub

[agile jay]

The TOS for what?

[zealous kayak]

Yeah, Sonnet's also capped - at least according to the people that told me here

[fluid mauve]

leagal shizzle for a company im starting

[zealous kayak]

see, maybe it's not, but that's the problem with not publishing the context window

[agile jay]

Well I wouldn't trust anyone unless it's a mod (yellow) or dev (green).

[zealous kayak]

it depends on the language a lot

python is usually pretty decent

I used GPT-4 to do pyqt UI code for example

[zealous kayak]

Yeah, but they won't share that info since the company doesn't (assuming they even know it in the first place). So I just have to rely on other users, y'know.

[agile jay]

For data science, it's probably enough.

[fluid mauve]

I have run some C and C++ stuff through codellama and it was quite decent

[agile jay]

But not for anything actually being used by others.

[fluid mauve]

maybe i should try assembler 😛

[agile jay]

Lol, the gcc compiler is more intelligent than a person manually writing assembly.

But I guess future LLM's will be the best compilers...

[fluid mauve]

LoL, my last assembler was on a 6502, and that was back in commodore 64 days circa 1986

[agile jay]

Mine was a few days ago when going over some wasm code.

[fluid mauve]

ta suck, fast though

[agile jay]

To see how changing stuff would change the wasm outputs, for optimization reasons.

Yep, hopefully the web moves over to wasm.

[fluid mauve]

Assembler, propper engineering

unlike my spelling

[agile jay]

Lol you go too low level and you're not making any progress.

[fluid mauve]

Queue tumbleweed - When i first started out as a computer service engineer in 1990 i was regularly doing board level repairs on customers sites. a few years past and SMT came along, it then became board swapping and a lot less engineering.

I'd say I miss my ocilloscope, but i dont. Heavy piece of shizzle.

[modest scarab]

Yes

[whole lark]

With GPT-4, NSFW content is an absolute no-no, but with Claude3, I can use it enough to write a novel. In Japan, someone who used GPT4 to write a novel won a famous award and became the talk of the town, but when you try it yourself, it's easy to write a novel.

[warm cave]

Are you saying that Claude 3 can do NSFW content? I thought it Anthropic had it locked down tight for that type of stuff

[whole lark]

That's right. But when I asked him to write some phrases for a sensual novel, he also wrote a description of a very racy scene. It's something like.

[warm cave]

Claude 3 usage 📈 Guys I swear I use Claude 3 for the context window, haha 😅 that’s it 😐

[tame current]

does anyone know if we will get grok on perplexity anytime soon?

[wicked zodiac]

How is the Sonar model? Anyone notice any differences?

I just selected it for the first time so I guess I'll find out

[minor shard]

My experience is that its super fast but not as smart as something like Opus or GPT4

[sly minnow]

Hello guys. Can you guys see how many PRO quota left while sending search?
It left only hint (Ctrl+.) while hover over the [PRO] icon.

I could still see the quota yesterday iirc.

[minor shard]

This is the update from the team regarding that issue: #1228400472151294012 message

Are you a pro member?

[sly minnow]

Yes

[sly minnow]

Got it, thanks.

[mental finch]

Yes

[tame current]

I really feel like the model in perplexity was changed but no one's talking about it

the output is different

[tame current]

the answer also seems a bit worse than before

somehow still better than all the competition...

[hasty musk]

Why when im using claude 3 opus and asking him are you claude? im getting the answer?:
"No, I am not Claude. As I mentioned, my name is Perplexity and I was created by Perplexity AI, not by Anthropic who developed Claude. Claude and I are different AI assistants."

[halcyon coral]

#💬│general message

[arctic timber]

Hello can somone help me with the image generation I put the prompt but only return me text and no generated images, or just tell me i cant create images xD #💬│general

[halcyon coral]

https://discord.com/channels/1047197230748151888/1194794305362071552

[austere kestrel]

exactly right.. i don't get what is so complicated about this...
more people than not seem to think input token inputs mean that the model's context window has been shortened... as you say that's not how it works

[zealous kayak]

yeah, the model still has the same context window - you just don't get to use it because what you send it gets truncated

so in a way to the end user it appears like the model has less memory

[sinful pine]

guys, my perplexity working wrong...

my output setting is setup to turkish language but get output sometimes in english language or sometimes not accurate turkish... how can i fix?

[austere kestrel]

you get to use it. Every message you send (and receive) goes into the context window stays there - well, until it fills up (e.g. with 128k tokens, or 200k tokens)

[zealous kayak]

Nah, that's not how these services operate

[austere kestrel]

The problem / limitation is, you can only send so much in one message (using perplexity, that is).

[zealous kayak]

like, chatGPT for example

it truncates your message history after 32k tokens

even on chatGPT plus

[austere kestrel]

yeah that's true, for chatgpt tbf

[zealous kayak]

perplexity does a similar thing, going by people's testing

[austere kestrel]

but not for claude via anthropic.com

[zealous kayak]

and like, it doesn't make sense for them to just limit the amount you can send in one message

[austere kestrel]

or gpt via API

[zealous kayak]

yeah, that is the one exception (besides APIs) - they lower your cap instead

[austere kestrel]

maybe, but my testing suggests otherwise ig
https://www.perplexity.ai/search/I-will-provide-.MaKRen9TQumbfMQF0CVKw

[zealous kayak]

well, it's a bit fuzzy is the thing

perplexity does RAG

I believe

[austere kestrel]

for file uploads

[zealous kayak]

which is kind of like keeping more stuff in memory, but an order of magnitude less accurate

[austere kestrel]

I wouldn't say an order of magnitude

it works pretty well tbh

but not for needle in haystack tests

[fast meteor]

upload screenshot of files is a workaround for short docs

[tough orchid]

I want to subscribe to Claude pro but the credit card not accepted

[austere kestrel]

yep, that's a clever approach (though just thinking about it.. if it's a short doc, wouldn't 32k be enough anyway..?)
(there also labs, if just happy to use Haiku and working with text)

[fast meteor]

if there is a surcharge that costs ~6x pro searches per query when working with claude opus that bypassing the limited contexts, this would be perfect for me

[feral marlin]

They’ve said before they do nothing special beyond capping file uploads at 32k tokens each

[torn ravine]

There’re many alternatives to Claude Pro that invoke Opus API and don’t cap usage.

[zealous kayak]

I... really can't see how that's possible, honestly

like, if they just cap every file at 32k
you could still fill up the 200k context

which would be crazy expensive for them

[feral marlin]

You could

[torn ravine]

Most without Internet access capability though

[zealous kayak]

the context window is 100% limited

otherwise they would be burning money

like the economics literally don't make sense
anthropic themselves can't provide unlimited messages with 200k context

[feral marlin]

They’re pretty large and get a lot of people to use opus they probably have a lot cheaper api access than others

[zealous kayak]

ok but if anthropic themselves can't do it
I cannot see anyone else being able to

[zealous kayak]

yeah, but they all cap context

I've also been trying to get a claude pro sub for that reason but, well, europe.

[torn ravine]

Some less known Chrome extensions have unlimited offer, which is very nice if one insists on complete context length

[feral marlin]

I use Claude pro but it’s about the same I just use interchangeably with perplexity since temperature is different

[torn ravine]

I think most of their users won’t bother switch to expensive models like Opus, so they manage to make ends meet.

[austere kestrel]

the costs are per token, for each input (message you send) and output (response the AI gives). You don't get charged for the LLM to just do what it is meant to do - start at the top of the context window and generate a completion (response) at the end

it's literally how they are meant to work

[zealous kayak]

You... literally do?
The more content there is in the context window (aka in the chat) the more you have to pay

that's literally how LLMs work

[austere kestrel]

openai does weird stuff with chatgpt; changing stuff dyamically

[austere kestrel]

yeah, for each message you send

[zealous kayak]

for example, if you run a local LLM, the more filled up the context window is, the longer it will take to generate

[austere kestrel]

each input (and response) accumulates in the context window...

[zealous kayak]

yes?

and you pay based on how filled up the context window is

[tough orchid]

which APIs are better than anthropic (excluding perplexity pro)

[austere kestrel]

no, you pay by the number of tokens in your prompt (input) and the response (output) per turn

[zealous kayak]

but your prompt isn't just the new message you send

your prompt is also the conversation history

like
I send a prompt that's 1k tokens in length.
AI sends back a response that's 1k tokens as well.
I send another message that's 1k tokens in length, and get another 1k length response back.

For that second message, I will be paying for 3k input tokens

the first input, the first output, and my second input.

that's how they work.

[fast meteor]

openai's Assistants api is stateful

[zealous kayak]

well, yeah - but you're still also paying for the convo history

they just do the conversation management themselves

rather than you having to do it in your API calls

[torn ravine]

You can try Monica for five days free. I literally fed nine or eleven chapters of a book into it and the summarization worked perfectly.

They don’t do RAG. When your input exceeds 200k, they just return error.

[tough orchid]

Can give me the link for monica

[torn ravine]

Sure. https://monica.im/home

You can notice (the slider) how long the conversation went until it reached limit.

[undone thorn]

[austere kestrel]

correct. But you don't get charged for the accumulation of those messages, you get charged for each message (and response)

Whatever the number of tokens for input and output per message, that is what you get charged for - but the LLM can still see the whole context window to make the completion...

the costs would be totally prohibitive if based on each prompt you sent and response you received was the totality of what's in the context window, not just the respective number of tokens in each for that particular turn

that would be insane

[agile jay]

Did the perplexity team remove the remaining message count on hover?

[torn ravine]

I also noticed this today. Maybe Pro search doesn't cost much so they decide to remove the cap?

[agile jay]

The cap is still there, just not shown.

"gpt4Limit": 580

So looks like the cap hasn't changed.

[fast meteor]

only shows when less than 100

[agile jay]

Wth do that?

[torn ravine]

I'm curious if anyone gets close to the cap. I use Pplx intensively, hardly reaching 400+

[agile jay]

Guess another thing I have to add to my extension.

Including the upload, image gen limit too.

[austere kestrel]

never came close

only so many hours in a day lol

[agile jay]

"createLimit": 50,
"uploadLimit": 999,
etc

"queryCount": 249,
"queryCountCopilot": 176,

Yep, those last 2 are the total number of prompts of my account...

[fast meteor]

when testing and tweaking the prompt manually,

[torn ravine]

Not sure. The usage count encourages me to ask more. Now the count is gone, I use it much less.

[agile jay]

So I use copilot around 70% of the time.

[fast meteor]

"queryCount": 5748, "queryCountCopilot": 2830,

[agile jay]

Same, also it was just nice to see it.

[agile jay]

Lol

How long have you had your account?

[fast meteor]

roughly 40 a day

since sept 2023

[agile jay]

Okay, makes more sense then

I barely use 10 a day... Since most of the tasks I do can't be completed yet with just AI.

[fast meteor]

this is why:

pplx is profitable per pro user
xd

[agile jay]

Yep, I also have a lot more AI services I'm subbed to.

There, fixed it

Removing features we like, and not adding ones we ask for, great job perplexity 👍

[austere kestrel]

it's been 600 since i've been using it, which ig is like a year now. meanwhile models have come / gone and pricing has changed

I don't think it represents the point at which they do or don't make money

it seems more like an arbritrary limit to prevent abuse

I find it so hard to believe the company is profitable yet (and there's nothing wrong with that - it's a start up)

[agile jay]

Depends on what deal they have, and how much the average PRO user uses it.

[austere kestrel]

it depends on a bunch of things

[agile jay]

Don't know their margins

[austere kestrel]

lol it's a private company why would you

it's not all about API usage...

they're a company

with devs, an HR department, marketers etc etc

[agile jay]

Yep, but they are a pretty small team currently

[austere kestrel]

i have no idea. but anyway it would be interesting to know which AI startup is profitable

[fast meteor]

i truly believe they would reconsider about the file uploading behaviour as they now accept/have more (big) cooperate users

[agile jay]

They don't even have a team/business plan yet...

[fast meteor]

not publicly👀 👀 👀

[austere kestrel]

yeah they're meant to be a search/answer engine..

[agile jay]

If it's not public, then it's a very small number

[austere kestrel]

what mistral is doing is intersting. on prem hosting

i think a lot of large entreprises would be attracted to that

[fast meteor]

if you digging deeper, you would find the answer👀 👀 👀

[austere kestrel]

being able to use LLMs on their internal data without a third party in the middle

[austere kestrel]

what is the enterprise use case / pitch exactly?

[agile jay]

Probably a lot more expensive

[austere kestrel]

honest question

[austere kestrel]

ofc

[agile jay]

Yep, the perplexity layout doesn't look that good for enterprise use.

[austere kestrel]

if you're a multinational bank or consultancy or whatever, the whole idea is that they would be willing to pay to host on prem and cut out openai or whoever

[agile jay]

Good news, they listened to my advice and fixed the thing which added the overflow to the html tag.

[agile jay]

Yep, but they also have to make sure their models are competitive though.

[austere kestrel]

this is why they have such a lean team 😂

[agile jay]

And that the models are more reliable.

[austere kestrel]

they just outsoure their work to discord ha

[agile jay]

It took a long time...I literally found the problem in a few minutes.

[austere kestrel]

yeah for sure, agree - that'd be the trade-off/consideration

like the on prem solution only works if the llm is actually pretty decent (ideally SOTA, but mistral isn't there yet)

[agile jay]

Would really only make sense if you were gonna finetune it with your own data.

[austere kestrel]

not really

just basic RAG

think about how much data massive corporations that have been aroudn for decades would have

[agile jay]

Yep, but their data tends to not be the most organised, lol

[austere kestrel]

it wouldn't be about fine tuning a model to be a great investment banker

[austere kestrel]

which is what LLMs strive in

doesn't need to be structured, like in the same sense a db has to be

[agile jay]

Yep, but you need to know where the data is, to add it to RAG...

Which is most likely gonna be in email attachments, lol

[austere kestrel]

yeah ig i'm confusing structured with organised

but anyway, regardless of the exact implementation, I don't think the appeal would be fine tuning a model (they can do that already; pretty sure JP Morgan or Goldman have one which is pretty well known)

[agile jay]

It depends on the type of business.

The main problem I would have is with current models still hallucinating.

So you need a way to check results automatically.

[austere kestrel]

yeah still need a human in the loop, esp when it comes to generative stuff

but for retrieval, i see it differenly. Like I think it can perform poorly (return 5 relevant docs when there's really 10, or return non-relevant docs), but there's less scope for actual hallucination.. like it's retrieving stuff - either well or poorly ig ha

[tight prism]

Me

[agile jay]

Sometimes it just makes up the files

Finetuning it for RAG would probably help

[austere kestrel]

I have never had Gemini give me a link for a file in Drive that doesn't actually exist

[agile jay]

They might do link validation

[austere kestrel]

well whatever they do, it works

and I assume that would be the kind of approach a large company would take

[agile jay]

Yep but drive data is more uniform, compared to a large volume of random files.

Would probably be better to reformat all the data, to make it normalised

[worldly tinsel]

Hello, it is normal that when using perplexy when I am in a chat, many times I ask him related things that I asked him before and he answers me with things that have nothing to do with it, is this normal? For example, I told him about a player's career in a soccer game (pes 2021), and after a while I spoke to him about the same thing, and he sent me a message in English (I'm Spanish) looking for information about pes 2021, and he didn't respond. to the thread in question.

Do you understand my question? Any solution?

[fast meteor]

data security law compliance

[agile jay]

Yep, if you are in search mode.

[mental widget]

Hello, does anyone know whether it is possible to disable AI learning for my data input (for privacy reason) when I use/buy the pro version

[mental widget]

Thank you very much, I was not able to find the information. Maybe I am blind. Where did you find it ?

[fast meteor]

settings account

[halcyon coral]

https://www.perplexity.ai/settings/account

[mental widget]

Thank you very very much

[worldly tinsel]

how i change mode to search mode?

[agile jay]

what is the focus mode? You normally choose it at the bottom left when starting a new chat?

[worldly tinsel]

I saw that, when starting a chat it does let me, but having it already started it doesn't let me? That's a serious mistake, isn't it? Well, I might want to change the search type in the entire chat (PS: could it be because the chat is in a thread?)

[agile jay]

You can't change modes while already in a chat.

But if you were in search mode, it's pretty nomal for it to ignore your past messages, since it's designed more for search.

You probably wanna use writing mode, if that is not what you want.

[worldly tinsel]

I understand, I will copy my message and put it in a new chat, but without a search mode, thank you very much man!

[agile jay]

np

[worldly tinsel]

and last question, what mode is for not search in internet?

[agile jay]

Writing mode

[worldly tinsel]

nicee

thx you!

[agile jay]

Writing mode is like normal Claude/ChatGPT

[worldly tinsel]

aah ok then nicee

thanks you!

[torn ravine]

Yeah I also found this. Hopefully there could be something in between, something like ChatGPT Plus, which can access the Internet while staying context-aware.

[agile jay]

Yep, or just a dynamic mode, which just does what you tell it to do.

[mental widget]

Hello, does anyone know how I can change the verification from email to password ?

[agile jay]

You should really use a 0auth2 login, like Google login

[mental widget]

But I a already used my mail address for authentication. Is their an option to add oauth to my mail address ?

[agile jay]

You would probably need to contact support

Or if you're a free user, you can just make a new account

[mental widget]

No pro

just 2 minutes ago

[agile jay]

In pro, there is a contact support button in the settings

Try that, since pro users have access to the tech support

[mental widget]

I will do. Thank you very much

[fast meteor]

didnt you get the email to sign in?

sometimes the email delivery can be delayed

[mental widget]

I also got this information. Thank you

I have one last question: I am able to see the pro settings in my account but I also see that the number of pro requests are 0

Does the payment information need time ?

[agile jay]

It's strange if it's zero

I'm pretty sure that free users even have access to it.

You could try #1140622008086970420

[mental widget]

I had access to it then asked around 5 question and than this happened

[agile jay]

Looks like the preplexity devs are currently offline

But if you post in bug-report it should get solved

[mental widget]

I will do, thank you very much

Login / Logoff solved the issue

[tawdry roost]

why the hell it keeps the old sources in the convo when am asking a diff question
and because of that the response is mostly wrong as its considering all the sources at once

I always have to remove the sources (min 2 times) to get the actual response

🤨

And if u ask another ques or prompt, all the sources are back again 🤦‍♂️

[rigid kestrel]

Hi how can you see that number?

[agile jay]

You need mad skills

[devout cargo]

Hola .
Un artículo de revista Forbes de México habla como ha ido transcendiendo Perplexity. El artículo en muy bueno lo dejo aquí .
https://www.forbes.com.mx/como-si-wikipedia-y-chatgpt-tuvieran-un-hijo-asi-es-la-startup-buzzy-ai-madre-de-perplexity-el-buscador-que-acabaria-con-google/

[solid whale]

¡El titulo! Jajaja.

[devout bronze]

does anyone else get connectivity issues sometimes?

i will say, i do use a vpn, but the page loads fine then says "Refresh: Connection Lost"

doesn't matter if i'm logged in or not

[agile jay]

Perplexity uses websockets, so the VPN seems to be affecting them somehow.

How are you using the VPN and what kind of connection is it?

TCP/UDP/Wireguard/OpenVPN etc

[devout bronze]

worked fine like 2 weeks ago lmao

[devout bronze]

wiregaurd on mullvad servers

i checked the network logs on DevTools and it shows 403s

from cloudflare

[agile jay]

On windows?

[devout cargo]

Encantado de ver gente de habla en Español j

[devout bronze]

which is odd because my browser is sending the cloudflare clearance cookies after solving the captchas

[devout bronze]

linux

firefox

i disabled uBlock for it too no difference

[agile jay]

Through a dekstop client or an extension?

[devout bronze]

website

[agile jay]

I meant the vpn connection

[devout bronze]

sorry

desktop

[solid whale]

Hablo un poco español solamente, pero es sufficiente.

[agile jay]

Since it's 403, I would guess that the IP is blocked by cloudflare

Try another VPN location

[devout cargo]

Riao una pregunta. Cuando hablas a Perplexity en Español te responde en el mismo idioma?

A mí no

[devout bronze]

but the initial page loads fine?
also on my own websites with cloudflare waf it's not blocked (although they do use cf enterprise so i wouldn't be suprised)

[agile jay]

In Pro, yes

[devout bronze]

i'll try to switch servers

[agile jay]

Sometimes in free mode, it doesn't

[devout bronze]

well i guess that worked 🤷‍♂️

would be nice to have a more descriptive error than just Refresh: Lost Connection

[agile jay]

Did it?

[devout bronze]

yeah, switched locations

[solid whale]

Puede, pero normalmente uso inglés con eso.

[agile jay]

Mullvlad normally doesn't get hit, but I guess you were unlucky

I normally use mullvlad through tailscale..

[devout cargo]

Ice un test de idiomas devuelve la respuesta en inglés

[devout cargo]

Since March he has been having problems with the language

[devout bronze]

might've been the datacenter, the ones i tried were denver and ashburn but seattle is working alright

[agile jay]

¿Estás en el modo libre?

[devout cargo]

Si

Espera modo libre que te refieres ?

[agile jay]

A veces, el modo gratuito responde en inglés, porque el modelo gratuito se ajustó con ejemplos en inglés.

[agile jay]

Si eres un usuario gratuito, utiliza el modelo gratuito.

Para los usuarios profesionales, pueden elegir, y los otros modelos como Claude y GPT4Turbo no están ajustados.

[devout cargo]

Siempre uso el modelo gratuito tengo configurado todo en Español

[solid whale]

Uso “Claude 3 Opus”.

[devout cargo]

Yo uso el que pone Perplexity

[agile jay]

La opción de idioma en la configuración simplemente reemplaza las palabras en inglés por español. El modelo en sí y el mensaje del sistema no cambian.

[devout cargo]

Antes no pasaba esto 🙁

[solid whale]

Ese modelo probablemente no entiende muy bien el español. Necesitas usar uno que sea más inteligente.

[devout cargo]

El modelo ere GPT 3.5

[agile jay]

Es un problema bastante común para los usuarios que no hablan inglés. ¿Quizás puedas intentar decirle que solo responda en español usando una "colección"?

[agile jay]

Es un ajuste fino de GPT3.5 que está capacitado para utilizar fuentes.

Not sure if that is how you say sources properly in Spanish

Hacer una colección como esa probablemente funcione.

[solid whale]

Empezaron a usar su propio modelo, que es una versión modificada de un modelo de código abierto. Es más barato para ellos. Creo que solo lo enseñaron en inglés.

[devout cargo]

Bufff eso explica todo

[agile jay]

Parece que está funcionando para mí

[devout cargo]

En Pro Search usa GPT-4

[agile jay]

Me aseguré de cambiarlo al modelo gratuito.

Parece que definitivamente está funcionando.

[devout cargo]

Jajaja ahora que hablo con vosotros va bien

Llevo dos semanas hablando en español y me contesta en inglés ahora que hablo con vosotros parece que la cosa ha cambiado no sé por qué

[agile jay]

Soy simplemente mágico, supongo.

[devout cargo]

Prueba a poner Hospital la fe

Te saldrá en inglés

[devout cargo]

😯😯

Ahh en pro Search sii

[agile jay]

¿Hiciste la colección?

[devout cargo]

Sii

[agile jay]

No estoy usando la búsqueda profesional

Me cambié al modelo gratuito.

[devout cargo]

Pero en búsqueda rápida?

[agile jay]

Se nota porque el nombre del modelo no está ahí.

¿Quieres decir sin el interruptor del copiloto?

La misma cosa.

[devout cargo]

Como lo has hecho

[agile jay]

Estoy usando una colección

[devout cargo]

Creo una colección y sale

[agile jay]

Como esto

[agile jay]

¿Escribiste lo que yo hice?

[devout cargo]

Voy hacerlo ya

[agile jay]

Lo mantuve simple y di la orden en inglés.

Porque el modelo entiende mejor el inglés.

[devout cargo]

Correcto funciona

[agile jay]

👍

[devout cargo]

En el perfil de IA lo tengo todo configurado en Español

También

[agile jay]

Ojalá eso lo solucione

[devout cargo]

Antes así funciona perfectamente

[agile jay]

¿Esto fue antes o después de que hicieras la colección?

[devout cargo]

Eso es antes de hacer la colección

Con la colección si responde en Español

Code vente de Beta a Android en Perplexity hace falta gente así 🙂

[agile jay]

Jajaja, no hay problema

[devout cargo]

Tú si tienes Android también está muy bien tenerte 🙂 de beta en Perplexity

@signal hamlet
Hello Alex, the colleagues have come up with a solution for the problem of the Spanish language of the AI that answers in English, so it works perfectly. We are talking about the free fast search model. I think the problem is that the model does not integrate well with the AI profile that is the response preference configuration
Code has found this solution

[tranquil orchid]

Hey @devout cargo!

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

[agile jay]

@remote frost Did you figure it out?

[tepid pilot]

anyone tried you search engine? how does it compare?

[weary crescent]

What happened here?

[agile jay]

It's the RLHF kicking in...

[weary crescent]

why does it think I want it to become bri'ish

[devout bronze]

might be the content inside of your file

[weary crescent]

OH you're right. in the transcript somebody jokingly said he should be british for literally a few seconds. lmaoo

[devout bronze]

lmao

[agile jay]

...

Prompt injection successful...

[devout bronze]

you might be able to modify the prompt to make sure it doesn't follow the file as instructions

[weary crescent]

nah because it was a 3 hour meeting and I completely forgot about that part

gpt4 does it fine, just gonna work with that for now lol

[strong tiger]

Does anyone know if you're upgrading from Raycast Pro to Raycast Pro + AI if you get the free six months of Perplexity Pro? I can't seem to find an answer to that, but if upgrading Raycast to include AI gets me six months of Perplexity I'll probably end up doing that.

[agile jay]

You should probably ask someone like Alex, when he's online

[agile jay]

@signal hamlet I assume he should message you guys, if he doesn't get an email after upgrading?

[fluid mauve]

I just had a thought whilst contemplating why blood from a nose bleed tissue sinks to the bottom of the toilet. I thought I would ask perplexity the question but it then occurred to me that I might not 'believe' what Perplexity tells me but I would believe it if Google told me.

This thought revelation, to me, makes me think that although I love everything Perplexity and other AI vendors are doing in the generative AI space, there is still a 'credibility' gap. I believe People 'trust' Google because it has always been here, and has been a pretty trustworthy companion. AI, being the young up and coming challenger, will have to do more to reduce 'hallucinations' (fake/incorrect) to reduce this credibility gap.

Just my two cents on a Sunday morning

[devout cargo]

Hola esto es Google SGE ?

[fluid mauve]

Obviously Perplexity providing cited sources is a good start

[umbral ember]

it is one part of the Google SGE, yes.

[devout cargo]

Aquí Google SGE solamente funciona para dar información de animales . Gémini está muy capado

Aún alucina

Microsoft nos actualizo la interfaz en España los primeros días podría poner 4000 tokens subir PDF de tamaño considerable por un error . Luego lo bajaron otra vez a 2000 tokens sin PDF .

He notado que han mejorado las búsquedas tiene acceso a muchos más contenido

Las búsquedas se las disputan entre Perplexity y Copilot .
Perplexity sigue teniendo más rango de búsqueda . Pero parece ser que Microsoft con Open AI está desarrollando un buscador conversaciónal también . https://www.lavanguardia.com/andro4all/tecnologia/openai-planea-lanzar-un-buscador-para-competir-con-google

[umbral ember]

Almost certainly - multi-modal interaction with a language model which can reason across modalities, apply domain-specific functions or agents, and which is grounded in both historical/archive as well as real-time Internet data sources, is probably the next step.

[devout cargo]

GPT-4 , Gémini 1.5 Mixtral ya funcionan así

[fluid mauve]

@signal hamlet good news about Raycast. Question I am sure I and any rabbit purchasers have is 'if I take out a Raycast subscription will the free 3 or 6 months perplexity pro access be tacked on the end of our already enabled 1 year perplexity pro access through the Rabbit deal?'

[dim ginkgo]

Real?

[fluid mauve]

Define real? It is a legitimate site. Like most things you pay your money and take your chances.

No real difference with what I do with perplexity. The bonus with perplexity is the community.

[dim ginkgo]

I found the claude 3 Opus API very expensive, but YouPro allows unlimited queries, which makes me suspicious. Or maybe its context window is 32k instead of 200k

[devout cargo]

Antes hablamos de funciones multimodales , me acaba de llegar una actualización que combina texto y imagen esto lo hacen posible por los enlaces de la imagenes y la compatibilidad de la interfaz GT-4 es capas de escribir el. Contexto con sus ilustraciones ordenadamente
WoW

👌

[torn sierra]

They have 200k context window. You may check their Discord serwer - it’s officially said

[torn sierra]

They have many things that differ them from Perplexity making them potentially a better alternative (like problem solving chat, creative chat etc.). Perplexity is mainly focused on searches. But when it comes to making searches Perplexity is the best option available.

[drifting raptor]

Hi! A few questions) The daily request limit is gone, is it really unlimited? Or a marketing ploy? When will gemeni 1.5 be released?

[serene sand]

#💬│general message

https://www.perplexity.ai/hub/faq/what-is-pro-search

[tame current]

genius mode is so useless

but I'm on the free version so maybe that's why

or maybe it's just that idk how to use it

[buoyant kindle]

I don't understand why these platforms have these gimmicks. Just stick to the core concept and make it work well.

[tame current]

agreed

some have so many features even though they don't work well at all

[sly charm]

Hi, why can't I enter a discount code when I create a new account and sign up for the pro plan?

[boreal idol]

How do pro users feel about how many pro searches i have left each day

I only can see cmd + .

[serene sand]

#1227595427373776906 message

[young dew]

drink water

[serene sand]

the documentation and the application experience constant change very often

[cinder comet]

can i use perplexity pro ( claude opus ) without the search?

someone help this poor soul

[fast meteor]

in the writing mode

[wicked zodiac]

Change Focus to Writing

[cinder comet]

whats the context limit?

+cap usage per day

[fast meteor]

currently 600 every 24hours interval. at least 32k tokens. img gen is 50/24h

[cinder comet]

wait

thats actually quite good

[tame current]

anyone tried https://www.cognosys.ai/ ?

seems pretty good

it's like perplexity

[cinder comet]

this is different from perplexity

it automates your workflow

[cinder comet]

hmm 1000 messages per month

gonna go for perplexity ig

@torpid thorn uhm, does the cap limit shows at the bottom or smth?

ive seen it somewhere, but i dont have it

[fast meteor]

only shows when less than 100
when mouse over the pro toggle switch

[cinder comet]

hmm i see

[lavish quarry]

How do I report an issue?

[fast meteor]

#1111765634267742408 #1140622008086970420 and [e-mail]

[lavish quarry]

Cheers

[tame current]

not really, sure it has some features like that but the core is same, LLM + web search. it even has the clarification questions like perplexity

also I wonder where meta got the inspiration from for their AI Chat 😛

[cinder comet]

https://docs.perplexity.ai/docs/rate-limits

this needs to be updated

[cinder comet]

oh

is this new?

[tame current]

yes, it's in whatsapp and rolling out in only some countries but I haven't gotten it

[cinder comet]

so its based on llama 2

[tame current]

yes

I think so atleast, they don't say which one specifically

but it's definitely llama

[lone citrus]

I saw someone posted in reddit, the context limit is over 32k but you have type all texts instead of upload documents, for somehow reazons pplx don't want user to use long context for documents

[zealous kayak]

money

(I mean that literally, it's expensive as hell to have convos with long context)

for uploaded documents, RAG is an order of magnitude cheaper and works well enough in most cases

[torn sierra]

For now as I believe there’s nothing better than Perplexity. I tried many services and found nothing better. Of course there are some things that I dont like but I can set them aside. At least for now 😃

[lone citrus]

agree

compare to poe, the claude or Gpt4 in perplexity is so cheap

acceptable limit

[tame current]

so much more accurate too

[dim ginkgo]

I would like to ask if the recently upgraded GPT 4 has been integrated into Perplexity?

[dim ginkgo]

can you give me the link?

[torn sierra]

Sure. #958041026415718414 message

[dim ginkgo]

Tks

[ripe void]

What is raycast?

[wise granite]

yooo, can perplexity read aloud on the pc version?

i have pro too

[tame current]

https://retrododo.com/first-official-emulator-igba-now-on-apple-app-store-to-download/

[warm cave]

rn its only on mobile

[lone citrus]

I belive the you.com have their own LLM

they already devoloping their own llm for a while, so it's make sence

[livid mantle]

😂

[warm cave]

lol, posted the wong place 😆

[tame current]

Cognosys seems to find difficulty reading some PDFs that Perplexity read.

I do not think it's that good, but it's too early to say.

[hasty musk]

Hey guys, where is can see how much queries i have left?

i *

[gaunt crater]

What is the Claude Opus context window (as in message history) for Perplexity? Is the 32K input size inclusive of file uploads? I assume the input is the same as Phind but I am finding it hard to compare services on the window size. Leaning towards Phind because of the extension and the better web search but having to choose between one or the other service this month.

[agile jay]

They changed it so it only shows the number when you have less than 100 left.

[hasty musk]

Its 600 per day yes?

[agile jay]

Yep

[tepid pilot]

whe are we getting gemini back?

[agile jay]

Never

[topaz eagle]

Anyone know how to get rate limit increases for the API?

[boreal tiger]

hmm has anyone been facing trouble with opus ?

after a few messages it starts hallucinating or answering random questions

[shut dagger]

is perplexity pro now using latest GPT 4 Turbo model? non preview variant?

[halcyon coral]

#🧪│api-general message

[warm cave]

@south kindle Scratchpad has been really big for me, supper helpful, I can use it as a collection and upload my class notes as a PDF and its way better at helping me with problems I am stuck on compared to Claude 3 without Scratchpad. Thank you! 🙏 🙂

[rugged epoch]

https://vm.tiktok.com/ZGeuW63qm/

https://vm.tiktok.com/ZGeuWJMes/

[gaunt crater]

#💬│general message
Still trying to get an answer on the Opus context window. 😕 I realize they may want to reserve the right to adjust it later but it'd be nice to know.

[shut dagger]

It's 32K

that's what they told me in pro support

[halcyon coral]

Please check: #👑│pro-lounge message
But the context size is completely irrelevant in 99% of cases. Firstly, the context is larger than you think and secondly, RAG is used for file uploads anyway, which results in a MUCH larger context ☺️

[shut dagger]

no access channeL?

[halcyon coral]

Pro channel

[shut dagger]

how to get access?

I got pro

[halcyon coral]

1. Leave the server
2. Join with the Pro Discord link at https://www.perplexity.ai/settings/account
   

[shut dagger]

oh okkk

[torn sierra]

I uploaded a file that has one hundred pages of text and it didn't capture the full context, it was something around 30k. For most applications it's enough, and in ChatGPT Plus / Team there isn't more anyway 😉 However, it's no surprise that everyone would like to have more and have access to 200k right away 😅

[shut dagger]

Got access, thanks man!

[torn cape]

hello

[torn sierra]

Must do the same 🙂 Thnx

[halcyon coral]

RAG is not there to memorize the whole document. RAG is there to find relevant parts of the document for your question and give them to the LLM. And a larger context of 200k, for example, is not possible (maybe with fewer uses...), the costs are far too high.

[dawn junco]

hmm

[torn cape]

are u guys ai + crypto project or only AI

[halcyon coral]

Perplexity AI is an AI search engine (answer engine)

[warm cave]

most of the time RAG is great, but sometimes the whole document is relevant, Like yesterday I had a transcript that was 160 pages for a class, It was the combined transcripts of 10 or so classes, and I asked It to make a cheat sheet/formula sheet. you.com did better than perplexity, but you.com also stoped working after a couple of messages, so most of the time RAG is great, but occasionally it lacks

[south kindle]

I’ve been having issues lately with formatting using it. Trying to get it away from the giant wall of text mashed together. All the info is still there but it’s a mess to look at on mobile.

[warm cave]

I don’t use mobile often, I mostly use web and your original prompt has been working well. I have made changes for specific uses, like telling it to never simplify math operations or approximate. this has helped overcome some of the accuracy issues

[south kindle]

OG prompt being which one

[warm cave]

this one, top of the discussion thread

Honestly just have not had the time to read through the whole disscussion to see what changes i want to make

#1223058316662538331 message

[torn sierra]

Thnx. So if I understand correctly, what we provide to the model (e.g. a file) is treated as a knowledge base, and the model does not process the entire context, but rather RAG selects the relevant parts? And for this reason, the context window appears to be larger than it actually is?

[halcyon coral]

That was just a simple explanation, but yes, this saves costs on the one hand and avoids errors due to an overloaded context on the other. It is also faster.

[torn sierra]

that's simply how I understood that RAG more or less works. I have no idea about the exact details of how it functions. It makes sense. Thank you for the explanation ☺️

[agile jay]

Basically uses embeddings to match the semantic meaning of the text.

[torn sierra]

I understand it in a way that RAG is coupled with LLM and provides additional info, which fills the context, and only then LLM works on it. Am I right?

[agile jay]

LLM's and embeddings are different things. Even though embeddings are generated during the training process of LLM's.

[granite quartz]

Is RAG always used for files with Perplexity? Or only when context goes above ~30k tokens?

[torn sierra]

the LLM model you are using is an integral part of the RAG system, which means that every query you make must pass through the retrieval module. From what I understand, the query is first converted into embeddings. These embeddings are then used to search for relevant data (for example, in your file, which is also converted into embeddings). The retrieved data is subsequently fed into the LLM, which generates the final response 😀

[agile jay]

the LLM model you are using is an integral part of the RAG system
This part is kind of wrong. You can pretty much use any embeddings for RAG, doesn't have to match the LLM you are using.

[torn sierra]

Sure, from what I understand, we have our query, which we convert into embeddings. We then perform a matching process to find relevant information. After that, the entire query along with the retrieved context is passed to the LLM (at the very end). That's essentially what I was trying to convey. But of course, if I'm stating anything incorrectly, please feel free to correct me 🙂 Just a few hours ago, I had no idea at all how this whole process worked 😄

[young dew]

drink water

[tepid pilot]

why not?

[tepid pilot]

what's scratchpad?

[devout bronze]

genuine question, Perplexity or You.com? i have pro on both but eventually am going to cancel one of them because i don’t want to pay for 2

[agile jay]

You have both, so shouldn't you know?

[icy zenith]

copilot pro is the obvious choice 🙄 /j

[agile jay]

Too lobotomized

[icy zenith]

not when you pay

also turbo fixed that up anyways imo

[agile jay]

Out of all the models, copilot pro and Gemini ultra are pretty bad at following commands.

[timber eagle]

Hey folks, what’s the best model to use for academic purposes 🙂

[warm cave]

Like research? or like personal tutor?

[timber eagle]

Research.

I’m currently obsessed w Claude opus 3. Love the references

[warm cave]

I would probably use Claude 3, very smart and great context window, and if you need specific formatting, give it to GPT-4 for that, somtimes GPT-4 is beter at following commands to format in a specific way, but default go to I would say Claude 3

And what is great about perplexity is if you don’t like the output, there is a rewrite button so you can try GPT-4, Mistral Large, etc

[timber eagle]

Yes agreed.
The thing with GPT 4 is sometime. It doesn’t give the actual information what I need compare to other models. I’ve tried.

Earlier I was using only GPT PRO

also it sucks with providing sources

[warm cave]

Yeah, I sometimes copy and past stuff into GPT because I need something very specific, Like give in Latex, or use "Blank" format citations, and its just consistent at that, but for the bulk of it i use Claude 3 Opus

[timber eagle]

Ah yes!! Bummer if using for university 😤

[warm cave]

Like Claude 3 was having a hard time giving something to me in the format of a wikipedia article, yesterday I randomly made a Wikipedia article bc a youtube video said that there was no wiki on the topic, and GPT got the formatting right first try. have no idea if it will get approved, just did it on a whim using a AI, haha

[timber eagle]

Oh, that’s really great. I mean it should be approved. Looks very legit to me.

What are the special keywords you use with claude 3, or you have learnt overtime because I guess using these AI is not that easy as you think of it is.

[warm cave]

Im not sure, Its hard to think what I do specifically, I just work back and forth with AI using various prompts until I get a result I like, sometimes copy and pasting. This was done Research done with Perplexity, Microsoft Copilot, and You.com. I then gave the full research to Perplexity with Claude 3 to write the Wikipedia article with a few back and forth , and then finally perplexity using GPT-4 to get the formatting to match Wikipedia formatting.

Mostly I just have a feel for how to get the AI to do what I want with prompting, and which AI is best for the task. I am a big fan a Chain or Though, Tree of thought, Scratchpad. also having the AI "role-play" as an expert in the task. Like having it take the personality of hank green to explain a biology topics.

[warm cave]

Any specific things you have learned that are helpful?

[devout bronze]

i do, but they both have their strengths/weaknesses and i wanted to just get some opinions

[warm cave]

I just got you.com yesterday and have had perplexity for about a year now so I know a lot more about plex, I really like plex, and If I could only chose 1 I would do plex, Great search, different focus modes for web, writing, etc. re-write with different models. Also having Collections where you can have system prompts are invaluable, the only thing it lacks it code interpreter and context length. Tho you can get pretty decent code interpreter for free with Microsoft copilot. If you really need a lot of context like over 100k then thats the only reason I would way look elsewhere. I got you.com bc I had a long context use case and as a student its $10 a month, and I am trying to get rid of chatGPT, chatGPT i only pay for because it has code interpreter for math problems and other problem solving. but $40 a month on AI is killing me, haha. I will just use perplexity if free code interpeter with MS copilot is good enough, otherwise i will see if you.com genius mode is any good

what are you thoughts on it?

[devout bronze]

so you're saying plex is better, besides

• no code interperater
• short context length
  ?

[warm cave]

yeah, sorry long message

[devout bronze]

i do agree collections are very nice

but having the code interperater is amazing for my programming needs

@grand mauve sorry for ping but will this ever happen?

[agile jay]

Code interpreter only runs python code...

[warm cave]

yeah, i have a lot of system prompts for lots of tasks, would strugle if i lost them

[devout bronze]

and i do appreciate the long context window of you.com

i'm using it rn to summarize some AI research papers

i would say the pplx seems more like a research tool and you.com seems more like a chatbot though

[warm cave]

Yeah, I have been using the long context, and Gemini 1.5 and Claude 3 Opus is very nice, tho I have been geting erros often

[warm cave]

Yeah, you.com really has the 2 selling points over plex in my eyes, but plex has a lot more. so I cant say one is better, its really the use case