💬│general

[agile jay]

NineInchNails is enough

But normally I have it related to the site.

[fluid mauve]

password entropy,

[agile jay]

No need for it when it's being hashed.

[fluid mauve]

so you hash your plain text password and remember the hash?

[agile jay]

Don't need to remember it.

I just remember the password and have an extension which autohashes it.

[fluid mauve]

nice

[devout geyser]

unless I'm mistaken Opus is capped at 50 indeed, I just sent two messages for Opus a moment ago and now I'm at 48

[agile jay]

Yep it is

[livid mantle]

????

silent update?

HAHAHAH

it doesnt say unlmiited or 35

we will have to find out

someone test already?

ping me thanks

[thorny canopy]

Capped 50

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

[livid mantle]

wow that sucks

we want unlimited

maybe its a typo they meant to make it 500

missing. an extra 0 there

[fluid mauve]

I reckon they are stair-stepping it up in a controlled manner to make sure there is no reoccurance of the abuse where they have to pull us all lower again. Granted better coms would be good, but as others have said, other platforms seem to be having similar issues and have also reduced/cut opus connectivity. Sucks, but I get it from an engineering point of view. Maybe not comms.

[glad pilot]

Maybe if there is still a cap they can make it visible as the 30 cap. When I check my account setting I thought the cap is lifted but then I came to dc people told me there is still one

😂

[fluid mauve]

now im being blocked posting 'constructive criticism

[vernal remnant]

are y’all sure it’s still capped?

did someone test the 50 messages?

[tame current]

It's a start for sure. But, they need to constantly raise the cap back up. After all, they said it was a temporary measure so, they need to stick to their word.

Here we go. That says it will go back up to 600.

[tame current]

you dont use a manager?

i use bitwarden

also finally pplx doing something good, if they limited it to 50 or 100 none of the backlash would have happened

W for us

NordPass is very very good.

You also get email masking included.

[rough flower]

No 2FA

[tame current]

its paid

bitwarden is free and opensource

[pseudo valve]

Hi

[keen pendant]

Hi there,

I upgraded to copy.ai pro plan. How to claim the 6 months pro of Perplexity ?

[wide mesa]

It still would have happened honestly

[tame current]

shrug

[wide mesa]

Also, when perplexity cites sources are those like, really old sources ?

[warm cave]

Yes they are, but AI still hallucinates some details, so it its important than I recommend you double check by visiting the links

but for most things it will be right since it is giving information from the web

[wide mesa]

Is it capable of citing stuff published recently? (as recent as 2023-2024)?

[warm cave]

I think so, i believe i have used it by asking questions about recent papers that were on arxiv only a few days old and it worked

[wide mesa]

Actually ye it just showed me a reference from 2023. neat

wtf it listed something posted from 1991 as 2023 🙄

[warm cave]

lol, hallucinations

I think that there is an operator that you can use, let me see if i can find it

[wide mesa]

id really appreciate that

[warm cave]

No problem, this works for normal searches, and i think its working for academic, but give it a try and see if it works

after:xxxx-xx-xx

[wide mesa]

Thx thx I'll test it out in a bit

[foggy whale]

I got nudged out of Claude Pro pretty quickly too. may be an upstream issue?

anyway, guess it makes me thrifty with my prompts

[arctic wharf]

Looks good

[small hemlock]

Just tried coding an indicator for mt4 in Perplexity and it did it perfectly. Great stuff

[keen pendant]

ib me

[agile jay]

Pretty sure you are not allowed to advertise that here.

[sweet jasper]

Nothing users monitizing hard

[cyan ice]

This was Genuine. got my premium.

Thanks @foggy parrot

[livid mantle]

@reef cave

[arctic wharf]

Btw have you guys tried llama3?

I really like it

What do you think?

[tame current]

It's great.

[pine valve]

It is really fast, especially the 8b model, like lightning.

[cinder comet]

whats the deal again?

is it limited again?

mine doesnt show anything

[serene sand]

#👉│feedback-general message

#💗│sharing message

[devout geyser]

yes it's limited to 50 per day at the moment, and they've hidden the number until you get to 10 or less I believe

[loud garden]

Нарисуй мир

[serene sand]

https://www.perplexity.ai/

#1194794305362071552 message

[loud garden]

СПС👍😋

[north orchid]

No trial?

[frank hill]

can't login to perplixity in iOS App

[reef cave]

Eh

[golden lark]

When I use writing, I upload some files. How can I write separately from the files?

[glad pilot]

yeah, they raise the limit to 50 and remove the counter, and it will only show up when you got 10 left

[tame current]

@golden lark

[left gazelle]

did they rlly remove 7 day free trial

[tame current]

Hmm. I've noticed that they now do not show the model selected for pro searches. Possibly so you don't notice it going from Opus to Sonnet?

Or a bug.

[cinder comet]

still shows it for me

[tame current]

It does if I specifically click to change models, after an answer.

Web or mobile?

[cinder comet]

web

[tame current]

Android here.

[cinder comet]

also i have a script installed that lets me chose the model from the go

[tame current]

Perhaps it's a mobile only thing.

[cinder comet]

maybe

[tame current]

Which I imagine is a heavy proportion of users.

Hoping it's a bug. Cause that would be sneaky.

[left gazelle]

who can share perplexity pro with me 🙂

[cinder comet]

you seem rich

[left gazelle]

nope 🙂

broke af

[tame current]

That would be explicity against the ToS.

And quite bold asking in the official Discord.

[livid aurora]

Yea

[cinder comet]

try codellama3

[left gazelle]

bruh whos gonna check if i am sharing an account with someone

[livid aurora]

where

[tame current]

Apparently they are hot on that at the moment.

However, very very bold to ask here. I'll give you that!

[cinder comet]

🙂

🙂 🙂 🙂 🙂

[tame current]

To be expected!

[fluid mauve]

What is to be expected?

[wide mesa]

I really wish you could use the writing focus alongside the provided sources from the search function

What an odd comment to make man

oof it got deleted

[obtuse pagoda]

while we can’t use the opus normally, I want to know which revision of gpt4-turbo is used by Perplexity, any info?

[tame current]

Sounds juicy.

[tame current]

Lack of Opus messages.

[wide mesa]

they basically ranted about using you.com over perplexity and said perplexity wont last the next 6 months, citing EU law some such etc whilst pinging some perplexity folks

It was kinda unhinged

[tame current]

The You UI is terrible.

And buggy.

And looks horrible.

Well. Perplexity is funded by Bezos.

[wide mesa]

You UI just looks like default chatgpt tbh

[tame current]

I think they'll survive.

Chatgpt is also ugly.

[wide mesa]

nah man clearly they wont last the next 6 months because random users aid so

honestly a pathepatic attempt aat a scare tactic

[tame current]

I could see Amazon buying them out.

[wooden karma]

my first experience with You is that i literally cant click an X button lol

[wooden karma]

or i can click it if i click the actual icon but it looks like i should be able to click it outside the icon too

[wide mesa]

ew beggar

go ask your parents for their credit card

[tame current]

Get a job.

[wide mesa]

That or do your assigments yourself

There's a free trial, use it

[tame current]

What do you want to do with it anyway?

[wide mesa]

sucks to be you

[tame current]

To do what though?

To achieve what?

And I don't need to share my account.

[wide mesa]

He's gonna use it to generate AI NSFW haha

[tame current]

And wont.

[fast meteor]

here you go, free of charge

[wide mesa]

@tame current btw do you know whats the maximum length that Perplexity makes responses in?

[tame current]

I'm not sure.

[wide mesa]

Also they got banned lol

[tame current]

I know the context is 32,000 for Opus. But not response length.

Good.

The nerve of them.

[wide mesa]

I tried to have it do a 2500 response and it didnt really work

doesnt seem to reach much beyond 1500 without references

[tame current]

Why would someone beg in the official Discord?

Why did they think that would work?

[wide mesa]

wont know if you dont try ig

[tame current]

In ways, they will be celebrated as a visionary.

Historians will look back on them as a freedom fighter.

[wide mesa]

Truly one of all time

Perplexity with writing focus can get to 2640 words from my 2500 word promp neat

[tame current]

I think the Claude app can get much higher.

But. Higher context blah dr blah.

[wide mesa]

Tbf with writing I just told it to take my established text and entend it t 2500 words by elaborating on the given arguments and adding more detail as needed

[halcyon kite]

Are there plans for Llama3 70b to be integrated into Perplexity?

[livid mantle]

we have yet to have the latest gpt 4 turbo added

maybe it will be added togetger with Llama 4 400b

[orchid magnet]

How do I have pwrplexity make images

[livid mantle]

perplexity dont pay me to help or teach so I choose my right to not help you, im so sorry

but there are 4 voluntary community mods who love helping ppl, wait for them to assist you

disclaimer: it doesnt imply that voluntary community mods are getting paid either

[wraith crow]

the latest gpt4 is already added. #👑│pro-lounge message

[livid mantle]

oh sorry i just woke up Spezi

thanks

[wraith crow]

there is a guide on how to make images here : #1194794305362071552 message

[south kindle]

I think new gpt4 has been active for a few days

Uh oh I lost my roles 😦

[livid mantle]

you lost your orange role why

[south kindle]

I only have plexer

[livid mantle]

Ah

[sweet jasper]

You guys get funny roles?

[tame current]

yeah

[wide mesa]

Why the weird passive aggressive comment

[livid mantle]

hey moonbacon_, i wasn't trying to be passive aggressive, i was just stating my position and letting you know that there are other voluntary community mods who might be able to help you out. no need to take it personally!

[fluid mauve]

yeah, it wasn't very community focused

[wide mesa]

I noticed formal wording tends to get snagged on AI detectors often

[livid mantle]

WOMP WOMP

[wide mesa]

I do believe you are in the "nobody asked" territory

[winged moon]

And you had to make this comment?

[inland bear]

In March, I sent a message to the support email asking what the context window of Claude 3 Opus and Claude 3 Sonnet at Perplexity is. Here is the response that was sent to me: 200k.

[hasty musk]

It’s 32k not 200k, idk…

[cursive jacinth]

Hi sir, now i am really confused with the token limit sir.

[tranquil orchid]

By default Perplexity can read at least 4000 tokens per question

It can read many more with File Upload. Longer pasted text is converted to a file automatically.
File uploads can be read with a context window of at least 32000 tokens by GPT-4 or Claude 3 with a Pro subscription. To unlock this longer context length, please subscribe to Perplexity Pro.
Models can write up to 4000 tokens at a time, you can prompt them to continue in a follow-up.

[inland bear]

As soon as the Claude 3 Opus was launched on Perplexity with a daily limit of 5 questions, I subscribed to Poe.com and Phind.com for a month. With the return of a 600 daily questions limit at Perplexity, I ended up canceling both services.

I primarily work with legal documents (court cases), in PDF format, in Portuguese. Generally, these are lengthy files.

Phind.com does not allow (or at least did not allow) the upload of PDF files.

Poe.com did allow this and offered the possibility of multiple context windows (and, depending on them, the remaining usage points varied). Later on, they doubled the points required for using Opus in any context window.

In my experience, I received very similar responses using the Claude 3 Opus from both Perplexity and Poe.com.

The day before yesterday, given the new reduction in the number of questions, I ended up subscribing to You.com. Definitely, the answers I get there (asking the same questions and using the same PDF files), using the Claude 3 Opus, are much more concise than those I get on Perplexity, using the same model.

You.com impressed me in other respects, but not with the Claude 3 Opus.

I am not an expert in computer science and I may be talking nonsense, but according to the official response from Perplexity to my question and the comparison I made using the Claude 3 Opus from Poe.com, explicitly using 200k, at least for my use, I noticed no differences (and I observed a significant difference for the Claude 3 Opus from You.com).

[buoyant kindle]

My use case is very similar to yours.

So far, I've had the best experience using Claude 3 Haiku through Perplexity Labs. Which is weird, considering it's free.

[stuck ferry]

Oh

[tame current]

can someone please confirm this ?

it seems to me like perplexity used to make lists a lot, but now it rarely does

I'm talking about free version

[torn ravine]

In my case where I frequently use Opus to expedite studying new books, longer context window helps by providing more thorough answers that can only be obtained by reading the whole book. RAG doesn't bother me except this.

But for papers especially with many figures, I like Pplx. It passes relevant figures to the model while most competitions out there only pass in the extracted text. That makes a lot of difference.

[inland bear]

I've never used Perplexity Labs. If there's the possibility of uploading files, I'll stop by to take a look.

On Poe.com, there was the option of Claude 3 Haiku. I wasn't getting as comprehensive responses as when using Claude 3 Opus to analyze my PDF files, but without a doubt, the response time was infinitely faster and, for certain cases, such as a first read, it served my purposes well.

I requested in a message sent to Perplexity that the model be made available on Perplexy Pro. I admit I was unaware that it was already available at Perplexity Labs.

Don’t get me wrong: although I really want to see Claude 3 Haiku available on Perplexity Pro (with the capability to analyze PDF files), I am eagerly counting the hours until the limit of 600 daily questions is restored on Claude 3 Opus, as contracted.

My experience with Perplexity's support has been very good, and I would very much like them to resolve this question limitation issue as quickly as possible.

[buoyant kindle]

No, you can't upload, but you can just copy the text from any pdf and paste it into the prompt. It works very well, 100% of the times.
Opus or any other LLM in PPLX Pro cuts off the document but makes it appear like it has analyzed the entire document. It most certainly hasn't. Try asking it how the document ends. I've gotten responses from like 3 pages into a 100 page document...
You can also try it with a transcript from a podcast. The summary appears to be complete, but when you ask it how the interview ended, it responds that it appears to have ended abruptly in the middle of a conversation.

Gemini 1.5 Pro through Google Dev works very well even with 300k+ token documents.

[haughty lodge]

Yeah with the 1m context window but it is really slow (800k token of code took around 90sec to process)

[buoyant kindle]

Yes, but it's so damn good and useful that I'd wait 24 hours.

[stoic wave]

Damn, you must have so much free time.

[buoyant kindle]

I do. 😁

Using AI to have even more.

[inland bear]

Thank you very much for the tips! I will test them as soon as possible!

[fluid mauve]

I do love that you think 90 seconds seems a long time to analyse 800k tokens of code. How long do you think it would take to do it manually? Ai has opened up so much that we start to take the simple things for granted.

[buoyant kindle]

Nice! Let me know if you find anything better. We seem to be using AI in similar ways.

[torn sierra]

You.com can analyze long documents. The context window in you.com (Claude Opus) works fine.

[haughty lodge]

I think you got the wrong idea. I am not complaining 1.5 is a great model, but it's a lot slower than other models. It's a lot of text but 90 seconds before it even starts to write is quite long compared to the others.

And that this does not matter what you ask it as long as it has these context.

[buoyant kindle]

Right...

[honest basin]

ig they're still doing the fake error messages

once you send too much

[buoyant kindle]

Oh, wait... What... I refreshed the page and the summary showed up.

[coarse lark]

Will you explain to these people about how amazing truebit protocol is? It is.

How important it is to web three AI healthcare privacy and verifying it’s ahead of its time. I’m hoping Perplexity will break it down.

[agile jay]

Isn't that just blockchain stuff for running smart contracts and similar stuff.

[torn sierra]

Yup. It happens sometimes. From what I saw this error is already mentioned on their Discord server

[mighty fox]

When Grok-1 ?

[inland bear]

I just used the Claude 3 Haiku on the Perplexity Labs! It works just as well as the one from Poe! I copied and pasted a 37-page PDF document and it summarized it very efficiently in just two pages! I'm going to start using the service for my urgent readings! Thank you so much for the tip!

Unfortunately, Gemini 1.5 Pro is still not available through Google in Brazil (just like the Claude Pro is not available directly through Anthropic either).

I used Gemini 1.0 Ultra for two months and found it to be very weak (it still referred to itself as Bard...).

You.com offers Gemini 1.5 Pro and I've already noticed that it's much superior to the 1.0 Ultra. However, I gess You.com limits the window to 32k tokens.

As soon as it's available here, I'll definitely test Gemini 1.5 Pro with the 300k tokens!

Once again, thank you very much! If I find something better, I’ll let you know, for sure!

[buoyant kindle]

Great to hear!

As for region restrictions, I use and appreciate PIA VPN. 55 EUR/USD gets you like two years of subscription.

[upbeat tiger]

It's impossible it's that condition because it happened to me on the first try after not using 'you' for 1 month, yesterday :/
I have no idea what is causing that, maybe too many people using at the same time so it hits a quota.
I didn't even try to send a very long prompt. iirc it was about 100 lines.

[cinder comet]

oh the limit is 50 on claude opus

they'll agree on a number later, probably 100, and then it'll be standardized

the reason is system abuse obv

[inland bear]

I'll try it as soon as possible as well!

[fluid mauve]

Ahhh, that makes more sense.

[unreal otter]

I can’t input

I can’t click button

[serene sand]

known bug, try w/ shorter input

[south kindle]

oddly also only a bug with new threads

if you copy/paste that same chunk into an existing thread, it should work

[serene sand]

#1216589575107903588 message

[unreal otter]

I can‘t. He can’t operate and can‘t close it. Turning off the keyboard will make the text I entered disappear.

[crude spoke]

where to go if i have questions

[serene sand]

https://discord.com/channels/1047197230748151888/1118264005207793674

[crude spoke]

abt the app

ty

[serene sand]

restart

[unreal otter]

ok im sleep im china

Entering Chinese will cause some problems.

[serene sand]

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

[south kindle]

https://console.groq.com/playground?model=llama3-70b-8192

[warm cave]

This is amazing with Open WebUI

[south kindle]

how do you have that setup?

seems slick

[agile jay]

I'm guessing you add the API key, or something. Since it sounds like an open-source project.

A web UI for ollama models?

[warm cave]

Here is an easy setup guide: https://decoder.sh/videos/use-your-self_hosted-llm-anywhere-with-ollama-web-ui But you groq you just put in under openai api key and base url

[warm cave]

yes

[warm cave]

https://docs.openwebui.com

[agile jay]

I am guessing you're hosting the webserver on your PC, or something?

And accessing it through localhost on your phone?

[warm cave]

Yeah, 3 option, run it on the same computer, or use ngrok so you can use us from another device or lastly you can use an API key on the same device'

[agile jay]

If you only wanna use the API key, is there a url just for the client?

[warm cave]

I am hosting on a PC and have ollama and ngrok setup

[agile jay]

Or I still need to host the web server for the UI?

[warm cave]

Yeah, you still need a server for the UI

[agile jay]

Okay, guess I'll just wrap it in an APK, to get around that then.

[warm cave]

so you are making it into an app?

[devout cargo]

New Search Engine ?

[agile jay]

No, it's a self hosted client for ollama models.

[devout cargo]

View new Search Engine Open source

[tame current]

endpoint?

[warm cave]

soo many, haha

[agile jay]

It is ollama so

That's to be expected.

[warm cave]

Well those are the ones i have downloaded

i need to delete the ones i dont use

[agile jay]

Yep, since it's a TS project, just gonna wrap it into an APK.

[warm cave]

Nice

[devout cargo]

Delerte video im no problem to Perplexity

[warm cave]

btw, i guess i sent you the wrong link the other day, this is what i though i sent: https://github.com/developersdigest/llm-answer-engine

[tame current]

-0.01 💀

oh

thats their way of saying

[serene sand]

https://discord.com/channels/1047197230748151888/1059504969386037258

[tame current]

ur broke

[devout cargo]

Code mira esto
https://www.morphic.sh/

New Open source beta

[warm cave]

Thats interesting

nice, llama 3

[devout cargo]

No problema IA speak Multi-lenguajes

Speed IS GPT 3.5
QUALITY is GPT-4

No es Perplexity pero al ser abierto la comunidad lo está desarrollando muy rápido

[agile jay]

GPT-3.5 is pretty bad nowadays. To be honest Llama 3 70B should probably be the default.

[warm cave]

Yeah, way too outdated

[agile jay]

So many models perform better than it and at quicker speeds and lower prices.

[devout cargo]

Claude 3 Haiku, Llama 3 8B

Why doesn't Perplexity use them?

[agile jay]

They have them in labs, but not in the normal client.

https://labs.pplx.ai

[lone citrus]

I guess for many people, chatgpt sounds like better than .... hummm what is llama or haiku? Never heard, not a real ai thing

[devout cargo]

Morphic is open source, I understand, not at the same pace as Perplexity. They already have llama and Grok implemented

[serene sand]

https://x.com/AravSrinivas/status/1781721468180767002

chefs already cooking

Aravind Srinivas (@AravSrinivas) on X

8b is so good. Can create a lot more experiences with it. We have some ideas. Stay tuned!

Twitter · Apr 20, 2024, 4:27 PM

[agile jay]

Yep, the llama 3 models are my main models currently. Through the groq API. Surprised how well they perform.

Only downside is currently the context length.

[devout cargo]

I use the free subscription and Perplexity worked very well now with the default mode, it is a disaster in Multilanguage

[stoic wave]

@agile jay in your experience so far, which one do you think is the best Model for programming/coding cases?

[torn sierra]

Llama 3 is a game changer. Btw. Just played again with you.com and its context window in Claude 3 Opus. Performs pretty well. I gave it around 100 pages pdf and it wrote me a very long article based on it. Something what Perplexity was not able to

[devout cargo]

Copilot Programming

[stoic wave]

GitHub Copilot?

[devout cargo]

Depends. copilot in its free version has also improved a lot

@agile jay I proposed it to the team Being able to upload images or PDFs in the same way that you can use ProSearch to a limited extent in the free version Copilot in Spain now allows you to upload PDFs and images as many as you want. Gemini also allows you to upload all the images you want. I think Perplexity can do that, limiting file and image uploads to a memory limit as well.

[agile jay]

In theory, there's nothing stopping you from handling RAG all locally and not needing to upload the files to the service.

[agile jay]

Depends on the language. On average probably Opus because it's good at following instructions.

[cinder comet]

do you have any other websitess like this? this is similar to perplexity tbh

quite good

even the UI is similar to perplexity

[devout cargo]

I guess at Google IO in May they will release Gemini 1.5 Pro intentional multimodal If not, they surprise like they did with PaLM2 and release Gemini 2.0

[devout cargo]

I saw that by accident in the #Perplexity hashtag a user posted it

The most curious thing is that in any type of search they have Pro search 🤔

The good thing is that Perplexity and You.com are no longer alone, now there is also Morphic AI that cannot be neglected

[stoic wave]

Damn, MorphicAI is insanely fast!

And results not that bad either

[torn sierra]

Llama 3 in version 8 and 70B has not yet fully shown its potential. And yet, version 400B is still waiting. It's going to be interesting 🙂

[devout cargo]

Llama 3 400B if it is a (MoE)

Multimodal Modelo Meta AI 🙂

[devout cargo]

In a very short time the open source community has achieved this in such a short time and it is a beta. I can't imagine giving more time to what they can do.

[south kindle]

this has been interesting to just load up and keep running

this is some weird new religion stuff.

"in the beginning, there was @agile jay "

[agile jay]

Lol, older than time itself.

[devout cargo]

@agile jay
https://www.google.com/amp/s/www.xataka.com/aplicaciones/ultimo-openai-llega-a-copilot-asistente-programacion-evoluciona-nuevo-modelo-ia/amp

[slate condor]

It would be great if Perplexity had an iPad app. One that actually works.

[half venture]

Is that GitHub copilot running in copilot ?

Or is that code interpreter

What openai has etc etc

[devout cargo]

Copilot in its free version is improving a lot 🙂

[half venture]

Meh too restrictive

Maybe mustafa can make it less full

Dull

And more like pi

[devout cargo]

But the restrictions, believe me, don't matter if you can't pay for a subscription, people go to the most functional and complete thing, which is why copilot is a success.

[half venture]

Ehhhh

Chatgpt is more of a success

Than copilot

Even if copilot offers more

[devout cargo]

Copilot is a rebranding of chat GPT . but with free internet and more functions. That's the business of Microsoft and Open AI. why GPT3 does not have internet in open ai. for Copilot to be used

[half venture]

Yeah and still chatgpt is more succesfull

Copilot marketing sucks

Period

Chatgpt has more pro users

More users in general

And I am sure sam Altman will soon find a way to get better gpus

[devout cargo]

Paying yes, free no

[half venture]

Hm

Fair

Yeah Microsoft will have a upper hand

As always

[devout cargo]

Look at Google and Deepmind Microsoft has copied Amazon in its businesses such as Audible Bling. Microsoft has left OpenAI as independent but the one who gets the say and has 50% of the shares of Open AI is Microsoft

[tame current]

is the claude 3 opus limit gone now or is it still there?

[agile jay]

It's currently at 50

[devout cargo]

Perplexity does not use Mapbox, by chance it is on AWS like Perplexity's infrastructure, however Anthropic is also on AWS and Perplexity does not use claude, this is the biggest question I have.

[agile jay]

Mapbox?

[devout cargo]

I thought you knew

https://youtu.be/hFUaXEXfNnA?si=KWY0eyvRZNac2Gzt

https://youtu.be/znOlwELyt8g?si=UDq4joNqi1n7z8i3

https://youtu.be/YKMDw7ERxZ4?si=t0ybyzaEgUZNsihl

[devout cargo]

Do you understand why I ask why Perplexity doesn't use Claude as the default model if they both use AWS?

[agile jay]

Oh, I've not used perplexity for navigation

[agile jay]

It doesn't make a difference though, since they would still be consuming anthropics API. Unless you mean somehow by both being invested by Amazon related groups, that they would be able to share resources, or reduce costs.

[tame current]

The only good thing about Copilot is that it allows copyrighten characters in Dall-e 3.

[devout cargo]

https://twitter.com/perplexity_ai/status/1768307093075874241?t=6G56jPHtGSG5oIDrCRIIxQ&s=19

Perplexity (@perplexity_ai) on X

We've rolled out improvements to local searches through our integration with @Yelp and maps to help you quickly find information on local restaurants, and businesses!

Twitter · Mar 14, 2024, 4:03 PM

[devout cargo]

Correct

[tame current]

Chatgpt doesn't allow this.

[agile jay]

Yep, because they have a simple filter for the input

[tame current]

It's possible to trick ChatGPT though because of the way it's trained. So, if you ask for a yellow bear with a red top, it's probably going to create Pooh.

Dall-e 3 is quite interesting. If you ask it in ChatGPT to create a copywritten character, it is to choose 3 key adjectives of the character to create a generic alternative.

[agile jay]

Yep, that's the method I used. Find an image of what you want, pass it into GPT4V and ask it for a detailed description of the character, and then use that description to generate the images.

[devout cargo]

I think Amazon is seeing how Mapbox Perplexity and Anthropic evolve. I think Alexa is going to use Claude sooner or later to answer questions from Perplexity users. Alexa already works in General Motors, using Mapbox is a possibility. Amazon has invested in all these companies, you could end up like Audible and Blink

[agile jay]

There is a difference between investment and majority shareholder.

[tame current]

It didn't go so well!

[agile jay]

Lol

We have disney at home...

[tame current]

My thoughts exactly.

[devout cargo]

Looking at the AI landscape, I imagine that it is tempting for Amazon to get a good piece of cake from these companies to have them under its umbrella.

[agile jay]

Yep, but if they're not a majority shareholder, they can't do Microsoft level of stuff.

[devout cargo]

Maybe it's something simpler, for example. I let you use AWS resources and you let me use your technology

[agile jay]

Very unlikely to hand over their IP for some training resources.

They have other investers too

Such as Google, which are competitors lol

And probably have more AI inference than Amazon.

[devout cargo]

https://www.google.com/amp/s/www.genbeta.com/actualidad/amazon-invierte-4-000-millones-dolares-anthropic-para-hacer-frente-a-chatgpt-lucha-mejor-ia-solo-acaba-comenzar/amp

They also want to make us pay part of the Prime for using this modality 😅

[agile jay]

Yep, but when they say that Amazon will be able to use there models, it will most likely still be through anthropic servers though.

[agile jay]

Lol, the sub will just get higher and higher

[devout cargo]

The Alexa developers are against it because you already pay in Prime for that service

[agile jay]

Well the Alexa devs haven't achieved much...

Especially in comparison to modern LLM's

[devout cargo]

So I see Antrophic developing Alexa 😅

[agile jay]

No, the Alexa team should just make a wrapper that uses Claude 3.

If they are using haiku or sonnet, it will be really cheap.

Even with opus, it would be cheap.

Since alexa doesn't speak that much

[devout cargo]

In the Amazon app I think this is claude

[agile jay]

Most likely. Maybe ask it a task that gives different results for different models.

[warm cave]

amazon has a chatbot?

[agile jay]

No, it's just support in the amazon app

So either GPT3.5 or Claude haiku/sonnet

[warm cave]

oh. lol

I want its system prompt, haha

[livid mantle]

its using llama3

[formal parcel]

Just even quant Sonnet a lot

[tame current]

what..

its a dollar...

so 2k tokens per .1 point

so every ip gets 8k tokens

[haughty lodge]

I think the price is actually Chinese yuan

So it's like 14 bucks

[tame current]

that makes more sense 💀

[haughty lodge]

Yeah at that point openai is just the better option

[tame current]

not if u abuse their site Trollface

[tawdry chasm]

pretty sure they still make a heck lot because they use non legit API accounts

[glad pilot]

actually the site offers opus

And it is way cheaper than using API

[arctic wharf]

What do you mean by non legit? 💀

[tawdry chasm]

as in they were able to get some accounts with credits at no cost or something
there's a pretty big market for trading these accounts

these days u can get accounts at a price of a few percent of official

[arctic wharf]

Thats crazy..

[buoyant kindle]

Normally I'd disregard it as an annoying conspiracy theory. But when a service is randomly wonky, I'm at least slightly open to the idea.

[tawdry chasm]

there are many routing platforms out there that provides these really cheap API

like they just collect(buy?) these accounts at a bulk rate, since they expire pretty quickly they are very cheap, then make a routing platform

[buoyant kindle]

It makes perfect sense, and would explain the wonkiness.

I'm guessing Poe, Phind, Julius and the others, which are more expensive than ChatGPT, Claude.ai, Gemini Ultra, are legitimate?

[tawdry chasm]

yea these big platforms are using legit sources

they probably have an agreement with openai/anthropic/whatever

[buoyant kindle]

@tawdry chasm
Got it. Thanks.
I'm leaning towards using ChatGPT Plus only, but find their maximum context length of 32k tokens lacking.
Got any recommendations?

[tawdry chasm]

what use cases do u need more than 32k context?

[buoyant kindle]

Analyzing large text reports. But also for summarizing long podcast episodes (2-4 hours). Multimodality isn't needed - just text.

[civic flax]

get some picture

[tawdry chasm]

chatgpt's rag implementation isn't the best but rag is definitely the only way to go

[buoyant kindle]

RAG? Really? I need it to "understand" the entire text and respond in context. I don't need simple look-up answers.

Gemini 1.5 Pro and Claude 3 (even Haiku) works very well.

[tawdry chasm]

u might want to try feeding it to gemini pro 1.5

[buoyant kindle]

Oh, yes, I've done that. It's amazing.

[left gazelle]

ey who down to share pro

i have 30k vbucks

[tame current]

I think you'll last about 10 more minutes if you say that.

[halcyon coral]

Still not allowed

[tame current]

If you're going to do it, maybe don't do it in the official Discord.

I don't think they'll like that.

The guy who was begging last night was unbelievable.

[half venture]

Ok what is aravind going to release tommorow

[tame current]

He basically said. Let me share your account, and I'm not telling you why.

That sounds very legal.

[primal sigil]

Hola ! Hay hilos específicos para novatos en castellano? Gracias

[fluid mauve]

No me parece. Creo que la mayor parte del sitio está en inglés. Podrías intentar hacer una búsqueda en el cuadro de búsqueda usando vocabulario en español.<- Google Translate

[primal sigil]

Gracias

[fluid mauve]

De nada

Not Google Translate

[serene sand]

#💗│sharing message

[vapid onyx]

Good morning!

[tame current]

Peepo_Wave

[vapid onyx]

How are you?

[tame current]

doing alright, curious though....any idea why Aravind tweeted 04/23

👀

[vapid onyx]

No idea tbh. Hoping to find out just like you!

Curious what it entails

[tame current]

hehe

link ?

[tame current]

https://twitter.com/AravSrinivas/status/1781902284844421624

Aravind Srinivas (@AravSrinivas) on X

4/23

Twitter · Apr 21, 2024, 4:26 AM

[nimble field]

perplexity in house model surpassing gpt-6 and claude gamma 3

100$ per million token output and input

[tame current]

still making a bot and prob api to abuse their authcodes

already have a working one

though i can only get ab 8k tokens per ip

[fluid mauve]

Your making a bot to ABUSE the system and posting it in the main discord or am I misunderstanding something

[tame current]

uh

[tame current]

no :3

[fluid mauve]

Hey, like i said i may have mis understood. you said "still making a bot and prob api to abuse their authcodes" and i womndered if that what you say you are doing?

[tame current]

what u even mean main discord

[fluid mauve]

as in the General section is the main comms part of the Perplexity discord channel

[tame current]

if I understood correctly, he is not talking about perplexity

[tame current]

but this site

there are many chinese sites like that because chatgpt is blocked in china iirc

[tame current]

official discord

[fluid mauve]

Ahh, that makes more sense. Thanks fort clarifying.

[tame current]

yerp

[tame current]

i ripped this site from gpt4free

their stuff was ass

[tame current]

makes sense, I like this github https://github.com/xx025/carrot but you need to sort through a lot of sites to find a working one lol

[serene sand]

https://tenor.com/view/robot-depressed-marvin-hitch-hikers-guide-to-the-galaxy-gif-4931652

[tame current]

💀 thanks tho

the first site i visited straight up uses actual openai keyss

💀

[tame current]

not all are free, the ones with "😄" emoji are free

[eternal robin]

i wasn't expecting this from perplexity 🥲

[tame current]

💀

[tame current]

prob gonna go with this one

https://share.wendabao.net

or netfy

idk, maybe even multiple to stop outages

i need to be on telegram more 💀 whole ass underworld

multiple sounds like a good idea

[tame current]

yup

sucks with these is that its really slow

[stable fossil]

sorry. what is this about?

[tame current]

hes edging us with some feature drop or something ig

[stable fossil]

ohho. i hope he is talking about something related perplexity. not another partnership

[harsh verge]

https://youtu.be/LGuA5JOyUhE?si=AzhxwS7mCeYXwTGA

[cinder comet]

did they increase the limit?

for opus

[scarlet pagoda]

Yes but not by a lot

[slender nexus]

I cant seem to use a coupon code for pro

[agile jay]

I think you can only do so in web

[slender nexus]

Ah

[tame current]

1. the pfp 💀
2. at least dm him...

[scarlet pagoda]

WHAT ARE YOU SAYING BRO

[tame current]

bro is not professional...

[scarlet pagoda]

CAN YOU HELP ME BRO

[stoic wave]

What sort of challenges? can you explain it so people can help.

[scarlet pagoda]

I'm currently engaged in AI development but recognize the need for a mentor to navigate the intricacies of expanding the business successfully

[carmine cedar]

bro is in too deep even his messages sound ai generated

[vapid onyx]

If you require support from the team regarding your Perplexity subscription, you can reach out to the team via email: [e-mail]

Otherwise, mentoring in AI is more or less community driven. People help people all the time here.

[agile jay]

Or... Ask perplexity

That's what it's designed for after all.

[vapid onyx]

https://tenor.com/view/think-about-it-use-your-brain-use-the-brain-think-brain-gif-7914082

[scarlet pagoda]

but i need a large funding for the ai development

[agile jay]

Yep, you're not gonna get it here...

Unless you have a PhD in AI research, worked for a large AI company, or have a really good idea.

[serene sand]

https://www.ycombinator.com/apply

[agile jay]

But since you are looking for a mentor, I doubt you have any of them.

[scarlet pagoda]

so how can i get mentor

[agile jay]

Mentors don't come from nowhere. The easiest way is to join an AI research company.

If you have the qualifications.

[serene sand]

https://www.youtube.com/@AndrejKarpathy/videos

[scarlet pagoda]

but i am 19 year old boy and i don't have qualifications

[agile jay]

Nowadays your best mentor is gonna be the internet.

[agile jay]

Go to uni then...

[serene sand]

https://tenor.com/view/yt-youtube-logo-gif-27453294

[agile jay]

Yep, and huggingface

[arctic wharf]

Hello 😄

[agile jay]

Hi

[scarlet pagoda]

by the way who are you

[arctic wharf]

Btw does perplexity use huggingface models?

[agile jay]

In labs

[arctic wharf]

And do perplexity self host models or just buying them from somewhere?

[agile jay]

https://labs.pplx.ai

I assume they are hosting them.

If they wanna maximize their API revenue.

[arctic wharf]

Aren’t you developer for them?

[agile jay]

Nope, I don't work for Perplexity.

[arctic wharf]

I thought you were

Because you are always helping people here

[agile jay]

Nope, I just know a lot of stuff

Employees are the pink names

[arctic wharf]

So you know how to develop?

[agile jay]

Yellow are the mods

[agile jay]

Yep

[arctic wharf]

What’s your favourite language?

[agile jay]

Probably Go

Depends on the task

[arctic wharf]

Wow that’s nice

Is go similar to js`

[agile jay]

Yes and No

They have similarities and differences

[scarlet pagoda]

what are you saying about hugging face

[agile jay]

Since they are both C based langauges

[arctic wharf]

Fair enough

[agile jay]

Hugging face is where the biggest AI open-source community converges.

[arctic wharf]

Yes they are

[scarlet pagoda]

who are you

[arctic wharf]

Btw if you want to selfhost models, you need really fast servers right?

[agile jay]

Go has a very small syntax, great compiling and other stuff, which makes it my preferred language.

[vapid onyx]

@scarlet pagoda https://www.perplexity.ai/search/How-can-I-IQl7x61RSImPFp0c7MyJUA

[agile jay]

Depends on the model

[arctic wharf]

Like 48 cores and 144 gb ram

Or something like this

[agile jay]

<Code/>

[arctic wharf]

Or am i wrong?

[scarlet pagoda]

thanks

[agile jay]

Doesn't need to be that powerful for 8B models

[arctic wharf]

For example llama 3

[scarlet pagoda]

i mean you are boy or girl

[arctic wharf]

80b is their highest right?

[agile jay]

For the 8B models, a lot of hardware should be able to run it.

70B model needs around 2-3 4090's, and 8+ for fine-tuning.

[arctic wharf]

Really, i mean i have a raspberry pi 4,. 8gb at home, do you thing it can handle it, lol

[half venture]

70b model is a pain in the ass to run

[agile jay]

Still waiting for the 5090's to update my server.

[agile jay]

Use groq for the llama models.

[half venture]

A bunch of my friends shared there gpu and almost toasted them all

[arctic wharf]

You have a home server?

[agile jay]

Yep

[arctic wharf]

Ouch

[agile jay]

It's normally not used for AI, but I've been testing out llama 3 on it.

[arctic wharf]

And if we compare cpu/gpu whats better to run a model on?

[agile jay]

GPU in most cases

[scarlet pagoda]

how much cost of making ai

[agile jay]

Yep, it's just cheaper overall to use groq and it also has better performance.

[arctic wharf]

Groq is really fast

[scarlet pagoda]

and also running an ai

[agile jay]

Llama 3 70B Cost around 20M

[arctic wharf]

I wonder how the can afford that powerful servers

[scarlet pagoda]

not bigger cost

[agile jay]

Well their 405B model costs a lot more.

[buoyant kindle]

When's 405B expected? What are the reasonable rumours saying?

[agile jay]

It's supposed to be coming out around a month after the other releases. But since they haven't actually finished training yet, might get delayed.

[half venture]

Quick question

Nvidia will soon start shipping h200s

From april to June

Will be there shipping dates

Do you think Microsoft was waiting for h200

And openai too

[agile jay]

I imagine they are arranged but the buyers and their data centers.

Installing such a large amount of hardware is no easy task.

[half venture]

All the releases are now set for between april and June

Like Google

Etc etc

[agile jay]

Nope, they weren't.

It took only 25K GPU'S for llama 3

And mark said they will have 350K soon...

[vapid onyx]

https://www.perplexity.ai/search/How-can-I-IQl7x61RSImPFp0c7MyJUA

[agile jay]

You can either double the compute, or double the training time.

[vapid onyx]

I added a follow up to your question

[half venture]

You didn't understand me

[agile jay]

You are asking if they were waiting before training their next models?

[half venture]

I am saying I don't think it's a coincidence all the tech companies major events are between april and June when nvidia starts shilling

Shipping

[agile jay]

For the h200 shipments?

[half venture]

No my point was

H200

They are starting to ship h200 now

[agile jay]

That's what I meant, lol

[half venture]

Yeah

[agile jay]

Yes and no, because of how long training takes and then safety tests afterwards.

[half venture]

Openai and Microsoft definitely were waiting tho

[agile jay]

Google, Openai and Anthropic do a lot more safety tests from what I've seen.

[buoyant kindle]

It doesn't make sense to wait with anything until H200...

[half venture]

It does

If you want to run I dunno

A huge expensive model

Like sota

Sora

[agile jay]

Maybe before releasing them. But not before training.

I guess that's where open-source shines.

Since they don't have to worry about hosting

They just release it for others to host.

[half venture]

Yeah before releasing them so they don't end up in the current situation openai is in

Completely bottlenecked

At 40 messages 3 hours

And anthropic is even worse

The only exception is Microsoft copilot

And Google

H200 will solve a lot of capacity issues I think

[torn sierra]

Therefore, we shouldn't expect much from GPT-5. In fact, we shouldn’t expect GPT-5 anytime soon, at least not in the form we would want, and certainly not available on a large scale. After all, OpenAI/Microsoft does not have the necessary technical infrastructure to support it 😉 Instead, we should anticipate opportunities in solutions from Meta and expect the Llama 3 400B model. I use the Team plan in ChatGPT - there are a lot more queries 😉

[half venture]

Umm they do have the infrastructure to support it

Microsoft definitely does

They are just waiting for h200 for that EXTRA EXTRA use case

[torn sierra]

Dunno. But certainly not on such a scale to be widely available yet (at least I believe so), especially since there are still frequent issues with GPT-4, although it has improved significantly. Unless the GPT-5 model is not as large as we expect 😄 As Meta has demonstrated, smaller models can also perform exceptionally well. In any case, there will be an opportunity with the availability of the B200, but for now, only orders for these chips have been made as I guess 😉

[warm cave]

So there are a few possibilities I can think of, 1. This is the day that the Rabbit R1 gets dropped. 2. Grok 1.0 is done being fine-tuned and will be added to perplexity 3. Llama 70b is added to perplexity. 4. Big update to the platform. 5. Opus limits are removed. Hopefully that if it's a big improve to the platform than they will remove the opus limits beforehand, otherwise there is a good chance no-one cares and just fills the comment section with questions about the Opus limits. https://x.com/aravsrinivas/status/1781902284844421624?s=46

Aravind Srinivas (@AravSrinivas) on X

4/23

Twitter · Apr 21, 2024, 4:26 AM

[half venture]

Sonar will be upgraded to llama 70 b

That's it

[warm cave]

Hopefully if that's the case sonar is actually good this time

[wraith crow]

I vote option 4

[halcyon coral]

Same

😏

[warm cave]

But it is the same day people get there hands on the Rabbit R1, and perplexity is partnered with them, so if I had to guess what was most likely, I might say option 1. But I guess we will see

[agile jay]

They are mods... Better not be trolling us...

[serene sand]

https://tenor.com/view/heidi-klum-number-two-2fingers-deuces-your-second-yes-gif-25857953

[agile jay]

Wow 3 mods now

They are up to something...

[cinder comet]

how is dawn a mod

he

he's everywhere

[agile jay]

No idea who dawn is

Some random mod

[cinder comet]

yea

agree

[agile jay]

@golden anvil who are you?

Don't think mods would randomly be trolling, so it's probably #4

[halcyon coral]

Who knows...

[wraith crow]

https://tenor.com/view/superstore-amy-sosa-im-just-guessing-just-guessing-wild-guess-gif-24963833

[agile jay]

Sure...

[tame current]

Is llama coming to perplexity?

[serene sand]

#💬│general message

[golden anvil]

Apple buys perplexity on 4/23

[agile jay]

That's the day perplexity is finally solved, and becomes simplicity

[wicked zodiac]

I wouldn't mind. either Perplexity or Pi (but I prefer Pi, its more "Apple like")

[minor shard]

That would be wild.

Also very disappointing because it’s just another company Apple is eating up

[torn sierra]

Seems like a solid choice, assuming it’s legit

[tame current]

Or maybe it's just ceo's birthday date

Maybe it's when the cap goes up to 300!

[tame current]

this is gonna be a pain to code 😭

gpt 4 turbo tho

not even

gpt 4

guh

[minor shard]

Watch OpenAI announce GPT 5 tomorrow for like November 5th lol

Just to keep people looking at them

While Llama 3 405b model comes out in a month and possibly beats it

[tame current]

its not unlimited it seems

womp

[orchid magnet]

Apple bought perplexity?

[cinder comet]

oai tryna stay relevant

[cinder comet]

fake news

[agile jay]

What is it?

You are just looking at http requests in postman

[tame current]

event stream

legit the site

makes a request for every character

[arctic spindle]

Do someone know when GPT-4-Turbo-2024-04-09 is going to get added to perplexity?

[idle hull]

Hello, good evening, I wanted to ask a question, why did you remove the 7-day free trial? I'm just asking because a colleague of mine wanted to try the service and get his first impressions.

[agile jay]

People were abusing it for making bots etc

[brisk patio]

hi there, yesterday my opus seemed to be unlimited again, but this morning i got 0 left... Any official announcement????????💢

[agile jay]

It was just raised from 30 to 50

And it only shows the number at 10 left...

[idle hull]

Well, that makes sense, but maybe it also has a solution to bring that possibility back.

[agile jay]

Very unlikely in the near future

[brisk patio]

thanks, but omg, then 50/24hrs??? just thinking about making my 2nd pro account in PPLX

[agile jay]

Lol, just wait a bit, or change it to GPT4Turbo

[idle hull]

Well, at least I have been able to show the service to a couple of my acquaintances, I will tell my partner to try it and if he likes it he can try to pay for the next month.

[agile jay]

How many requests do you make a day?

[brisk patio]

around 50-100 i guess, TBH I'm from JP and I work for article production and making PPLX for assistance

[agile jay]

How many prompts do you normally use for each article?

[brisk patio]

not so many, some prompts and some sample such as style, rules,,,, it takes larger tokens than English one, though

[feral marlin]

if you're just using it for claude opus then maybe just use claude.ai until they raise the limit

[brisk patio]

ty lunar,yeah, I have claude API tiar4 too, but PPLX can do web research you know,that why I love PPLX from very first

[tame current]

Is that perplexity copy still up?

[warm cave]

Devv.ai, morphia.sh, Komo.ai, you.com, or what one?

[young dew]

drink water

[tame current]

Github one

[warm cave]

Yep, https://www.morphic.sh

You can try it here, or run it locally by downloading it from GitHub

[tame current]

I was talkin ab clarity

[warm cave]

Oh, have not heard about that one

[tame current]

They've since changed the looks ig

[warm cave]

https://github.com/mckaywrigley/clarity-ai

This it?

[tame current]

Yeah

[warm cave]

Wonder if I can get groq working in it

[stable fossil]

higuys. i have requested refund for my yearly sub which i bought on april 2. they have sent an refund invoice to me via email. but i still don't see my refund amount in my card. (they have sent the refund invoice exactly 11hr before from now. ) how long it will take to reflect in my card?

[tame current]

pplx slow for anyone?

[minor shard]

Who is Lequn? https://x.com/aravsrinivas/status/1782607360323924446?s=46

Aravind Srinivas (@AravSrinivas) on X

Lequn has joined Perplexity full-time

Twitter · Apr 23, 2024, 3:08 AM

[haughty lodge]

they said 70% of the 4t request already go to it in #👑│pro-lounge

[tame current]

Finnally found a way to keep my gpt bot completely free, supports claud opus, gpt 4 turbo, and gemini pro

[arctic spindle]

Well, I did some tests with perplexity's gpt 4 turbo, and it said his knowledge cutoff is December 2023. I tried asking it to write a user review about November 2023 movies and December 2023 movies, and he didn't know about any of those movies, saying like "as of my knowledge cutoff of December 2023" bla bla bla. So in conclusion perplexity's gpt 4 turbo says its knowledge cutoff is December 2023, but still has the April 2023 knowledge cutoff.

[gritty pasture]

Hi, how do I enable Pro search by default in the search url for default Chrome?

currently I have https://www.perplexity.ai/search?focus=internet&q=%s

[warm cave]

How?

[tame current]

Code is public, it's on anyvm bot repo (if u go through the link in description)

It's not fully finished tho

It works but no memory

[warm cave]

Cool, am taking a look

[austere kestrel]

https://www.perplexity.ai/search?focus=internet&copilot=true&q=%s

[austere kestrel]

I dunno but the model cards say the knowledge cut off is up to December 2023.. This test suggests the model used has training data up to at least Oct 23

[mental widget]

Good Morning. Does anyone know how fast the perplexity.pro support is. I clicked on pro support and wrote a question but didn't receive any feedback since one week. It was not even read from the support (visible in the status info)

[halcyon coral]

Feel free to ask your question here 😉

[tribal gust]

https://en.m.wikipedia.org/wiki/Yann_LeCun ?

[gritty pasture]

Thank you so much!

Can i control the model used from the url parameters? If there are docs you can point me there

[halcyon coral]

No that's not possible

[mental widget]

It is a account related question. I would like to change my auth method from email verification to a more secure method. I would like to use my email address. Therefore I asked the support team which methods are possible.

[halcyon coral]

You can't change the auth method, login with email is pretty safe unless you share your password/lose the account.

[tame current]

It's weird how a lot of AI try to correct quotes in paragraphs. It just seems like an easy thing to fix.

[mental widget]

Usually (if possible) I use TOTP which would be a more secure way and sometimes I don't have access to my mail (when I am in the office)

[cinder comet]

How can i access playground on mobile?

[rustic meadow]

https://labs.perplexity.ai

[fluid mauve]

I am such a cynic. this could be a legitimate offer.....

I know the nothing 2a comes with a perplexity pro subscription. I know a perplexity Pro subscription costs $200 a year. This just seems to good to be true. As I said i am a cynic. years in cyber Security makes me a very paranoid person. Please accept my appologies if this is a true offer, but I would need more proof. The old Cyber adage of "Trust, but verify!" springs to mind

UPDATE: looks like that post has gone. Good work Mods!

[halcyon coral]

🔨

[fluid mauve]

@halcyon coral is there a way to highlight a message to Mods?

[halcyon coral]

Long click/right click -> Apps -> report message or just @mention

[fluid mauve]

as i don't have anyting under Apps i will try and remember the @mention

[austere kestrel]

wonder whether was it just an embarrassing typo.. or deleted for some other (more interesting) reason ha

[willow flare]

is Perplexity down?

[mental widget]

No, I am able to get a response from the website when I ask perplexity

[tame current]

which one would you say is the best (and free) ?

I think I'll switch over to them till perplexity improves

[fast meteor]

do you have a screenshot of the tweet?

[austere kestrel]

it was something like "Excited for Lequn to be joining full-time"

something like that anyway

[fast meteor]

probably Lequn Chen?

[austere kestrel]

@fast meteor if it's a typo, this guy would seem a likely candidate I'd have thought for the actual intended 'announcement'

but i'm not really sure why everything is cryptic

or just not officially announced...

unless it's some partnership with the latest new gimmick

[fast meteor]

he might be the talent pplx needs. he is specialized in optimization of distributed systems..

[austere kestrel]

if he's a comms guru they should get him on board for sure

[livid aurora]

Guys why I get sometimes unrelated information to my questions

Sometimes I get only about 5 sources

and then the info is messed up

[fast meteor]

pretty weird that the ceo announced it on twitter ...

[dire gate]

The pro membership renewal failed bc the bank account didn#t have the funds. I received an email to register another payment method but I dont have another credit card, I use this one only for perplexity bc it doesnt offer other payment methods. The acc has money now and I woulod like perplexity to charge it again, what to do???

[tame current]

Copilot is kinda stupid sometimes.

Why would ChatGPT use Google's Lamda?

[fluid mauve]

It is hallucinationg. I get similar results if you ask your model its cuttof date or what model it is. I am using Mistral-Large

[tame current]

Yeah. It seems like it's not a standard hallucination. E. G. training data. It nicked the 'fact,' from a website.

[safe jolt]

not able to subscribe for pro
i click get started
and it just reloads the same page

[safe jolt]

no replies, going to purchase claude itself then...

my subscription in claude ended today, so thought id get this, but then ig ill stick with claude now...

[austere kestrel]

what source [4] is it citing there?
just out of curiosity

[tame current]

https://www.chooseoxygen.com/en/blog/chatgpt-vs-notion-ai-comprehensive-comparison-for-ai-writing

[tame current]

I mean. This isnt the support room. This is general.

[wise edge]

https://www.bloomberg.com/news/articles/2024-04-23/ai-search-startup-perplexity-valued-at-1-billion-in-funding-round?cmpid=socialflow-twitter-business
The startup is announcing a new enterprise version of its chatbot, priced at $40 a month, with added features including stronger security and data protection measures. Bridgewater Associates,Zoom Video Communications Inc. and the Cleveland Cavaliers are among the early group of businesses using the product.

AI Search Startup Perplexity Valued at $1 Billion in Funding Round

[tame current]

1 billion. Yet, they reduce our Opus by 90%. The nerve.

[warm saddle]

please help me to learn c shrap langaue with unity

[wise edge]

ask llama 3 using vs code with extension ~

[austere kestrel]

thanks. that's interesting. baffles me what an "enterprise" customer would get for $40 a month for this chatbot that is not currently offered to general subscribers paying $20 - but will be interesting to see what it involves nonetheless

we also learn that it is a team of 55

[tame current]

Yeah. No logging. Probably.

[austere kestrel]

but you can already opt out of that, no?

[tame current]

For ai training.

I meant for privacy.

[austere kestrel]

Or ig there is perhaps a distinction b/w training vs gen logging

right right gotcha

[tame current]

Me doubts Opus will return.

But, they'll probably add it back for more money.

*More Opus.

[austere kestrel]

but just to be clear, you're aware of this in Settings yeah

[tame current]

Yup.

Turned off.

[austere kestrel]

http://perplexity.ai/
https://chat.openai.com/
https://claude.ai/

[fast meteor]

The 55-person startup

w8 40 usd? thought it was 15 ...

[austere kestrel]

for "enterprise"

[meager parcel]

hola

[fast meteor]

10 days ago, it was 15 in the code👀 unless i hallucinated

[tame current]

It's going to turn into one of these 5 tier things.

[meager parcel]

o hy

[austere kestrel]

there was a code for this new / unreleased enterprise version of the product?

classy

[fast meteor]

maybe just a placeholder

[austere kestrel]

do you know if they take steam vouchers?

[fast meteor]

dk, maybe vbucks are accepted

[halcyon coral]

https://fxtwitter.com/AravSrinivas/status/1782775219733844256?t=Oo_2sf1Yj-XImPRrzO19nA&s=19

[half venture]

Where's the llama 70b

Sonar model

[winter owl]

hi, is there a way to set cap on how much is spent on pplx-api? on OpenAI we can set soft and hard caps to avoid incurring in unexpected costs

[tame current]

Hello everyone: The rabbit r1 comes with Perplexity, you should join their Discord at the bottom of the page -> You can also watch the livestream of the pickup party there later https://www.rabbit.tech/. Perplexity rules

[placid coyote]

I thought Opus was back – there were no daily usage limits shown in the settings, so I switched back. But now I've been kicked back to Sonnet? That's disappointing. I wouldn't have wasted my daily Opus quota if I had known it was still so limited... 😞

[sweet jasper]

Its le enshittification

[tame current]

hey ther ai friends

[hallow linden]

Comically dramatic video with very stocky stock footage

[livid mantle]

LOL

[hushed drum]

Enterprise Pro!?!?!?!?!?!?!?

What's that brother

[fast meteor]

enterprise ultra safe*

[sleek tapir]

it's normal pro but they promise not to sell your data

[livid mantle]

enterprrise 35 pro

[feral jay]

So is our data being used for training now??

[livid mantle]

always has been

🔫

[feral jay]

"Increased Data Privacy
Your data stays yours, period. We never train our LLMs on our enterprise customers’ data."

makes me think I shouldn't have been using perplexity for work stuff at all

[stoic wave]

fake promise, they sell it anways.

[halcyon coral]

No, you can always opt out

[livid mantle]

know your rights

you have the right to toggle that off as icelavaman is showing

[feral jay]

Right yes, I did that. Ok so I can't understand what's worth the extra $20 here.. single sign on??

[fast meteor]

as individual user, i also want more control over data retention time :/

[livid mantle]

consult with your local general attorney

[bold gull]

If I go with the new plan, will updates to model limits and throttling be clearly communicated so I don’t need to go to Reddit or Discord to find out?

[lone citrus]

I think the enterprise had infinite messages

[bold gull]

Oh that would be interesting

[lone citrus]

Also unlimited profile upload, pro user only have max tp 5 files in one thread chat

[livid mantle]

unlimoted profile picture upload

[bold gull]

Thank you

[lone citrus]

I wonder is there any new features for pro user

[lone citrus]

U welcome

[vagrant spear]

Will there be a promo code for normies?

[livid mantle]

embrace your wanderings

[lone citrus]

Could we buy the enterprise? Or it's only for companies, organizations

[livid mantle]

get LLC, you are considered a enterprise alone

[lone citrus]

Never think about open a LLC just because buy a AI subscription

[icy cloak]

For enterprise, can we get mroe details what this means?

Unimited Pro Searches per day
Pro Search includes three times as many sources and provides longer, more comprehensive answers.

[feral jay]

Oh my goodness is this why they keep shrinking the number of returned sources (it's down to 20 now right?) ...so they can make it a feature of paying for more?

[halcyon coral]

No...it's just a short explanation of Pro Search

[feral jay]

Oh, so it's still gonna be limited to 20 😭

[halcyon coral]

More sources ≠ better answers

[feral jay]

It's like they hate academics, between the 32k context and these shrinking limits

[fast meteor]

top P instead of top K pls

[feral jay]

I have old pplx searches that gave me better answers because they included more sources, remember that in academic focus 50-80% of the articles it finds are basically unrelated

[lone citrus]

I agree, but I do think it would help to find more perspectives on the topic.

Maybe it's should be something optional

[feral jay]

If that were true, then why would pplx advertise more sources as a benefit?

[sleek tapir]

https://tenor.com/view/money-mr-krabs-gif-18326632

[warm cave]

Be honest, you are probably a little disappointed about the news. Privacy is important but not very exciting, haha 😆

[serene sand]

#💭│ideas message

[halcyon coral]

Well 20 sources are better than 4 but you don't need 50. The more context there is, the more often your AI profile may be disregarded.

[tame current]

We who paid for the Pro version could well have the same privacy benefits as the Enterprise version....

[warm cave]

32k context dose not go very far with web search sources, more can be better, but higher quality searches is the most important, after that is optimized you could bring in more high quality sources by utilizing a 128k or 200k model

[minor shard]

phi 3 mini is out now

[hollow flame]

grats on the raise announcement!

[warm cave]

Very excited!

[minor shard]

i know! Thats where i saw it xD

Really would love these fast models to be hooked up to the internet stares at pplx

👀

[feral jay]

I guess. From my very old threads, 30 or 40 is also better than 20. Or maybe their re-ranker got worse for papers, the obvious thing was that sometimes the 30th paper was the winner and now it's gone

[cinder comet]

Stap it

How bad is it?

[meager parcel]

men

[minor shard]

Trying to understand why Microsoft is releasing Mini models. Unless they plan on putting them on their new surface devices

[warm cave]

Keep in mind this is a 4b model, but it is very impressive from my testing so far

[minor shard]

Supposed to be equal to Llama 3 8b

[warm cave]

But have only had it for like 5min. Haha

[warm cave]

That would be insane

[minor shard]

Yeah 100%. I am waiting for the bigger model that = the llama 3 80b so I can hopefully run it on my local server

[livid mantle]

400b

[austere kestrel]

honestly right

the big cryptically communicated bombshell: enterprises get single sign-on and a bunch acronoyms that apparently = better privacy (but still isn't on-prem... and makes one wonder what the non-enterprise user have in place for protecting their data..) whoop

[warm cave]

phi3 essay

[warm cave]

Our data is just stored in a bunch of .txt files spread across a bunch of free google drive accounts 😆

[solemn fox]

Introducing Perplexity Enterprise Pro! [insert a bunch of buzzwords without explaining what it actually is]. By the way it's $40 / month!

so what is this thing anyway

[warm cave]

my takeaway was $20 a month for normal, and $40 for better encryptions and such

[austere kestrel]

ha that's very thorough (and correct) ✅

[sinful smelt]

Does the context window still remain at 32k for Perplexity Pro Enterprise?

[warm cave]

I think mistral is the best big company for if you want 100% privacy, since they work with company’s do do on prem hosting, but at this point use llama3 its soo good

[austere kestrel]

lol yeah I'll happily give em 200k tokens of my data if they'll take it as an eterpriser user lol

[upbeat tiger]

Again another trash thing that no one asked, meanwhile we're still waiting for our opus restriction to be removed.

[warm cave]

I’m sure that enterprise was using perplexity and they wanted to cash in on those who would pay extra for security, but it would be nice if they focused on there current supporters by being transparent and bringing back normal usage

[serene sand]

#💗│sharing message

[slate condor]

Wouldnt it be cool if they used some of this money to make an iPad app that worked? Now that would be something.

[agile jay]

I find it strange that high security is not the default...

[signal hamlet]

Hey, @slate condor! Which version of the app are you using? What kind of issues are you facing? If you are using a VPN something is not loading, please try swithching the hotspot/provider. Please let us know more details in the #1140622008086970420

[warm cave]

Because if it’s not default, then you can charge for 😉

[agile jay]

Lol, security is something you shouldn't need to pay for.

[slate condor]

Hi Alex, I'm not alone in this and do not use a VPN. It is the latest version of the app and it does not function at all. I have sent several requests, Sorry just quite dissapointed in the product.

[agile jay]

Do you get a similar problem on the website?

Such as cloudflare verification?

[upbeat tiger]

As pro users we're already paying, and we don't even have security, that's how bad it is

[slate condor]

Nope. IOS is fine, Web is fine, iPad has been broken for weeks. They sent an update but still useless on iPad. They are aware but are not fixing it

[agile jay]

Rip, and apple doesn't support webapps

[halcyon coral]

But what exactly is not working?

[slate condor]

Cannot search, spins on sign in, etc. Broken.

[agile jay]

Yep, IMO any apps that are handling user data should have max security even for free users.

[upbeat tiger]

I agree with you, but this is even more unforgivable when you are actually paying.

[agile jay]

It's like saying, we don't value your privacy to your users.

[heady quiver]

What discount codes / coupons available for this month?

[warm cave]

There are refiérala codes

[heady quiver]

No; real discount codes or 1-2 free months, like they’ve done in the past.

[warm cave]

In that case, no. They stopped doing free trials due to abuse of the platform

[livid mantle]

Thank you for your hard work Mr. @signal hamlet.

[signal hamlet]

Please check your DM, @slate condor.

[tame current]

congrats on the investment round team!

is ewnterprise pro release already?

[serene sand]

https://www.perplexity.ai/enterprise

[tame current]

So it's just pplx pro with the security compliance for mid to large companies?

[serene sand]

you have FAQ on bottom of the page

[feral marlin]

Does this mean normal paid users get their data trained on

Also does enterprise still have the context limits and opus limits

[serene sand]

#💬│general message

[serene sand]

#💬│general message

[feral marlin]

This doesn’t answer either of those questions I don’t think, just says infinite pro searches and doesn’t mention context or model usage anywhere

[serene sand]

exactly, [e-mail]

[tame current]

💀

[fast meteor]

the expand and collapse sign should switch places💀

[minor shard]

lol i just died laughing, idk why

[tough wolf]

Been prompting all day for my react three fiber, next.js and shaders project but dude keeps asking me what specific thing I wanna about the common cold every prompt 😂 Symptoms ? Causes ? Treatment ? Prevention ?

[fast meteor]

i bet it is the same person who wrote this: #1210704558800642098 message 🤣

[alpine marsh]

how much opus in one day for the version of Enterprise Pro?

[warm cave]

I am dead 🤣

[warm cave]

I emailed, will update the discord with there response.

[tame current]

Probably chatgpt dan would work, but it would be lost in the 8k context fast. Dolphin llama3 8b happily talks about making meth and writes keyloggers, but it's 8b

[tame current]

So, there we go then. They lowered Opus for pro so they can raise it for enterprise. Mystery solved.

[mossy fox]

Honestly i'd love a cheaper model with a larger context size. They can limit Opus, but why not add for example Haiku with a bigger context window?

[warm cave]

Yeah, same thought, why not have Hakiu or sonnet as a 200k option

I mean it's 200k on labs, so why do we have a worse context then the free labs version

[mossy fox]

Yeap, would love to hear a valid reason why it's not in the app

[half venture]

Simple the claude opus they use is actually 200k

Same with sonnet

But in the app

Your file uploads are limited to 32k

Simply because there lazy

To update the context window

Of the file uploads

Due to security issues

And concerns

Simple as that

[mossy fox]

From what i understood the bigger the context window the more expensive the query? Or did i get that wrong?

[half venture]

Yes but that's not really an issue

[warm cave]

That's true

[half venture]

They have compute

It's security reasons

Same with Microsoft copilot

They have plenty of compute

[warm cave]

But perplexity is paying for the API and more context = more $

[mossy fox]

Now i'm curious of the security implications involving that