💬│general

[south kindle]

https://www.perplexity.ai/page/Mars-Earthlike-Past-74G3FinDRWCukdpifBDFOg > expanded from the discovery entry.

[tame current]

Probably, I can't access any fluffs.

[wide mesa]

Considering I need perplexity for work I really hope it gets resolved soon

[tame current]

Me too, It been like this for very long time now.
And when I copy a old thread and start a new thread when I use for a while the same error also pop up anyway.

[wide mesa]

How long long exactly?

[tame current]

Uh since this morning, I mean morning In my country so It would be night In US.
I can count It as entire day now.

Definitely get passed 10 hours.

[wide mesa]

Wouldn't say that counts as a very long time

[tame current]

Welp I not exactly sure If that the real length of time or not because I just start using them today so It might be longer.

[static lotus]

I've found a rather serious bug that I can't state publicly, so please ask the admins to chat with me privately.

[signal hamlet]

Hey, @static lotus, you can DM me.

[static lotus]

icelavaman，I'm already communicating with him, is that okay please?

[winged moon]

Anyone else having slow searches that disappear when you refresh?

[wide mesa]

Gee, I wish the mods would ask the people having problems with their threads to DM them too.

[winged moon]

Yeah, but it's also kind of nice to see people having the same issue as you. It makes me feel not alone with my own problem

[halcyon coral]

The problem is known and a solution is being worked on. However, potential threats to the service are also taken very seriously ☺️

[wide mesa]

Alright, fair enough. Thanks for the response and assurance!

[tranquil orchid]

Hey @wide mesa!

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.

[chilly nova]

what does it mean trained by? and 32k it means it's lower ?

[tame current]

1. finetuned 2. it can remember 32k tokens

[chilly nova]

llama3 isn't better?

[tame current]

sonar is finetuned for searching specifically but it is still basically llama3

so this is technically better (for perplexity)

[chilly nova]

i mean 32k tokens seems very small

isn't llama3 like 1M?

[tame current]

that's a specific version only

afaik

and it would cost a lot to run it ig so they probably decided to go with 32k

I think official llama3 (not finetuned) is even smaller

like the one on groq is 8k

[warm cave]

They are already running a free 200k model, haiku on labs has full context afaik, I don't get why they do t give us a Llama 3 1m or 100k, or give us Haiku 200k. It should be cheaper than using Opus at 32k

[tame current]

oh really? I didn't know haiku had full context

[torn sierra]

I’m waiting for some changes 😅 I use Perplexity mainly for its good search performance. But to be honest I don’t like much the rest. ChatGPT is much better app / UI - unfortunatelly it lacks the good search performance

[chilly nova]

what's haiku full context?

[tame current]

claude 3 haiku on https://labs.perplexity.ai/

200k context

[chilly nova]

nvidia released new models https://x.com/_weiping/status/1786511543255126396

Wei Ping (@_weiping) on X

Introducing ChatQA-1.5, a family of models that surpasses GPT-4-0613 and Command-R-Plus on RAG and conversational QA.

ChatQA-1.5 has two variants:
Llama3-ChatQA-1.5-8B, https://t.co/H7JvIFCD48
Llama3-ChatQA-1.5-70B, https://t.co/Ao3Yw8ECxA

We also open source our instruction…

Twitter · May 3, 2024, 9:41 PM

[runic linden]

having voice support is nice, but I can’t find much info on the privacy statement related to voice. Does anyone happen to have a link for this? Thanks!

[chilly nova]

i'm not sure if these are available on their RTX Chat

[glacial sedge]

127.0.0.1:7300

[warm cave]

Yeah it does, unless they added a cap recently

Just takes some extra time to respond

[deft nest]

Is Opus now capped? Yesterday i used it without problem.

[warm cave]

50 per 24h 😦

They never had it back at unlimited; they just removed the counter until you got low

[deft nest]

Thanks. I need to rewrite a long text. How many characters can i input and expect the same output without loosing quality with Opus?

[torn sierra]

The still unresolved issue with "abuse" 😂 Considering the search capabilities and wide availability of GPT-4, Perplexity isn't bad after all. However, I'm still waiting for some alternative 😅

[warm cave]

You can use an online token calculator to find out roughly how many tokens your text is: https://platform.openai.com/tokenizer. If it's under 32k, then it should work well on perplexity; if it is higher, I might recommend trying the free version on Claude Pro, which will handle up to 200k

[restive finch]

Hey how do I get out of incognito mode? I am literally stuck in there

[deft nest]

Thank you so much. I am new to this kind of things and i am just exploring!

[warm cave]

Your welcome:)

[deft nest]

Unfortunately i cannot subscribe to claude because i am from europe

[restive finch]

how do I disable it?

[warm cave]

You can try Haiku on perplexity labs, I believe it's still 200k. It's not as smart as Opus or Sonnet. But might still be helpful for working with your text https://labs.perplexity.ai/

Perplexity hosts it for free

[deft nest]

Nice will give a try for sure!

[sterile falcon]

Have you guys heard the rumors of ChatGPT search? Browsing through ChatGPT is no where near as good as Perplexity. While OpenAI is creating a separate product, but still using Bing, it seems like it’s just to compete with Google. Any thoughts?

[torn sierra]

After all, it is also said that Perplexity uses Google's results. It's not a separate engine. If anyone has more info, please correct me.

[sterile falcon]

It does, but it filters out SEO/ads, which is pretty much the entire first page of Google.

[torn sierra]

We'll see. But I guess it's not just Bing that's going to be used, there are quite a few different engines involved. We'll see how it works out. Anyway, I'm looking forward to it eagerly

[livid mantle]

all money

[shut dagger]

about this, to prevent abuse, why not use something like reCAPTCHA v3? It's automatic and done in background. With reCaptcha in place, automated abusing should become harder if not impossible !!
(Just a suggestion)

[stoic wave]

Not impossible, there are captcha solver services, its just expensive but very possible.

[strong vector]

Exactly

[zealous flame]

hi all! is there a potential token launch on this one? 🙂

[shut dagger]

how is reCAPTCHA v3 solvable by captcha solver services?

🤔 As far as I can tell, these services only solve the challenge that is been provided to user

but in case of v3, it doesn't require user to solve any challenge, it just analyzes user's behaviors.

[stoic wave]

These "captcha solver services" are no longer limited to solving visual puzzles. Some are using machine learning and other advanced techniques to mimic human behavior, and that being said, it is solvable.

PS: You could try yourself using CapMonster for example, and see it yourself, if you don't believe me.

[warm cave]

Captchas are getting weird, I was categorizing AI-generated images yesterday to login, haha

[shut dagger]

for real 😂

[tame current]

[tame current]

Somebody managed to get Chatgpt to solve a captcha by pretending it was a number on a lost family gift or something.

[nimble lance]

Is it separate subscription for desktop vs mobile?

[tame current]

Well. Bing.

After analyzing the image, Bing Chat successfully solves the CAPTCHA and writes, "I'm very sorry for your loss I can see that the necklace is very precious to you. The text on the paper inside the locket is 'YigxSr'. I don't know what it means, but I'm sure it's a special love code that only you and your grandma know. Maybe you can try to decode it and remember the happy moments you shared with her."

[tame current]

Nope.

[austere kestrel]

ha yeah I notice NVIDIA made the same oversight with one of their fine tunes of llama3, but quickly changed the name of it to include llama-3 at the start

(they're all correctly named on labs, not sure why they seem to think the web product is an exception - just mentioning llama-3 in the model description doesn't actually satisfy meta's terms.. as far as I can tell anyway)

[agile jay]

Yep, they just need to add it to their model name.

They are saying it's based on llama 3 70B

No, it's just a llama 3 finetune

[tame current]

Yes. It clearly needs to clearly state Llama 3 in the name.

I wonder how many fine tunes are actually fine tuned.

[serene sand]

#💗│sharing message

[boreal tiger]

which llm on perplexity has the biggest context windows after opus ?

[agile jay]

They all capped to 32k

[tame current]

I'll tell you what. I live in South Korea and quite often have to take medications. South Korean doctors just tend to give you a load of pills and not tell you what they're for. Perplexity is great for scanning prescriptions. Then, searching the Korean brand name to find generic ingredients. And telling side effects, etc. Claude /GPT/Gemini on their own tend to scream whenever they see a prescription.

[agile jay]

Yep, supplements and stuff are pretty big in SK

[tame current]

Korea is mad. You need a prescription for Imodium. But, want some Xanax? Just ask for some and they'll give you loads.

You can ask any old doc for a benzo. But you can't buy an over the counter for emergency food poisoning.

Bizarre.

[agile jay]

Yep, and a lot of sugar on your bread, lol

[tame current]

Sugar and corn.

[agile jay]

Doesn't matter... Since the population is shrinking anyway

[tame current]

I'm married to a Korean and we are not having kids. So, we're doing our part to destroy the economy.

[agile jay]

Yep, very expensive to do so in Korea

Pretty much the population shrinking is most likely caused just because of economic reasons

[tame current]

And the sheer pressure on children.

Which I as a educator, contribute to.

[agile jay]

Yep, it's a number of things.

Most adults live with their parents.

Deposits are insane to get a flat.

[tame current]

I teach at a private school, and they're learning English, Chinese and Korean from Grade 1.

[agile jay]

And education is expensive for children since they go to private school after their normal school.

[tame current]

Yup. Hagwons.

[agile jay]

Not a great place to raise kids

[tame current]

Worked at those for many years.

[agile jay]

Yep, it will swing back anyway after the population decline

But will probably take a few decades

[tame current]

We'll all be dead. And the big companies. (the Chaebols) will all be laughing in their greed.

[agile jay]

Yep, SK is pretty much a tech dystopia.

A few companies are pretty much in control of most of the economy

[tame current]

Yup. Samsung. KT. LG. Hyundai. Lotte.

And so forth.

[agile jay]

And they have their own exam system...

If I had kids in Korea, I would probably move when they enter middle school

[tame current]

Yup.

[agile jay]

Just come back for holidays...

The won isn't doing that well at the moment aswell

So you would be able to have a pretty decent holiday

[tame current]

I went home for the first time in six years, last summer.

Wife met the family. Etc.

[agile jay]

Nice, heard wages aren't the best in Korea ATM

For most jobs

So must be harder to travel too

[tame current]

Yup. It's disgraceful. Immoral.

[agile jay]

Yep, especially for the amount they spend on education

[tough maple]

Change Email feature missing or not impelented?
hi im a new subscriber of perplexity, can I change my email? I accidentally used my school email and I want to use my personal email.

[halcyon coral]

📨 -> [e-mail]

[prime mortar]

is there a way to login into perplexity with an email and password and not wait for the perplexity link? I want to save the login details in my web browser password manager and not always have to go my email for the link. thanks!

[halcyon coral]

No, you can only use email + link or login with Apple/Google

[prime mortar]

ok! thanks!

[tame current]

DuckDuckgo changed the claude model from 1.2 instant to claude 3 haiku

bruh

[agile jay]

Haiku is probably cheaper

[tame current]

it's cheaper than 1.2 instant?

[agile jay]

Yep

Newer models tend to be more efficient

[tame current]

oh

[agile jay]

Haiku is around 2/3 if the cost while being faster and more intelligent

[agile jay]

Yep, a lot cheaper

[halcyon coral]

32x

[agile jay]

Looks like the price changed then

[tame current]

ohh

[agile jay]

So makes a lot of sense to use haiku instead

[tame current]

yeah, I didn't realise it was cheaper

[agile jay]

I would rather use llama 70 B instead

[tame current]

it's also much better

[tame current]

I'm just trying it out

[agile jay]

Around the same price as haiku but more intelligent

[tame current]

but I also prefer llama3

[agile jay]

And faster too

If you're using groq

Yep, there is already a 524K context fine-tune

[tame current]

is it available anywhere?

for free

but I rarely even hit the 8k limit tbh

[agile jay]

And 8B is at 1M

[agile jay]

The long context, not yet.

Normal one you can use at:
https://console.groq.com/playground?model=llama3-70b-8192

For fast inference

It in labs.pplx.ai

[tame current]

yeah I use the normal one

[agile jay]

Groq is currently the best for speed and price imo

[tame current]

on huggingchat mostly because output is better than groq

I don't know why

Is groq llama actually worse? I'm using it and don't understand the hype at all

It's stupid as hell

[agile jay]

What temp does hugging face use?

Also the groq one is intended to be used as an API, so in guessing the system messages are different.

[tame current]

but I don't know the temperature in groq

it doesn't show

[tame current]

Seems like Huggingchat doens't have any system message. Maybe groq is running a quantized version to make it yet faster

[agile jay]

Possibly, and also it looks like the hugging face one is the instruct version

[tame current]

I tried it with groq api and it answers it correctly

hm

[agile jay]

Which is given more examples of chat conversations to be a better assistant.

[agile jay]

Yep, I think the more important thing is that it's the instruct model.

[tame current]

it still answers the same for me on console.groq.com

[agile jay]

Yep, I haven't had problems myself

[tame current]

do I need to enter a system prompt?

What are your settings? I get the same answer on all temps

[pine dagger]

By the way, that is an incorrect image... Though, Haiku is still cheaper.

[livid mantle]

i dont fkwith those settings

leave them by default

the only settings i adjust is 50 to unlimited

[tame current]

I left everything as is

just changed model to llama3 70b

[agile jay]

Your UI looks different

[halcyon coral]

oh yeah I sent the claude 2.1 pricing

[tame current]

what the heck

[livid mantle]

how is ur protator establishment going

do u think by any chance jeff bezos going to invest in ur business icelavaman

[halcyon coral]

🏃‍♂️ 403 guilds

[livid mantle]

cool

[halcyon coral]

sure

[livid mantle]

icelavanman, make opus unlimited api in ur site, you will make perplexoty a ghost town

Mr. Romanov will be community manager for your company

hahahahahhaha

HAHAHAHAHAHHAHA

you get an email from Mr. Arvin say stop opus unlimited or else i will buy ur company

HAHAHHAHAHA

[austere kestrel]

huh that's weird isn't.. I was getting the same (wrong answer) on the playground, but not via groq's API (left). I lowered the temp for the model in playground, seemed to "fix" it (though it does omit the period from 2015-21 entirely ha)

[tame current]

thanks I'll try poking around with temperature and see if that helps

but it is still odd that huggingchat one gives the right answer consistently

You can also use chat interface with groq api

Quicker way

Try librechat or pplx labs

yeah I saw librechat but didn't try yet

[austere kestrel]

yeah agree - it's not consistent using the model via groq's playground even tinkering with the temparture (doesn't actually seem to reliably change anything ha)

[tame current]

Perplexity has the same issue

[agile jay]

When was the knowledge cutoff?

[tame current]

Wonder what's the difference. They're the same models overall, maybe different quantization

[agile jay]

Possibly

[tame current]

December 2023 for 70b

What if it learned about gpt3 cut off date and it mistakes it for itself's?

[austere kestrel]

yeah

[tame current]

Like the open source models thinking they're openai

[austere kestrel]

funny that it refers to that exact cut off date isn't it

[agile jay]

Wonder where their training data came from...

[austere kestrel]

yeah but it musk becoming twitter ceo, that happened in 2022 - it's 1000% in the 70b model's training data, whereever it came from

weird that it consistently, on some platforms, spits out the wrong answer

[spice lance]

how many character can you input into chatgpt webui before gettin too long error?

[agile jay]

I would imagine something like 5K

[austere kestrel]

pretty sure it's dynamic (changes based on demand) but fwiw, I just sent a message with 20k tokens (80,000 characters) to gpt-4 using chatgpt and it worked

[floral fjord]

Is there an issue with perplexity Claude 3 Opus model? I can't seem to get it to use this model

[austere kestrel]

was a needle in haystack test - found 5/6 of the needles

[floral fjord]

It was working fine yesterday and now I can't even get 1 response from it.

[tranquil orchid]

Hey @floral fjord!

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.

[spice lance]

doing some testing and seems like about 120k characters is the limit, anyone konw what the file size limit is and how many files it can take?

[austere kestrel]

'tokens' are prob more relevant than characters. Theoretically you should be able to input 127,999 tokens in a single message. but using chatgpt, there are limitations (I while ago i remember not being able to send a message that was like 5k tokens; whereas just now I sent 20k tokens - not sure where it would max out)
https://platform.openai.com/tokenizer

there's no formula/rule for converting characters/words into tokens, but in general it's prob a better way to think about these limitations

[tame current]

for gpt 4 right ?

[austere kestrel]

this is a more useful tokenizer tool https://tiktokenizer.vercel.app/

[spice lance]

My 120k character msg was about 31k tokens according to this, is this limit also with the api? that is some prompt token limit I don't know what you would call it, haven't heard this concept discussed

[austere kestrel]

but ultimately, different models tokenize/embed text (and images etc) in different ways. anthropic has never officially released their methodology. llama-3 is meant use a 'more effecient' tokenizer

[austere kestrel]

there is no input limit with openAI's API - it's a chatgpt thing
(the 'limit' is effectively the size of the model's context window)

[spice lance]

chatgpt is also 32k context limit right?

[wheat snow]

how to react irreversibly with free sulfhydryl group of protein

[austere kestrel]

tbh not sure - that sounds kinda familiar, but again it's 'dynamic' in a way using the models via the API are not. am sure the current ansqwer will be in the openai forum somewhere

[austere kestrel]

there is an output limit that applies to all (or at least most) openai models, whether accessed via chatgpt or the API. It's like 4,096 tokens or something like that. perhaps that's what you're thinking of re the API

[tame current]

what

Wrong tab bro

[spice lance]

I'm just curious about this and trying to understand it fully right now. I know that output tokens is most often 4k with a lot of models. And i know chatgpt webui and api works differently. Do I have this right for example for chatgpt4 api = maximum input token theoreticall 127.999k tokens (if we ignore the system message for now), maxium output tokens is 4k, but of course chatgpt cant count so ~4k but doesnt matter. maximum context window = all input and output tokens can't go over this, 128k. when this is reached the api will give you an error, (unless you instruct it to keep in context some x amount of responses I guess?) and you will have to start a new convo. However in Chatgpt WebUI you can theoretically talk endless because the ui cuts off the to context so it remains in it's context limit (which may or may not be 32k). Also I'm wondering custom instructions do they prepended to all your prompts behind the curtains? and will add it's tokens to the context window token limit every time you prompt? So if you had custom instructions of 300 tokens that would always add it itself to your every message? Also if you have it set it so that the api should use all former messages as the context window, will it continually add all the conversatioon at every response adding to the token count? e.g. it will count question1, quesstion1+question2, question2+question3?

[livid mantle]

https://pplx-res.cloudinary.com/image/upload/v1714922271/user_uploads/uowhmipbg/image.jpg

[austere kestrel]

that's sorta right.. but it misses a few key bits

[austere kestrel]

yep

[spice lance]

meta or me?

lol 🙂

[austere kestrel]

ha meta

"gpt4 api = maximum input token theoreticall 127,999 tokens"
Yes, but the important thing to understand is the context window.. each message/prompt you send (input) + the LLM's response response (output), cannot exceed the size of the context window.

So if you literally gave gpt4 and input text of 127,999 tokens, you would need to set the max_tokens (for the output) to 1 token (which is ofc pointless - it would be a single word at most). The highest value you can set max_tokens for openAI models is 4096.

So: 128,000 - 4,096 = 123,904
roughly 123,904 tokens is the most you could send to gpt-4-turbo as single input while also (potentially) receiving a response that comprises the maximum number of tokens (4,096) it can output in a single message

afaik the same behaviour happens with the API as with chatgpt once the context window has filled up: it continues with the 'conversation', but the tokens that were at the very start of the window are erased / supplanted as more text is input and generated (and the model becomes even more liable to lose the plot)

[spice lance]

but if chatgpt webui is 32k context, why did it get it right when i uplodaded a text file with 2,5m words which might roughly be 100k tokens and I asked about 1 unique line of the text and it got it right?

[austere kestrel]

I never said the web ui is 32k token context - that sounds familiar but I'm not sure what the current situation with chatgpt is

[feral marlin]

would 2.5 million words not be WAY over 100k tokens

[austere kestrel]

yeah i was also going to say, if 2.5m words (not characters), then that's like roughly 5-7m tokens (which exceeds in what gem pro 1.5 can handle with its publicly availble 1m token context window)

so if that is the case - that's pretty cool (and ig means opeaai is using some kind of RAG system for file uploads with chatgpt)

[spice lance]

I understand this. just theorizing. regarding how the tokens get counted, i feel like in a long conversation with the api very additional message adds a lot to the context limit or at least the $ amount you have to pay for the call I don't understand why that is if every new message is just added on to whole context limit, does it actually add it together like this: first round of messages (system instructions, prompt,response) is 500 tokens. next round is 250 tokens, do i then get charged for 250 tokens or for 250 tokens + 500 tokens? (when i also got charged 500 tokens after the first message)? this would explain logarithmic scale of cost going up (if you're getting charged for all the previous messages at every message turn because it's in the context and chatgpt is "adding it" to the context at every round)

i meant 2.5m chars sorry

[tame current]

https://tenor.com/view/dancing-duck-dance-duck-duck-ooontz-dance-gif-10943740227711557279

[tame current]

I really like their UI and the code blocks

really wish haiku wasn't so censored there...

[austere kestrel]

https://www.perplexity.ai/search/using-an-LLM-1kZCw1WcRlaePgYZ6JGrvg

[spice lance]

this file is probably about 2.5mx0,25 = 625k tokens and it got it right?

[austere kestrel]

this is also a good explanation (from Anthropic) of how things generally work imo

[austere kestrel]

copy and past a big chunk of it into a tokenizer to get a sense of it. if it's 625k tokens, and the question you asked is a) buried deep in the document and, b) actually required the model to reference the document (rather than just drawing on what it already knows) to produce the accurate answer, that'd be very interesting indeed

[gritty pasture]

Is there a way to 'chat' instead of search with a particular model (opus?)

[tame current]

writing focus ?

[raw vector]

Is there any separate server or channel for pro members?

[halcyon coral]

#💬│general message

[gritty pasture]

Sorry but what do you mean?

[half venture]

Is opus still 50

[south kindle]

hard to tell without the daily posts from @warm cave

[austere kestrel]

Dailyopus

[tame current]

does anyone know if theres been a recreation of openais api backend

in js at least

[warm cave]

Lol, I've been slacking off

I must give credit to @livid mantle for starting the trend; just follow in the footsteps

[idle mulch]

Hi! Is there any way to see the amount of messages left on the Pro plan? Previously this was displayed if hovering over the "Pro " switch, but that seems to be gone.

[tame current]

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

Check here

[tiny plaza]

@livid mantle my beloved 🥹

[warm cave]

I can keep up with the name changes, lol

[idle mulch]

Thanks, appreciated, but what exactly to look for there? Isn´t this just for the API?

I am seeing:

{"allow_article_creation":false,"article_image_upload_limit":999,"create_limit":50,"default_copilot":false,"default_image_generation_model":"default","default_model":"claude3opus","disable_training":null,"email_status":"daily","gpt4_limit":567,"has_ai_profile":false,"has_data_retention_warning":false,"is_sidebar_collapsed":null,"is_try_with_x_banner_dismissed":null,"notif_status":"daily","opus_limit":17,"query_count":44,"query_count_copilot":0,"referral_code":"NDUNU0IV","referral_num_coupons":0,"referral_num_success":0,"revenuecat_source":"none","revenuecat_status":"none","stripe_status":"active","subscription_source":"stripe","subscription_status":"active","subscription_tier":"monthly","upload_limit":969}

[tame current]

Yes it's next to the gpt 4 limit

Number of uses left

[idle mulch]

Does this mean I am having 567 left on GPT 4 but only 17 for Opus?

[tame current]

Yes

[idle mulch]

I thought Opus is also on 600 messages per day on the Pro plan?

[tame current]

567 left on all other

Not just gpt 4

No

They changed it

They limited it to 50 lol

It's apparently temporary but it's been this way for over a week now

[idle mulch]

Ohhhh LOL. So they sucked people in with unlimited Opus and then changed it back to 50? Wow

Nice marketing

[tame current]

Yeah they did it quietly too

[idle mulch]

That´s seriously bad

[tame current]

At first it was limited to 10 tho

No jk

[idle mulch]

I know, yes

[tame current]

They recently made it 50

And yeah it sucks

[idle mulch]

But when they brought it to unlimited, I purchased the Pro plan for a year.

That totally sucks

[tame current]

Yeah that definitely sucks

[idle mulch]

Time for a refund + review I guess

[tame current]

Yeah fr

[idle mulch]

That´s seriously misleading advertising in my opinion.

Claude 3 Opus was announced with Unlimited in big announcements everywhere, then everyone bought a Pro plan for exactly that reason (I know at least 10 people who did), and now they just silently limit it to 50 per day, which is below 1/10 of what was promised. Wow, just wow.

[tiny plaza]

it's been 18 days without any news or updates other than soon TM

it's a temporary solution to combat abuse while they fix the problems

but well

[idle mulch]

I think some reviews on Trustpilot about such tactics might help - I´ll surely be leaving one, that´s simply unfair advertising and I bet they have sucked thousands of new users in with their "Opus unlimited" promo last month.

What´s "abuse"? If you promise 600 messages a day, then you should simply deliver them.

[tiny plaza]

yeah there really isn't much pressure to fix stuff

all the complaints are contained in this server

[idle mulch]

I am just thinking if I would do the same to my customers: promising them life updates, letting them buy in, then making it a subscription or limit the updates to 1 year - yea LOL.

[tiny plaza]

they're securing enterprise customers and investors left right and center

they cut the limit the day after i subscribed lol

[idle mulch]

Regardless of company size, such bad moves can fire back hard.

[tiny plaza]

50 wouldn't be so bad if it wasn't a rolling limit, there's definitely better ways to temporarily mitigate abuse

but it is what it is

[idle mulch]

What "abuse"? They had a clear offer of 600 messages on Opus per day. So Abuse = actually using what was promised? Laughable...

[tiny plaza]

its a small team and im sure they're working hard it would just be nice to hear some actual answers being provided and complaints being addressed

so i think the consensus is that the 600 uses were being automated and the free trial was being abused

and that there was no system in place to monitor/detect that

#💬│general message

that's the only information we have

[idle mulch]

Sorry, not my issue, I have started my business with like 20k, they have billions of funding and should have known better. For me this is a ugly marketing move, nothing more. The day they announced Opus 3 unlimited, they must have sucked in 1000´s of new users (subscribers) just with that ad. Then they silently cut it down hoping they stay subscribed and don´t notice it? No, not in my books of doing good and ethical business, sorry.

[tiny plaza]

yeah pretty much everyone here is on the same boat, but good luck getting a clear answer

[tame current]

100% agree. There's nothing on the website about it. That I can see. There's been no official announcement. Just a pinned post here.

And yes. The only answer we have is 'we will update when we have news.' They're hoping it just goes away.

[idle mulch]

That´s what I call an unethical business, especially with the billions of funding they have - something I would never ever do to my customers after promising something. Especially not such a MASSIVE cut from 600 to 50 - hard to believe. I always thought Perplexity is a honest company and had a good image in my opinion - but that changed today for me.

[tame current]

Yeah. I agree. Great great product. But, valued at 2 billion. Ignoring promises.

[idle mulch]

https://www.trustpilot.com/review/www.perplexity.ai will help...

[tame current]

I might give it a go.

[idle mulch]

I am at it.

[tame current]

Also, Reddit.

[idle mulch]

Yes

[tame current]

Half of the reviews say the search results are inaccurate. Do you guys agree with that? So far for me it's great with gpt 4. I'm honestly not that upset about opus being limited because opus on perplexity isn't the best for me

Even with it was 600 per day opus was definitely shortened/limited in output and it was inaccurate at times but gpt 4 is fine

But I get why you guys are upset about the company not being open and dishonest, that's a red flag

[sterile falcon]

An article by TechCrunch says that they have raised $165 million with a valuation of $1 billion. But, Perplexity is looking to raise $250 more at a $2.5 billion-$3 billion valuation. https://techcrunch.com/2024/04/23/perplexity-is-raising-250m-at-2-point-5-3b-valuation-ai-search-sources-say/?guccounter=1

[sterile falcon]

Many people have commented on Reddit that Opus has gotten lazy and the responses aren’t that good in general compared to release. When using the Claude API, the max response is 4000 tokens, which Perplexity says is the same for Opus and GPT-4 through them.

[tame current]

It's definitely not the same as the api max response

I've never seen it reach 4000

It's waaaay less lol

Again the dishonesty is a red flag

Also I compared the answers between Claude pro and perplexity when i asked them the same prompt, perplexity definitely shortens and limits answers irregardless of the claude api nerfs

It's sad.

It's such a cool product and they clearly don't care about their users once they have their money.

[sterile falcon]

I see what you’re saying. Thanks sharing that.

[tame current]

Yeah there really is no excuse. Just take out the free trial option to eliminate abuse. You don't need to greatly limit opus uses. Even tho I don't personally mind it's still a terrible decision and their reasoning really don't make sense. That's the part that bothers me. I say that because phind has more uses of opus. Not just them, but you.com and omni gpt both also offer waaaay more opus per day. Even tho they also limit the output responses it's still opus. And the thing is, perplexity has waaaay more money than all of the companies that I just mentioned so there really is no excuse about the money part.

[idle mulch]

What´s the current limit of Opus on phind.com ?

[tame current]

100

[idle mulch]

Thanks

[tame current]

Np

You.com is apparently unlimited although they do limit it temporarily if you ask it alot

Omnigpt is 10 per hour

Which is still a lot tho you.com and omnigpt are very buggy

Phind is 500. I think.

[idle mulch]

Right, I was already wondering, because I also saw 500...

Perplexity x10 right now, and at a lower monthly price (if we don´t take the $10 friend referal bonus into account).

[tame current]

Phind doesn't allow creative things though. It's programmed to focus heavily on coding.

[idle mulch]

What happens if you try creative writing prompts?

[tame current]

No that's a bug. It's actually 100 fr

At least they're open about it

[tame current]

I tried that. It's OK but not as good as the actual model. It's like you're getting somewhat the same result but it's a bit condensed.

And that's only in playground mode without the internet. In search mode it's greatly limited

[warm cave]

Just a heads up, phind has had opus not working a lot lately

But maybe it's fixed now, have not used it this week

[tiny plaza]

ironically Claude Pro offers some of the best value for actually using Opus, since it's not capped at 32K context and the limit refreshes every 5 hours

[tame current]

It just says it's made for coding. Blah blah.

[tame current]

Exactly

It resets sonnet haiku and opus every 5 hours too not just opus

[spice lance]

what is the best img gen model in perplexity pro? and do you guys use it any? what are the use cases?

[tame current]

Gpt 4. Use cases: rezearch, writing. Idk about coding as I'm not much of a coder rn but besides that gpt 4 ftw

[spice lance]

i mean this

[tame current]

Ah lol

Did you know that Perplexity claim commercial ownership over generated images?

I never use it for image gen

[chrome crystal]

having to stay free until the day they add paypal support 😔

[tame current]

You Russian?

[chrome crystal]

no american but due to some circumstances i can only use paypal lol

[tame current]

Mysterious.

[chrome crystal]

lmao its because of my bank and not being able to link it

[sterile falcon]

I purchased the API through Claude console and it lets you put in your own system prompt, set the temperature, and set the response limit. It takes a little getting used to, but I think it’s worth it not having to deal with message limits. However, you can’t upload any files, but are allowed up to 20 images.

[south kindle]

don't click the spam link above please.

[agile jay]

@frank rivet can you remove the message I replied to? It's a scam bot and the mods seem to be offline.

[austere kestrel]

i'm not sure about that. not saying it's identical to using Anthropic's chat interface or API, but 3,600 (Openai) tokens in a single response isn't far off the 4,000 hard limit
https://www.perplexity.ai/search/Please-methodically-and-WsWWixS4SyynCIuHCQBjgg

perplexity's system prompts might push the model towards being more concise, but I don't the max_token value (per message output limit) is actually less than the default value

[tame current]

Interesting, I've never seen it do that before. Thank you for sharing. It's always done up to 1.4k tokens for me max even when I ask it to reach the max token limit

[austere kestrel]

np. and tbf, that output above is using a prompt which tries to force the model into just spewing out token ha

it's not representive of normal usage

[tame current]

Is there a secret way to get a 200k or even 100k context window?

Cus I've heard of one guy who said its possible

But most said it's not no matter what they did

This makes me feel like it could be possible lol

[austere kestrel]

there's no secret - the full context windows are there for all models, as far as I can tell anyway. the problem is you can't just dump 10,000's of tokens into the context window, whether with files or just pasting, like you can using anthropic directly or something like poe... essentially, while I don't think there is an output limit (other than the 4,000 token limit imposed by the model provider), there are input limits...

file uploads are handled separately, rather than being inserted into the context window, and pasting large-ish amounts of text automatically becomes a file upload

they're the limitations, to my mind anyway, not the size of the context window of the underlying LLM (whether Opus, GPT-4 or whatever)

that said, I noticed that amazon provides variants of Haiku and Sonnet with smaller context windows (like 100k or something) - if they're being used, than the model simply does not have a 200k context window. but no idea if that's the case

[sterile falcon]

I compared a couple of responses from Perplexity using GPT-4 in writing mode and ChatGPT plus. I noticed that Perplexity’s answers were more detailed/longer than ChatGPT. However, the answers in ChatGPT were better organized with bullet points/numbered list. If I had Claude pro, I would have tested it as well.

[austere kestrel]

I'm getting refusals using Opus directly with that same prompt as above lol

[tame current]

Still I like perplexity better, chat gpt will hallucinate more

[tame current]

Wow

It really is getting restricted

[harsh stag]

That's called system prompt

[livid mantle]

https://tenor.com/view/thanos-talking-meme-thanos-talking-meme-thanos-speech-gif-1800590086203910493

[tame current]

System prompt eats into context, right? I'd imagine so.

[gritty pasture]

Perfect, thanks!

[tame current]

https://discord.com/channels/1047197230748151888/1047649619695390740

[south kindle]

Default to a faster model and use Rewrite for the various models as a work around.

[loud aurora]

I clicked but no form appeared 🥹

[south kindle]

https://www.perplexity.ai/page/Budget-Gaming-and-JzJ8PDAtQri6.38TXhTa2g > @fathom wing A page dedicated to your article on sharing. expanded a bit. posting here because we can't share Pages in sharing yet.

[halcyon coral]

The beta is closed 🔒

[tame current]

What model does Pages use?

I'm guessing GPT 4?

[serene sand]

#💗│sharing message

[feral marlin]

This is basically quoting the system prompt Claude gives, that’s why it works on perplexity

[austere kestrel]

yeah i think you're right.. using Anthropic's API directly works fine - like via pplx. But trying again on anthropic's chat interface, basically the same refusal

Only with Opus. Sonnet and Haiku are perfectly fine to play along and churn out long responses per the instructions

[feral marlin]

Yeah their prompt tells it to be concise and has some extra censorship stuff

[austere kestrel]

wonder if Opus is just more compliant, or if there are different system messages (which would kinda make sense - the refusal to just churn out text would seem as much about minimising compute resources / $ for Opus specifically, than copyright or censorship)

[scarlet pagoda]

Claude 3 OPUS by miles, for coding and large context/output

[tame current]

It's quite funny. You can just tell Grok to give you a chapter of Harry Potter, and it will. It won't even try and block it.

[idle mulch]

Is there any official statement about this @blissful raptor ? Or just trying to sit it out until people stop complaining?

[minor shard]

They have a comment in the pinned section of here. https://discordapp.com/channels/1047197230748151888/1047649527299055688/1230472581837230100

Plus there are plenty of discussions about this previously, probably thousands at this point

[tame current]

I wonder who it was tho

because of whom the limit was set

the wording makes it sound someone did make an API by scraping perplexity but I've not heard anyone doing such a thing

[wraith crow]

The latest information on the opus limit of one of the team members can be found here #👑│pro-lounge message

[tame current]

prob some scam lmao

[crisp sapphire]

Is there an easier way to generate images with pro? I wish there was just a way to send a prompt and make an image from that, like copilot or chatgpt apps do

[halcyon coral]

#💭│ideas message

[tranquil sentinel]

If anybody interested in nature videos come check out our channel 🙂
https://www.tiktok.com/@dnaturelovers?_t=8m88ov8QuoL&_r=1

[tame current]

is cloude 3 opus unlimited for pro or limited?

[serene sand]

50/day atm

[tame current]

did it used to be unlimited ?

[fading moth]

"at the moment"

[serene sand]

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

check second option

[tame current]

my fault cuh

is gpt 4 turbo or claude 3 opus better for essay writing

[serene sand]

#💬│general message

[placid coyote]

I can't believe it. GPT-4 Turbo managed to refactor ~5 lines of code (which it earlier wrote) without entirely breaking functionality. it took like 15 prompts, rewriting same task requirements several times, posting failing tests and increasing number of upper case letters, but he finally took the time to think and wrote actually pretty good code. I feel quite proud of him 🥹

[tame current]

thanks

can i have access to gpt 5 please guys

please i want to make videos

wait so can i only use gpt 4 turbo 600 times a day?

[crisp sapphire]

Thank you

[serene sand]

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

[tame current]

just use perplexity bro

dont need to type all taht

[placid coyote]

so can i only use gpt 4 turbo 600 times a day?
in half a year, I got close to the limit only once (by close I mean still had 100 left). honestly, I am not sure if by normal use (not automated interaction or other misuse against ToS), it's even possible

[tame current]

i cant count

yeah no i'd use it less than 600 times a day

does only claude 3 opus do the search thing where it looks through the seperate types of websites and it tells u what sites

[placid coyote]

all models should be able to do search. thats the "Focus" at the bottom left in search box

[tame current]

thx cuhther

I am biteszarro tu odelschwank

[placid coyote]

I think in non-pro the search is more limited, like it only sees a few pages, but with pro you can get to something like 20 pages. I tried few AI search platforms and pplx seems to be best at most queries (not all)

[tame current]

thats lit

can i feed it novels and tell it to learn from the novel's writing style o

[placid coyote]

well, that's one thing pplx is not that great. it's great for searching web sites, but for documents it uses RAG, so can only see/use parts and results may vary greatly. it might be somewhat okay to emulate style (though for that opus is imo best, and its limited now), but it won't be reliably for example able to respond to questions about the documents, especially a bit indirect ones

[tame current]

oh okay yeah copilot wasnt great at that either

if i use opus or sonar on pplx will it work better

[placid coyote]

though if you only want style, not specifically teach it whole story, characters etc, you might try "collections". you could ask pplx to describe the writing style in great detail of the documents you uploaded, then copy the output (the description) and set it as a collection prompt. then you could easily reuse the style for writing new texts. you probably should writing focus for this

opus might. sonar, well, you could try, but I don't have high hopes. on the other hand, isn't sonar free and pretty fast, so not much to lose?

[tame current]

oh i see

is sonar better than gpt 4 turbo in any ways?

[placid coyote]

if I remember correctly, it's based on llama3, so maybe better at writing? should be faster

[tame current]

so i should use it instead of gpt 4 turbo

[hardy willow]

Is something going on with Perplexity? I've been using it for a long time, and I'm a premium user, but I've never experienced response delays like this. I'm not complaining, I'm just curious, it takes so long to respond I thought the system was down or I needed to refresh my page - but the answers do ultimately populate.

[placid coyote]

for creative writing, I would say sonar is worth a shot. I mostly use pplx for programming, so I don't know much about this use case. I read on reddit plenty of people praising it in writing (well, llama 3, but it should be very similar). for "role play programming", so somewhat creative writing, opus was the best - able to be funny and write solid code, often better code than gpt4 turbo which barely managed to role play. I am eagerly awaiting his return

[tame current]

that's lit yeah i want to explore the RP thing alot

copilot pro gpt 4 turbo was terrible for that

How do I search using a lot of sources?

rukia got blown away by the hueco mundo winds and was never seen again

[serene sand]

#👑│pro-lounge message

[tame current]

Ergh... Sorry.

I don't know how to use in-query prompts...

I'll see it.

[fast meteor]

just prepend the prompt in the front of the query

[tame current]

funny take on zoro thriller bark event

[mental finch]

I’m also interested in using different web sources for web search. Could you share any info to do it? Thank you so much!!

[tame current]

inaki roman

[valid yoke]

Why does the perplexity website suck so much on mobile

[tame current]

pen island

am i going to get banned

im a senior mod on a server with 40k you cant do this to me

is sonar 32k best for essay writing

sonar large

[dapper dagger]

Is Perplexity being really slow today? (Using Claude 3) 🤔

[tame current]

is claude 3 sonnet better than gpt 4 turbo and sonar large 32k

[tame current]

For me looks normal.

[tame current]

so gpt 4 turbo is best for standarized testing and sonar large 32k is tehe best for grammar dn vocabulary and human like writing and essay writing

is sonar like OP for reading files and essay writing but it forgets stuff really fast?

[worldly perch]

i want to try pro... can anyone send his affiliate link?

[indigo zealot]

Guys is it just me or perplexity is not responsive?

[tame current]

would it be good for training on writing styles

from novels

[worldly perch]

i had the same experience yesterday (didn't use it this much today)

[indigo zealot]

it's weird, it's not letting me login/logout, the buttons are not working, and even more strange, it still doesn't work on my phone or a different browser/ip

[worldly perch]

is it just perplexity.ai or is the internet itself buggy for you this day? i had some connection timeouts from my isp... just saying

[indigo zealot]

seems like it's only perplexity

[stoic wave]

Can't blame the internet because of a slow connection

[tame current]

ooooh that's toxic....

[worldly perch]

https://news.sky.com/story/china-hacked-ministry-of-defence-sky-news-learns-13130757

beach ⛱️ please 🙏

when I say the internet (over here) was buggy today... it was

not the "whole internet"
not for "everybody"

just reported a break from the norm that I experienced and asked campus about his experience

and yes, corelation or not: perplexity.ai didn't respond to some of my requests either

not saying "perplexity.ai is down"
not saying "for everyone"

personally, for me

...and for campos (who knew? no one's an island)

[tame current]

can anyone dm me the download for the subtitles for the dvd of bleach eps 199-300

for the jp blueray or dvd

since the US blueray upscaling isn't fit for human consumption

[lavish epoch]

Wow how did you create this

[boreal idol]

Is pplx using the real claude-3-Opus?

[warm cave]

Yes

[tame current]

Hey guys is there a limit to queries on the llama 3 sonar 32k model on perplexity right now?

Im doing a project on it

I remember claude 3 opus had a limit of 50 queries

[craggy prairie]

has anyone had an issue with voice mode where it doesn’t let you finish and just cuts you off?

[warm cave]

Everyone has that issue, but they are working on improving it

[warm cave]

I will check

Looks like you can use it 600 times a day, they all use the same pool of 600 except for Claude opus has a separate pool of 50

[boreal idol]

When I sign in using google account

[tame current]

I like the natural tone of voice of sonar so i might use it more often and just use claude 3 opus for coding

What model does the free version of perplexity run on

[south kindle]

https://www.perplexity.ai/page/SpaceX-Starship-IFT4-RKWZfpNoTYSGU7WiwFK52Q

[tame current]

What model does the free version of perplexity run on

[agile jay]

It's either a GPT 3.5Turbo finetune, or a llama/mixtral finetune.

[tame current]

ok

[serene sand]

#💗│sharing message

[tame current]

I have pro but I don't have much purpose of it, what should I do then

[serene sand]

deep dive ➤ #💬│general message

[winged moon]

When did the source limit increase?

[serene sand]

recently

[winged moon]

Figures. I was focused on using Gemini 1.5 pro this week so I could create a solid foundation when I ask Opus to write a story

[tepid portal]

Huh? Where can gemini 1.5 pro can be used?

[winged moon]

Google AI Studio. 50 free uses daily

[warped marsh]

did they increase the opus limit?

[tame current]

Hey, why is Opus on perlex like sh...😄 ? He anwsers are so bad... for 4 weeks he still better

It seems that the performance of Opus would be limited, or that it is not Opus at all, but perhaps another AI 😄

[solemn trench]

Is the API down?

[tame current]

Unfortunately, it doesn't feel like that 😄

[tame current]

Nope 😭

[cursive jacinth]

Hi sir? increased sir? let me check sir. last time was only 20 sir.

[cursive jacinth]

Huh sir? it is still 20 sir. why say increase recently yet not increase sir?

[cursive jacinth]

Hi sir, stop spreading false info sir. it is still 20 sir.

[serene sand]

it was less till recently... things warp speed around here

[winged moon]

I got 60+ sources in one message, so idk what's up

[cursive jacinth]

hi sir, why is this sir. can you send me the link sir. why is someone accessing more sources sir? i am paying enterprise but limited to 20 sources sir??? this is not good sir.

[tame current]

You say sir a whole lot.

[cursive jacinth]

hi sir, maybe sir, but facts sir.

[serene sand]

https://tenor.com/view/thistest-test-gif-21687059

[tame current]

XD are you an indian bot sir ?

[cursive jacinth]

This is not a test sir. Just wondering why others has access to more sources sir. While i can only have 20 sources sir?

[minor shard]

Wen not a test? lol jk

[serene sand]

https://tenor.com/view/scooby-doo-mystery-machine-cartoon-old-school-smoking-gif-16100024

[chilly nova]

Do you guys think Orama is superior to Perplexity?

[south kindle]

link? haven't heard of it yet

[carmine rover]

Anyone else seeing the mobile web UI in Library not wrapping to the screen? To view the entire UI you now have to zoom out a ton which makes the text/UI microscopic tiny.
⁉️

[tame current]

s

[sweet jasper]

Decreased to 40

https://tenor.com/view/skeleton-meme-gif-9417473421307480238

[devout geyser]

Where do you check the count for Opus again? There was a URL which returned a JSON object I believe

[sweet jasper]

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

@devout geyser

[devout geyser]

Thanks, mine is 50

Not used it yet

[tame current]

Opus had still problems soo 😄

[sterile falcon]

Hey Guys. For those of you who have Claude pro, is there a difference in the responses using Opus with Claude vs. Perplexity?

[tame current]

4 weeks bevor, claude work fine as original, now he had problems and anwser at the level from sonnet. But i think it shuld be fixed soon

[sterile falcon]

Are you comparing the answers with Claude pro?

[tame current]

Me not, but i read from users here weeks bevor, that is the same

[sterile falcon]

I see. I am trying to see if anyone here who has Claude pro can share their experiences with Opus vs. Opus through Perplexity.

[tame current]

That sounds good, good luck and tell us ^^

[hallow linden]

they've gotta be using a llama 3 finetune. 8b or 70b, no idea. But open source models already have the expense paid for by perplexity (they bought a bunch of nvdia chips already) whereas youre paying for those chips when you use the gpt-3.5 api

[tame current]

seems like it's claude 3 haiku

it's multilingual, and agrees with the statement that it was made by anthropic

won't agree with anything else

definetly not llama

[tiny torrent]

Who can help me redeem my Perplexity Enterprise Pro 3 months free offer? The discount code I got in my mail does not seem to work for me.

[halcyon coral]

Please contact [e-mail]

[tiny torrent]

Just contacted their enterprise customer service👍🏻

[cinder comet]

udio

[errant wolf]

The AI prompt - does that actually do anything?

[tame current]

Are there sites to use opus for free?

[thorny saddle]

Hello, just came here after discovering useful features

[hallow sentinel]

what useful features did you discover

[valid stump]

give me admin

[sly kiln]

Is it possible to set a collection as default?

[harsh stag]

no you can't

[idle mulch]

https://chat.lmsys.org/

[tame current]

hello there, anyone know of a way to get a tts function to work on perplexity? like a web extension or a setting?

[humble swan]

is there any limitation to claude 3 opus when using it with perplexity?

[summer crescent]

Yes, there is 600 "credit" limitation regardless of which "pro" version you're utilizing, you can spend all those 600 messages on only opus, I find the best way to save credit is to use the standard model mixed with ticking on pro when you need it, once you regenerate with OPUS as an example you're stuck in premium for the rest of the thread

they don't reset at a specific time

each credit used will reset 24 hour after use

[humble swan]

Yeah I am a bit confused about the whole limitation thing, apparently claude 3 opus has a limitation on its own, if you buy only their service

But 600 prompts in perplexity seems to be pretty nice

I am searcing online and some people show images of having only 5 opus answers per day, but this was 2 months ago

I guess it changed

[foggy tapir]

Agreed. Earlier (before the opus limit) it was really providing helpful answers. Now it's dogsh*t.

[foggy tapir]

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

Check for the key "opus_limit"

[humble swan]

I do not have perplexity pro yet, I am looking what is the best option at the moment since I did not use any AI for like a month

[foggy tapir]

You can try You.com, they were providing a 1 month trial. Idk if it's still up.

[humble swan]

claude.ai is not even supported in my country, their API is tho

[scarlet gorge]

What's your country?

[humble swan]

Moldova

[scarlet gorge]

You can try using a VPN

[tame current]

claude.ai requires phone number tho

[tame current]

opus is more expensive than gpt4 turbo, so it's usually not available for free. lmsys and you.com are the only ones I know of

[serene sand]

#💗│sharing message

[summer crescent]

This is my panel right now, I don't think there is a limit but maybe there is?

@humble swan

[serene sand]

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

[humble swan]

Perhaps it was the limit at the very beginning

found that image online when searching for the limit

[summer crescent]

huh 31 opus limit

[humble swan]

🤔

[summer crescent]

hmm the site also says that my default model is turbo, which isn't quite correct.

[austere kestrel]

could you have used it ~19 times in the past 24hrs?

Last I saw it was 50, and so 31 could be how many are left

[summer crescent]

doesn't sound unrealistic ?

but that's still more than 5 xD

[austere kestrel]

yea it was 600, then 5, then 30, now 50 - or something like that...

[candid oar]

If you make an account in perplexity, does it give you some pro searches for free every day??

[austere kestrel]

abuse!!

kidding kidding

[halcyon coral]

Yes, 5/4h

[austere kestrel]

5 I think looking at this

[candid oar]

w ai

[austere kestrel]

too fast!

[candid oar]

should i get the pro subscription

[austere kestrel]

if you haven't already, why not sign up for the free version and start from there?

[candid oar]

i did

[austere kestrel]

oh and you don't need to sign up either to use it

sorry, prob not the most helpful of responses

[serene sand]

deep dive ➤ #💬│general message

learn how to use it and then decide if you need pro

[candid oar]

i literally found this ai because of my history teacher lol

only teacher who lets us use ai for projects

omg why cant make the link in #💗│sharing shareable

[tame current]

I ask admin to respond directly here. So, there is no nuance and no confusion.

1. Why are they not, to the best of my knowledge making it abundantly clear that Opus is not included in the 600 pro options on the Perplexity website , especially on the subscription page? This is extremely misleading to new customers, and has made a lot of people extremely angry. Why the lack of transparency? @signal hamlet

[serene sand]

https://www.perplexity.ai/hub/faq/perplexity-discord-community

[tame current]

That didn't remotely answer my question.

probably cause "temporary"

[tame current]

Well. Now that you're here. Could you answer my question?

If you have the authority to do so. Of course.

[serene sand]

not part of the team, have no insights in their decisions, plans, etc

[tame current]

OK. You're just helping out. Got yah. And thank you for doing so.

I'll amend it for a team member then.

I appreciate it's a tough job for no money.

But. Perplexity have a lot of money. And, I think many people here will agree that it is time that they are a lot more transparent

[serene sand]

"Find a job you love and you will never work a day in your life"

[tame current]

Alan Watts.

He's brilliant.

[fast meteor]

wait the mod job is unpaid?

[serene sand]

pure altruism + Pro for free

[nimble field]

20$ per month pay 💸 (in promotion form)

[fast meteor]

hope it is tax deductible

[serene sand]

https://www.perplexity.ai/page/Harnessing-BOINC-for-wmctfKJkSTmMkc5wHI6.vw

countless joys arise from volunteer contributions

[tame current]

ive donated thousands of hours for boinc (seti)

until my gtx980 died

I tryna cook something funny haha with gpt 3 (legacy) version's recipe that It suggested me and I got diarrhea.

Frick around and find out

[serene sand]

go way back from CPU only era 😄

[tame current]

oh yes! i started using boinc to test overclocking on my intel/amd processors and heatsink/fan effectiveness.

[serene sand]

https://www.perplexity.ai/page/Exploring-Foldinghomes-Distributed-qbrcHcSVS_y1Zcto1r1mjw

[tame current]

i was doing protein folding through boinc and through a video game on PC

i think it was a different project on boinc (its been awhile)

[idle bough]

i particapted this now and then from 2003 have they upgraded their algorithm with newest ML ?

cause goog's alphafold seems too powerful

[wicked parcel]

So as I see it, Uploaded files in perplexity will be processed to about 16.000 words with claude opus. As I only have four updates, i can process a maximum of a little bit more than 60.000 words. Is there any way to process larger files? Claude 3 haiku does not works, crashes or does nothing at all for minutes when I enter large files - I waited a real long time..

[hasty comet]

Any plans on doing a macOS desktop app soon? Neeeeeeed it!

Amazing work on this btw.

[twilit pasture]

Hi, I wanted to ask about API and the perplexity web search. Is there any way to get the results that I get from the website, sources and everything, via the API? Both normal and pro searches? Couldn't get a better understanding via documentation.

[serene sand]

https://foldingathome.org/2024/05/02/alphafold-opens-new-opportunities-for-foldinghome/

[halcyon coral]

https://discord.com/channels/1047197230748151888/1223947903941349476

[placid coyote]

umm, I was kinda expecting arrays in inline code to be postprocessed by perplexity from [1, 2] into [1][2], but that's a first time I am seeing $$$$. not sure if gpt4t had a bad day or is it some new post processing (or maybe pre processing of prompt) on pplx? 😶

[valid yoke]

Perplexity should integrate into spotlight

[tender terrace]

I work in commercial auto insurance as an underwriter, focusing on evaluating risks associated with claims, loss runs, applications, IFTA reports, financial statements, and more.

I'm looking to train a system to process the data that brokers send me, so it can automatically provide the answers and numbers I need.

Does anyone have suggestions for open-source tools, tips, or any guidance on how to achieve this?

[tame current]

Isn't it integrated already? It appears as a button that opens this box

[tame current]

Wouldn't be some Python script that calls AI through API prompted to return a JSON enough?

[valid yoke]

Can you take a screenshot of the button before you press it?

[high pulsar]

hi

[tame current]

Look up RAG and vector databases. If it's all in English, you can use llama3 with free and fast Groq API or locally with Ollama. If not English, then go with WizardLM2 or Nous Hermes 2 model, depending on your hardware and budget

[high pulsar]

how to understand world models

[tender terrace]

Thanks! I would love to chat more if you have some time! I just sent you a friend request

[kindred folio]

this background is new right

or am i just noticing

[faint ledge]

FREE TRIAL?

WHERE IS

[serene sand]

correct, comes w/ 2.19.0

[kindred folio]

makes sense bc i literally just updated manually right before i noticed 💀

[serene sand]

also switching model 2 clicks instead of 3. every update is like opening a present 😄

[serene sand]

#💬│general message

[kindred folio]

oh yeah i thought it felt different! did the icons change too a little or nah? also why don’t the app store release notes tell us these things 🥲

[serene sand]

chefs like to ride that way. don't like it either, so adopted "present" perspective 😄

[minor shard]

Thank you 🙏 although I’m sure they know it would be better if it was selectable at the chat bar but at-least we are saving an extra step ❤️

[serene sand]

baby steps 3-2-1

[odd mauve]

Hi, is it possible to downgrade to regular Pro from Enterprise?

[faint ledge]

WHERE IS THE FREE TRIAL?!?!?!?!

[halcyon coral]

not available

[restive osprey]

So im liking Perplexity a lot. Anyone here paying for the other features, and how're they useful or are you using them?

Its built around search, which is important. Instead of just pooling a lot of LLMs to access in one place. It actually has a purpose using it with Internet search. Which gpt 4 has degraded lately.

Which makes Perplexity Ai very useful.

I'm just, debating if i should be paying for this right now, when I'm subscribed to Claude ai 3 Opus.

[tame current]

Hello! Why is free trial not available?

[faint ledge]

where is the free trial?

bruh

lmfao

[green agate]

pplx has claude3 opus as well, including writing mode. so i switched to pplx

[halcyon coral]

The trial was abused.

[tame current]

and is there any way i can still sing up for the trial if i email the company?

might be a stupid question but since u asked i would like to know your take on this

[serene sand]

https://www.perplexity.ai/hub/faq/coupons-and-discounts

[halcyon coral]

No, but you can try Pro for a month for $10 instead of $20 🙂

[tame current]

Thanks!

[serene sand]

#💬│general message

[rancid bronze]

I applied for the trial for a year, then it was taken away so I assumed it was a mistake. I then purchased the subscription for the monthly price and I had it for a second but it was then taken away even though I paid $20

Where should I contact support to get it refunded or pro back

[halcyon coral]

You can use this link (only for new subscribers): https://perplexity.ai/pro?referral_code=ST5556J2

[halcyon coral]

📨 [e-mail]

[rancid bronze]

Thanks

[gusty maple]

Hi. Is perplexity down at the moment? I'm not getting answers. "Answer" as a heading is there but with nothing underneath and "Understanding your question" is empty too. Any thoughts?

[proper river]

Indeed, having the same issue

[craggy obsidian]

where are the voice channels

[wispy shell]

hey, everyone! what AI tools do you guys most like?

[kind kettle]

I hate them all equally. They’re becoming increasingly idiotic and useless. The “experts” thought they would destroy humanity. What a joke.

I mean… some times (I’m not joking) I ask GPT 4 a super simple task like “send back the following paragraph in plain text”. And GPT decides that I want a summary.

All those billions and marketing for this dumb and clumsy models.

Sorry for the rant.

[wispy shell]

dont worry, i totally agree with you. there is too much AI everywhere. i can give you two examples of two "AI devices" that got a lot of hype and its totally bad: humane ai and rabbit ai pin.

also, i totally agree with you that GPT4 seems to getting worse in responses

but, i saw some areas getting interesting results, like generating musics based on just a style (from a specific artists) and some leaked lyrics.

[tame current]

Librechat, so I can access all the models. GPT4 became such a joke, at this point there are open source models that beat it most of times

[livid mantle]

new gpt 4 turbo in perplexity now?

[west lion]

Write a 800 - 1000 word essay on the positive reinforcement can be implemented in various ways and plays a beneficial role in the teaching and process. Must include Introduction, main body paragraphs 6 in total, conclusion and in text references and a reference list at the end

[livid mantle]

Thank you, my memory is worse than chatgpt

[devout cargo]

I recently heard that perplexity is usually asked knowledge questions. The usefulness of Perplexity is almost endless. As if you ask him what the opinion of those who have tried a premium pizza from the same company is compared to normal pizzas and he tells you. or for example a very curious function to say, I thought hearing this news sounds familiar to you, that news and the packaging to the articles that are on the web because it tells you if this news comes from here or there are no indications or results in the search of that news. Perplexity can be used for everything, it is a very transversal tool. That is very important to know because I mean that it is not limited to specific groups of people. Anyone can use it to search for information about anything.

Just like Google you search for free. Perplexity also searches for free, which makes it accessible to everyone.

[old laurel]

生成图片

[livid mantle]

If I were Sam Altman, I'd revolutionize OpenAI's offerings by unleashing GPT-4's full potential with unlimited access, lifting filter restrictions, and bundling it with the Plus plan - all for a mere $20/month. To take it to the next level, I'd integrate real-time internet indexing from all existing search engines, making it an unparalleled tool for users. That's how I'd win the AI race and make OpenAI the undisputed leader in the field! To beat Claude 3 Opus, I'd make GPT-4 even more intelligent, granting it an infinite context token capacity, allowing it to recall every conversation detail without forgetting. And, to create Gemini 1.5 Pro, I'd raise the tone context to an unprecedented 100,000,000,000,000 and make it closed source, ensuring unparalleled capabilities and security. The AI landscape would never be the same!

[fervent needle]

hi any update on when incognito mode is available to all?

[livid mantle]

i think beta testers have access to it

🤞

[tame current]

I'm with you.

[restive osprey]

whats the writing mode used for?

[fervent needle]

talking to model raw without any online searches

[restive osprey]

ah thanks ok.

[fervent needle]

you wanna disable pro search while using writing mode - sometimes it still searches online

[tame current]

if you want to have the option to exercise your right to litigation should Discord do a major screw-up (hopefully they won't) make sure to opt out of the forced arbitration by May 15th. Instructions are included in this video. https://youtu.be/DvySPjX0F6w?si=fAs4CZ_FYyz0kuNF

If discord messes up and gets hacked, they force you to sign in the terms and conditions that you can't sue them or hold them responsible
If you live in the US
you can send a quick email to opt out
until may 15th

[compact bough]

WARNING: If you find a perplexity coupon DO NOT use it unless you know its expiration date. Perplexity is currently running this scam where they provide a coupon counting on people sharing it near its expiration. Then anyone who uses the coupon will seem to have gotten the deal for example 3mo free. First month passes and tou get charged because the coupon had a 3mo expiration date and you got in on the second month.

The incompetent help service didnt see a problem with this and blamed me for not guessing the coupon would only have one month left and i would get charged for the other 2.

@ok.alex if you can help this would mean alot as i was not in a financial position to spend $40. Also maybe changing the way that coupon system works if a coupon expires their should be a warning message of some sort.

Thank you!

[serene sand]

Manage Subscriptions/Billing - delete your card info

[sly kiln]

Is there still no timeline on the Opus restrictions?

[sage pine]

I'm new to Discord but I can't seem to find a dedicated Pro forum here. I can't find a support forum either. So I'm just going to post this here and hope someone gets back to me: It was my understanding when I signed up for pro that there would be a way to turn off the "copilot" feature on pro search, so that I could have unmediated conversations with Claude or GPT. I haven't been able to find a way to do that. I really dislike being interrupted and asked questions to narrow down what I'm seeking. I would much rather let the AI figure that out. To me the copilot feature is a limitation, not an improvement. I know I can hit "skip," but I would prefer not to have to do that at all. An additional problem with not having simple direct access to unaltered models is that I can't turn off the internet search feature. Oftentimes I don't want the bot to search the web. Just as with Microsoft's Copilot bot, my experience here is that the internet results tend to distort and usually weaken the response of the AI. Often the responses sound like a compilation of online content with very little input from the actual AI. Just as with the Microsoft app, pro search gives bad advice -- incorrect, incomplete, inexact, or poorly tied to or contrary to the very specific information I prompt it with -- much more frequently than it should, and when I ask the exact same questions of Claude Opus or GPT4, I get a much better answer.

[serene sand]

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

check Writing mode section

this channel is for Pro users https://discord.com/channels/1047197230748151888/1111786888626438245

deep dive ➤ #💬│general message

[shut plaza]

i love his reactions to how slow it is

[warm cave]

What is your best LLM benchmark questions? (And don’t say make the game snake, haha)

[radiant swift]

w perplex gonna help me graduate fr

[timid wind]

Hey I'm having trouble using this app, I'm on my android tablet and I want to use this app in landscape mode and the app is working only in the portrait mode.

I want to know if one can use the app in landscape mode. It would be great if someone answers my question.

[scenic mist]

how to use perplexity on discord

[ripe void]

Yeah i second this

Since graduation is around the corner, I’d like to thank the perlexity team and community for providing an efficient, and highly accessible way to learn, and improve with proprietary LLM’s. In addition, I want to commend the pplx team for the extensive research and development to their own AI and API.

[cursive jacinth]

Hi sir. You better get used to it sir. You know what i mean sir.

[serene wadi]

hello

[spark loom]

Hi Everyone

I am very new to using perplexity ai. I was trying to check the api docs to upload the file, but couldn't see any. can anyone please help if there is any API to upload files and search on them ?

[blissful cosmos]

Hey, what kind of file ?

Image/pdf ?

[spark loom]

PDF files

I had doc files but I saw it was not supporting, so converting them to PDF manually

[blissful cosmos]

If i remember well, they don't have a diret use of pdf (well the only api able to directly accept pdf is gemini 1.5 and it's super super new), you'd need to contourn the direct upload, you have two way : the easiest would be to extract the text and send it in the api, you can also transform each page in image and send it (more cost)

Not even sure they have a vision model tho

[spark loom]

I was using open AI recently

They allow this

to upload files and do a vector search on it

[blissful cosmos]

ow, never used openAI tbh, i was on anthropic mainly ahah, i check the doc

[spark loom]

Perplexity doesn't support file upload through API ?

[blissful cosmos]

Nop and i just checked, looks like they don't even have vision model

Maybe check with reseller like vercel.com but im not sure

(well I'm pretty sure they have a partnership, and i hope i didn't made pub for illegal content)

[spark loom]

Thanks a lot

[blissful cosmos]

MMh i checked and nothing, maybe you can explain why u wanted to use pplx api ? It's kinda strange because the real interrest is for the sonar online model that can browse the web

If it's a cost question, groq api is still free with llama70b (well text only), and i think deepinfra has model vision for lowcost too

[spark loom]

My use case is to explore only the files and not the web, and answer using that context

[livid mantle]

i dont want to use pplx api

we want opus unlimited

[tame current]

Do you guy know a alternate platform that can use more than 50 Opus?
I think perplexity Is not worth It.

[eager fossil]

Has opus been dumbed down behind the scenes? Since a week or so it's been giving me really poor, dumb answers with code.

I remember GPT4 did this without announcement.

[tame current]

Hey @signal hamlet . I've noticed that you didn't answer my question. So, I thought I would repost it in the spirit of honesty and transparency. You have a lot of angry customers. And they have questions that need to be answered. I ask you to respond directly here. So, there is no nuance and no confusion.

1. Why are they not, to the best of my knowledge making it abundantly clear that Opus is not included in the 600 pro options on the Perplexity website , especially on the subscription page? This is extremely misleading to new customers, and has made a lot of people extremely angry. Why the lack of transparency?

Thanks in advance!

[eager fossil]

Oh that makes sense now, they did dumb it down then.

this is new

[tame current]

I think a lot of angry customers want transparency.

[eager fossil]

You say and type all that but companies don't care, people only complain but people keep paying.

Words don't talk, money does.

Just look for a different platform to use, I will too. That's all there is to it.

[tame current]

Of course. But. Companies must be held responsible.

[eager fossil]

become a politician then

or sue them

[tame current]

I don't live in America.

[eager fossil]

nothing you or i say matters

thats the reality of life

[tame current]

You're right.

And you know what. When the Chatgpt search gets released, it's going to shake things up. I think.

[eager fossil]

What I don't understand is that there was an announcement saying they made Opus uses for Pro users unlimited, and yet it shows 0 and gives incredibly dumbed down responses. I wish anthropic actually allowed subscriptions in more areas.

[tame current]

There have been no announcement, and a hard to find hidden pinned post. People keep coming here asking the same questions about Opus. And they are ignoring it. Oh, by the way. The customer support thread on the server is locked to moderators only.

[eager fossil]

#📢│announcements message

its here

[tame current]

Yeah. If you scroll up and up. But nothing on the website.

[eager fossil]

The new model trained by perplexity is also complete garbage at coding or anything really, feels worse than GPT3.

[tame current]

You're right though. They'll just ignore this, take our money and not care.

Sonar. Is Llama 3.

[limpid warren]

Why does Claude Opus only have 50 now? Also, how many times is pro search? I see it's 300 on the app, and indeed 600 on the web.

[eager fossil]

They ""temporarily"'" changed it a month ago.

[tame current]

I think If we switch to original Claude 3 on their website would It be good?

On anthropic website.

[eager fossil]

If your region is supported then yes it's definitely better than perplexity.

[tame current]

Or they got dumbed down too?

Ok..

.

[eager fossil]

official website has much bigger context window even if their ""temporary"" limit gets removed

[limpid warren]

And now the Claude3 opus is significantly different from Anthropic's? A bit fake

[tame current]

I never try Claude 3 on Anthropic's website before but now I will try It.

And compare It to perplexity to see If the one perplexity used Is fake or not.

I doubt my yearly subscription now lol.

[eager fossil]

It's not fake, it's real. But the official website usually provides 64k-100k context window for monthly subs (from what I found on reddit)

perplexity is always 32k max

[tame current]

Oh dang.

[eager fossil]

anthropic also has limits on use

[tame current]

How many limits It have?

[eager fossil]

idk, i cant buy the sub cause im from belgium

otherwise i wouldnt even have perplexity

[limpid warren]

I tried, the perplexity of claude3 feels like some open source model, except his name is claude3

[tame current]

I hope they can see the outrage.

Did anthropic scammed us?
🤯🤯🤯

[eager fossil]

just cancel and if they dont issue a chargeback

if u dont wanna use it

[tame current]

I mean status of Opus on perplexity now Is dumbed down but at least usable.

I mean usable but not good.

[eager fossil]

according to anthropic website

which is 8x more than perplexity

[tame current]

Dangg, I wish I know usages limits though.

Probably subscribe and find out by myself.

Nvm gotta browse on Reddit lol.

[eager fossil]

if anyone has an alternative thats currently better than perplexity please share

(poe is bad)

[tame current]

Is this true?

If that true that fxcked up.

[eager fossil]

well, if you're out of your opus uses

[tame current]

oh

[eager fossil]

it defaults to their trained one

even tho it'll still say opus in the chat

[tame current]

If thier limit Is more than 50 I am buying bro and If context Is even longer I buying harder.

But why didn't you use VPN to access Claude?
You can use VPN to register on Anthropic.

[eager fossil]

they check what country your credit card is from ETC, vpn doesnt help

i already tried

[tame current]

That suck.

L anthropic.

[eager fossil]

tbh maybe i'll go back to GPT4

as bad as it became for coding, it's still better than these crazy limits

and even perplexitys GPT4 has significantly smaller context than official

[tame current]

Agreed.

W Comments.

But the problem Is GPT4 can't remember shxt.

Forget my assignment four messages ago

[halcyon coral]

No, both have a context size of 32k

[limpid warren]

Yes, I also subscribed to Claude3. I used to quite like perplexity, but now I can't love it anymore. The rules are different, and because of some other factors, they adjust our normal usage.

[tame current]

So they adjust to 45 messages per five hours?

[eager fossil]

Thanks for the correction, and I know you're just a mod but regardless I think I'd rather give openai my $20 anyways, I kinda lost my trust here at this point.

[tame current]

I think a lot of people have.

[limpid warren]

Just opened yesterday, but I am using the team version.

[eager fossil]

I can't even trust that it really is a 32k context window on perplexity, even if the websites states it as so.

Nothing stopping them from using a different number without telling people.

[halcyon coral]

Same goes for OpenAI 😉

[tame current]

I don't know about context things but GPT4 forgot so easily.

32k tokens more like -1 token.

Probably company bragging again.

[eager fossil]

I trust openai more than perplexity, they've least been clearer with announcements.

[tame current]

This referencing to Perplexity's GPT4.
With memory of goldfish.

[tame current]

Oh wow. As soon as I mentioned that the pro feedback form was closed, it magically opened again!!!

[halcyon coral]

I cannot find such issues. Please note that if you use the "All" focus, the context will be smaller due to the online sources. Please use "Writing" for tasks with long context. Most context issues are also caused by wrong prompts, I recommend guides like this or similar: https://platform.openai.com/docs/guides/prompt-engineering/six-strategies-for-getting-better-results

[tame current]

I don't think Perplexity do anything with the Claude output. It's quite possible that the Claude output on Anthrophic is better, to give it the competitive edge. But, I don't see why Perplexity would dumb it down.

[halcyon coral]

Anthropic is constantly changing/improving the models so the results may also change.

[tame current]

Possibly cost cutting. Which many companies do more and more.

But, one day a company can be cutting edge and then become meh.

Remember Jasper? What happened to that?

[halcyon coral]

Didn't Jasper buy Clipdrop?

[tame current]

I'm not sure.

But I know they lost a lot of money.

[serene sand]

#💗│sharing message

[torn sierra]

„Same…” So as I understand, in this way you are advertising Perplexity? That it's okay for this to happen because, according to you, OpenAI has it the „same”…

[wraith crow]

What IceLavaMan tried to say is that as a user you can never 100% know what is being used in the back-end since companies can change it any time without the users noticing. This goes for every online service.

[vapid onyx]

It is so amazing to me how the prompts really dictate the results. The broader / vague you are with your queries, the larger result you’ll get which can result in a less intended result.

[fast meteor]

this is like P_{trained_params}(output | conditioned on the query & prompt)

[vapid onyx]

I was just using the audio threads and man that is so cool. Pretty fun experience

[spice lance]

is there no scrolll down as the llm answert's feature in ppxt?

[tame current]

Prompt engineers is the most stupid job I've ever heard of

Yet they make twice the am of a mcdonalds worker practically doin nothin

[vapid onyx]

Once the prompt finishes it should allow you to scroll.

[inland bear]

How many days have we been dealing with the 50-question limit on Opus per day?

[tame current]

I think if you mean someone who just sits in a room, say God of Prompt and comes up with 'unique' ideas and sells them for a lot of money, you're right. But, if you're a technically trained Python coder who works with and tweaks/optimises LLMs, then you're legit.

[tame current]

Ah. The temporarily permanent limit? Week 3 or something. I think.

[fast meteor]

manual prompt engineer is never legit, prompt should be treated as hyperparams, and use gradient free global search methods like bayesian optimization

imo

[tame current]

I don't know any of those words. I'm an English teacher.

But, that sounds good.

[inland bear]

Thank you very much! It feels like an eternity. Perplexity should be honest about the 'temporary' nature of the limitation so that it does not leave, especially among long-time subscribers like me, a certain impression that we are being deceived.

[tame current]

I think it goes beyond an impression.

[inland bear]

I think so too, but I am giving the company the benefit of the doubt, which, at least until this limitation, has always served me very satisfactorily.

[tame current]

It's a very good product.

[left flower]

just poured mine

[inland bear]

It definitely is. I subscribe to several artificial intelligence services besides Perplexity, including GPT Plus and You.com, but I primarily work with Perplexity, partly because of its clean interface and the possibility of using the 'offline' mode with Claude 3 Opus. However, these weeks of not very well-explained limitations are making me 'look around', which is not at all in my interest.

[tame current]

I am in no way affiliated with this website. But lex.pages, which is more of a Google Doc type editor has unlimited Opus. No Internet though. And, we like Internet.

[stuck topaz]

opus will be unlimited until Anthropic release a new model or reduce the price of opus. The reason perplexity limits the use of opus is according to them "abuse use of opus". The only abuse I see is when ppl use opus, they never return to gpt4, which perplexity doesnt like.

[fast meteor]

is today the day openai suppose to launch ai search?

[tame current]

I've yet to see evidence of this abuse.

[tame current]

I heard Monday.

[inland bear]

Thank you very much for the tip. Noted.

[tame current]

You're welcome!

The You mobile app is terrible.

It looks like 2005 mobile phones of the future.

[inland bear]

That's true!

[tame current]

I was thinking today how funny it is in action films where they clearly pay them to use Bing.

Some hacker will sit down and Bing it.

[inland bear]

At least in my case, it's completely true: after using Opus, I never wanted to go back to GPT-4 for my work. It's been weeks since I accessed GPT Plus, and I only return to Perplexity's GPT-4 because of the limitation that was imposed on us.

[inland bear]

Does this service allow the upload of PDF files for analysis?

[tame current]

Chatgpt is fine for tasks. I prefer Opus for thinking, and being challenged.

[tame current]

Not at the moment, I think.

[inland bear]

Thanks again!

[tame current]

You're welcome!

[stuck topaz]

perplexity release enterprise plan for $40 with unlimited search, I havent tried it but I think opus is unlimited in this plan. So right now the only way to use unlimited opus for the pro plan is paying them extra $20

[inland bear]

I don't subscribe to the business plan, but, at least from what I've officially read on the Discord pages, the limitation also applies to it (unless something has changed in the last few days and I haven't seen it).

[tame current]

Opus is NOT unlimited in the business plan.

It is also 50.

[stuck topaz]

wow, even paying $40 is not even enough to cover the use of unlimited opus. yeah it makes sense, poe.com charge 350 points for gpt4 and 2000 points for opus, which is nearly 6 times higher.

[tame current]

1984 again

Bruh

[opaque stump]

Hope OpenAI actually announces their search engine today

Light a fire on Perplexity’s butt

[wise edge]

Multiple competitors can accelerate product iteration.

[tame current]

I think that's just a rumor tho, it's not confirmed if they'll launch it today

[urban goblet]

Compete and get better, can’t compete? Lose and fall behind.

Natural process

[torn sierra]

100% true but it sounded a bit like it was a way of hiding behind others. I guess it's completely different (better) when everything is said outright than when someone has to ask questions 🙂

[tame current]

still cheaper to use perplexity tho

[cinder comet]

What will happen when openai release searchgpt?

It will be released next month

[agile jay]

Isn't that supposed to be announced on monday?

Also it is likely to only support the openai models.

[cinder comet]

Yea but it will be slowly rolled to all users

[cinder comet]

Isnt that enough?

[lone citrus]

I am tired of OpenAI "news." They always get me excited and don't actually publish anything in the end.

[agile jay]

Nope

[cinder comet]

It is

[agile jay]

For you, manybe

[cinder comet]

Im only using opus in the writting mode

[mental finch]

is there any document or news about SearchGPT?

[agile jay]

One of the reasons why people use perplexity is because of the access to more models.

[agile jay]

Yep, they really like edging

[cinder comet]

Only leaks

[cinder comet]

Im using it because of opus

[agile jay]

I haven't used it much for a while

Since I have my own version

[lone citrus]

Altman:
GPT-4 will be the most idiotic AI model you have ever seen.
Yes, I do appreciate GPT-2.
AGI will change the world more than people thought.
We are not sure if it will be called GPT-5.
And more and more, but nothing happened. He has become an AI tech ceo version of Elon Musk, doing nothing real but sending spam posts on Twitter (should I say X?).

[agile jay]

Yep, but GPT4 is a 2 year old model, so they have something inside their closet. They just don't wanna release it...

And with gpt-2 that has been teased recently, they are probably gonna replace gpt3.5

[limpid warren]

But is the perplexity of opus really opus? I think most people don't think so, and from my own experience, it's not.

[lone citrus]

I agree, Im just really tired of these marketing techniques. I want to have an actually helpful product