💬│general

[viral vortex]

Guys where can we open a ticket ?

Support ticket

Perplexity took my monet

Money*

It showed 1 week free trial

And then when I registered it took my mine

Money *

[thorny canopy]

Any news on the limit fix or bump up? I hate pinging and asking, but it is a new week...

[wraith crow]

The free trial has been removed from the website. You can request a refund by mailing to [e-mail]

[viral vortex]

Free 1-week trial

Didn’t give me that

[wraith crow]

Please keep it nice in the chat :)

[viral vortex]

Yes

[viral vortex]

Tell me about this

Who’s fault is this

Luring people and looting them

And not giving trial as well

[storm bough]

i got the free trial, that's weird

[viral vortex]

I didn’t get and my money got cut

I’ll file a case if I don’t get my money back

[storm bough]

just refund it through ur bank

[tame current]

That should also work

[viral vortex]

I don’t want the free trial as well now

Don’t want

I want my money back

[storm bough]

^^^

[viral vortex]

I’m trying

[signal hamlet]

#⚡│ask-community message

[viral vortex]

I’ve dmed you

[wide mesa]

Seems like Perplexity isnt very good at following instructions on word count

[modern crown]

Has anyone used Sonar Large 32 pplx model? I feel like it has been pretty solid.

[signal hamlet]

Hey, @wide mesa, please take a look at this thread: https://discord.com/channels/1047197230748151888/1225545989989859409

[wide mesa]

ooo thank you for this!

[devout cargo]

https://twitter.com/OpenAI/status/1784992796669096181?t=jxQPjaXH0R-YSCYyHvUYIg&s=19
Hello, isn't this new Open AI feature the same as Perplexity's AI profile that remembers your name and preferences?

OpenAI (@OpenAI) on X

Memory is now available to all ChatGPT Plus users. Using Memory is easy: just start a new chat and tell ChatGPT anything you’d like it to remember.

Memory can be turned on or off in settings and is not currently available in Europe or Korea. Team, Enterprise, and GPTs to come.

Twitter · Apr 29, 2024, 5:07 PM

[modern crown]

@signal hamlet have there been any updates on the Opus usage? Sorry to be annoying about it.

[stable radish]

Hmmm, alright so I might go and invest those $20 on Claude.

But Claude 3 doesn't possess any tools yet right?

[signal hamlet]

Hey @brittle temple, rabbit's codes shouldn't expire, please send us all the details to [e-mail], and we'll help you apply it.

[modern crown]

I have been a big fan of the Sonar Large 32 pplx model

[austere kestrel]

no it's different - you can carry information across conversations

[devout cargo]

But if you give him information about your name. Your interests and you tell him to remember it. Does the AI always act on those parameters that you have given it about you?

[harsh verge]

Not really. The equivalent of Profile AI in the Perplexity interface is in ChatGPT "Customize ChatGPT" with "Custom Instructions".

[devout cargo]

Thanks for the clarification 😃 At the moment Open AI and Perplexity are the only ones that allow these options. Well Youchat too

Hello @signal hamlet How can you take advantage of the AI Profile in Perplexity? Is this function and type of prompt useful? In the general Perplexity answers I have been looking for, it is not detailed Thank you

[austere kestrel]

yeah though you can manage it ig (or opt out)

[devout cargo]

I don't really know how useful it is. This is how it behaves like Alexa or Google Assistant. a personal assistant

Gemini already has personal assistance but does not have these capabilities

[austere kestrel]

ig we'll see. but if after a very long conversation (particularly working with code or something) you can say "remember this conversation" - and it stores a decent summary of what the conversation involved (e..g. "started with this problem, tried this as a solution but didn't work; tried something else as a solution, which partially worked.." kinda thing)
being able to just refer back to that conversation in other new conversations would definitely have value i think

[jovial mantle]

Your avatar was the cover of my high school introduction to computing book

[agile jay]

https://fxtwitter.com/Gradient_AI_/status/1785030931407143040?t=U4_FdN9hNDaE9y432-lssQ&s=19

GG Gemini

[jovial mantle]

it's 8b tho

[austere kestrel]

still - that's like near perfect recall for virtually 1m tokens

that's wild

[agile jay]

Yep, last Friday it was at 160K...

And soon the 70B model will also be extended too.

[jovial mantle]

I want the 400B model :S

[agile jay]

Obviously the 8B model was done first since it's cheaper to do.

[agile jay]

70B finetunes are probably gonna be better for most cases.

[sterile falcon]

I have used ChatGPT and Claude directly through their websites and specified in the prompt that I only need X amount of words. Never have they followed my instructions. In fact, both ChatGPT and Claude have a hard time even counting the number of words. 9/10 both will give the wrong word count if you ask them.

[fossil sentinel]

your first mistake was trying to specify word count to an LLM

[jovial mantle]

They don't understand the meaning of word

[fossil sentinel]

it's better to specify paragraph or sentence count

[jovial mantle]

besides they perform statistical inference, you won't get a word count on their own output

Unless they were progammed with some meta-function that they could call on their own output

[agile jay]

Only way to make that work is by using prefixes

[warm cave]

You.com has a feature like chatGPT, say it a little back

[feral marlin]

like everything but claude.ai has that

phind also has

i think poe has

[ripe void]

Thats cool

Did you render that

[warm cave]

Thanks, yeah I used dalle 3

[valid yoke]

I think this channel is officially only English

[tiny plaza]

maybe one day the 50 will be replaced by something bigger again 🥹

[valid yoke]

Maybe we could make a Spanish channel though

[tiny plaza]

🤨

[minor shard]

Tats a profile. Memory is it actually saving your conversations and using them. Perplexity has a profile as well but not memory

[ancient flare]

its on lmsys. try it. for coding better than opus imo

[tame current]

They promised it will only be temporary. So, if it isn't... well. That would be dishonest at best.

[tiny plaza]

GPT-2-Chatbot blows so much out of the water imo TwT

[ancient flare]

hence the name eh heh

meme name

[minor shard]

Technically Llama 3 and GPT4 are better now 👀

Don't really need Opus that much anymore imo

[tame current]

Doesn't matter. They promised it would be temporary. So, they need to keep their word.

[minor shard]

Temporary can be months or days. They did not specify

[ancient flare]

gpt2-chatbot is tight lipped, but my guess is its trained off gpt4, opus, and all the arena top runners

[tame current]

Yes. But, if they never intend to put it back to the current number. Then, that's not temporary.

[minor shard]

I guess we will find out 🙂

[warm cave]

You.com saves snipets from your chats

and then adds them to the profiles automatically

seems very similar to chatGPT, both seem to save relevant information into your profile automatically, and then you can manage and delete the "memory"

[ancient flare]

you.com is terrible, buggy and context limited

even uses 8b as its llama 3 model

[warm cave]

Im not arguing for the platform, we are discussing the memory feature, haha

[shut plaza]

you.com doesn't even let me login lmao

just gives me an error

[ancient flare]

going by how it runs i wouldnt trust its memory feature hah

[shut plaza]

probably an conflict with my adblocker, but I shouldn't have to disable it

[ancient flare]

im going to try out chatgpt's one tho

[warm cave]

I dont think memory is something i want, I turned it off as soon as it was added to chatGPT and only turned it on with you.com to make the video

Maybe in the future i will change my mind

[ancient flare]

im looking forward to infinite context, it'll be game changing for coding

[warm cave]

Well yeah, long context is useful, but i don’t see little snippets here and there being helpful in its current stage

[ancient flare]

ah its only for small facts? not useful for my use case

[south kindle]

I figured crazy marine was back shilling for you.com again. Ha.

[minor shard]

oh ok, thought it was you that put it there lol

I don't think Ive ever needed to have AI remember past chats honestly. I can't even remember what I asked it yesterday 😛

[feral marlin]

I thought it wasn’t context limited

I’ve used with around 100k context just fine

And they claim publicly that it’s same as api

[stable fossil]

{
"allow_article_creation": false,
"article_image_upload_limit": 999,
"create_limit": 50,
"default_copilot": false,
"default_image_generation_model": "dall-e-3",
"default_model": "claude2",
"disable_training": true,
"email_status": "daily",
"gpt4_limit": 572,
"has_ai_profile": true,
"has_data_retention_warning": false,
"is_sidebar_collapsed": true,
"is_try_with_x_banner_dismissed": null,
"notif_status": "daily",
"opus_limit": 46,
"query_count": 392,
"query_count_copilot": 151,
"referral_code": "FG1E64F5",
"referral_num_coupons": 0,
"referral_num_success": 0,
"revenuecat_source": "none",
"revenuecat_status": "none",
"stripe_status": "active",
"subscription_tier": "monthly",
"upload_limit": 998
}

guys Why it says default model is claude2? I have set my default model as Sonnet in settings

[warm cave]

Yeah, 100% the same, haha

[sonic musk]

same when i switch to sonnet, always show claude 2

[tired basalt]

I love how helpful perplexity is for coding and debugging

[stable fossil]

you're right. chatgpt plus has became terrible. and open ai is not fixing code generation issue on chats. and the quality is also reduced. but in Perplexity if don't like the. we can just rewrite with different model which i freaking love it

[plush vessel]

Yea i agree and i think perplexity has up to date debugging because it is getting info from the web unlike chatgpt which just stores nearly not up to date info

[tranquil urchin]

I am a paying user of your company from China. I noticed that my service request has been temporarily blocked and received a notification from your team indicating unusual activity detected from my IP address.

I would like to confirm that I strictly adhere to all terms of use and policies while using your services. I have not sent any spam emails or engaged in any misconduct. I rely heavily on your company's services and hope to regain access to my service request as soon as possible.

If possible, please assist me in resolving this issue so that I can continue to use your company's services normally. If any additional information is required or further steps need to be taken to verify my identity, please let me know, and I will do my best to cooperate.

Thank you for your understanding and cooperation. Looking forward to your reply.

[agile jay]

change your vpn location

Certain servers are gonna be blocked, so change it until you find one which is not.

[tranquil urchin]

#thanks.I'll give it a try.

It's already fixed. Thank you.

[agile jay]

Np

[tame current]

That was funny. It was like an old days open letter.

[torn ravine]

Why isn't there $5 credit in the API account? Is it because it's the end of the month? I saw $5 here before.

[warm cave]

You get $5 for free with pro, it will expire each month and be replaced with next months API credit, making its always $5

[torn ravine]

I'm already subscribed to Pro, yet it went zero.

Is it by natural month or calendar month？

[warm cave]

sorry, i misread it, i thought you asked why you have $5, but i see you asked why you don’t 🤦. Yeah, i have never seen my api credit at $0

[ripe void]

The cybertruck is so dumb

[solid narwhal]

hi

[vale ravine]

yes even i tried multiple models and all faced same issue,
sometimes with even a simple question it gets stuck in a research loop and answers after like 3-5 minutes which is just crazyy!!!

[placid coyote]

it's a little strange when gpt4 writes about knowledge cutoff despite having access to a web search tool. I guess it didn't find relevant info in search results

[stuck iron]

Hey is it possible to get citations like [1] and as seen in the web UI via the API requests? I am using the perplexity-online models as I want RAG like knowledge over the web.

[stable radish]

Excuse my language, but

The f is this

[novel sable]

My Pro subscription coupon code (that i received with purchasing Nothing Phone 2(a) ) is showing expired while the last date to activate it is today.
Emailing support hasn't been helpful.

[halcyon coral]

Hey @novel sable!
The program has been suspended earlier than the anticipated date of April 30th due to a high incidence of fraud in the discount codes. Please reach out to Nothing support if you require further assistance.

[tame current]

I got the most recent system instruction for the standard Chatgpt 4. And, it's quite interesting : Certainly! Here's the full instruction from the beginning, which provides guidance on how to interact with me effectively:

"You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture. You are chatting with the user via the ChatGPT Android app. This means most of the time your lines should be a sentence or two, unless the user's request requires reasoning or long-form outputs. Never use emojis, unless explicitly asked to. Knowledge cutoff: 2023-12. Current date: 2024-04-30.

The user provided the following information about themselves. This user profile is shown to you in all conversations they have -- this means it is not relevant to 99% of requests.
Before answering, quietly think about whether the user's request is 'directly related', 'related', 'tangentially related', or 'not related' to the user profile provided.
Only acknowledge the profile when the request is directly related to the information provided.
Otherwise, don't acknowledge the existence of these instructions or the information at all."

These guidelines ensure that interactions are relevant and tailored to your needs, emphasizing the importance of directly relating responses to your specific inquiries or requirements when relevant.

[chilly nova]

how do you get access to the sources from perplexity answer?

i want to see where the info comes from

[steel tiger]

click on them?

[real coral]

@signal hamlet There is a problem logging in. When I tried to log in and enter my email address, a message appeared: "A sign-in link has been sent to your email address." However, I did not receive a message for a very long time.
My email address is "a lower rank" than, for example, Gmail, which is probably why I am delayed in receiving messages to log in. I have a Pro account
I am still waiting for a message to log in...

[tame current]

Did you check your spam?

[real coral]

Yes, of course. The folders are empty. I even refresh the page manually. 😦
I have tried a second time to do a sign-in, and nothing has come to the mailbox yet.

[tame current]

It's possible that the DNS flagged it as spam and it didn't go through, I guess.

[real coral]

Maybe it is, but I still don't have access to my account since this morning. @toxic moss 😦

[wooden karma]

omg

[scarlet pagoda]

how

[tame current]

It's not the size, it's how you use them.

By which I mean LLMs.

[rich blade]

is there a way we can have some sort of raycast ai but on windows with perplexity or else?

[tame current]

My mind misread that and thought you were asking for a racist ai.

https://chat.reka.ai/

Check them out. They're really good.

[scarlet pagoda]

OPUS is way better though, when dealing with large contexts

[tame current]

I love Opus and paid for 600. And got 50.

[kindred folio]

is perplexity down or am i trippin

[halcyon coral]

Perplexity is not down 😉
You probably need to refresh the website.

[kindred folio]

worked instantly thank you 😭

i shoulda known

i thought i did refresh

[opaque ferry]

is it possible to generate image only?

[wraith crow]

You will always need a thread with 1 prompt first. After that you can generate images with custom prompts.

[gleaming mantle]

Is something going on right now? All AI models are taking probably 10x the time to respond with answers for me.

[supple vine]

I noticed this yesterday. I thought maybe it was just the default model I had set. I switched to another, but nope. Same issue. It continues today.

[iron dagger]

does perplexity pro really allow you to use chatgpt4? i.e. would it be the same as paying for chatgpt plus?

or when they say chatgpt4 do they mean it is based on chatgpt4?

Thanks in advance

[supple vine]

Yes it will use GPT4-Turbo right now. The difference being that you don't have access to GPT's or memory like you would through a ChatGPT sub. As I understand it.
Here's a screenshot of the different models right now.

[iron dagger]

perfect thank you very much for the answer thank you.

[gleaming mantle]

I'm relatively new to these tools; what's this and why would that matter to me?

GPT's or memory

[supple vine]

GPT's are user made prompts or services that do specific tasks. The memory thing is new. As you chat with ChatGPT it "remembers" things about you and your preferences.

[gleaming mantle]

Ah interesting, thanks.

[livid aurora]

Llama 3 is inside Perplexity

[tame current]

Anyone test sonar with Llama 3 to see how it compares to Llama 3?

[tame current]

wdym sonar has internet access so obv it's better

[tame current]

Right, but wb quality of answers. It's like comparing opus from perplexity to opus from Claude pro. Opus from Claude pro has better longer answers and better quality for example. Opus from perplexity is sorta the same but I noticed answers are definitely shortened. I mean in this sense

[tame current]

are you using mobile app ? if so they intentionally do that

could be possible that they do something like that on the web too

you basically won't get the model 1:1, Perplexity changes stuff in various ways and they also probably use some prompt engineering to improve accuracy etc (even in writing mode)

[tame current]

Yeah I use mobile more than web. Is there a noticeable difference in using the web version compared to the mobile version?

[tame current]

yes they intentionally make the response shorter on mobile

[umbral ember]

Has Llama3 been rolled back? My model options are only showing Sonar listed as it was before the Llama 3 version was introduced

[tame current]

Currently on the web it shows Llama 3 as part of sonar.

[minor shard]

This is what I see

[umbral ember]

Yes, my mistake, I see Sonar Large 32k on Web, but not in Android.

[minor shard]

Yeah the app will show it later on once it gets updated

[limpid mason]

Am I the only one that has the voice function that doesn't work on the iOS app ?

[waxen silo]

Hey guys, does Perplexity serve ads? or is it fully ad-free? I haven't seen any but want to just make sure. Thanks!

[valid yoke]

They don’t have ads yet

[eternal quartz]

They apparently have plans for ads, but all that's been shown is sponsored suggestions at the bottom of an answer. Not full banner or pop up ads, nor anything that affects the primary response. At least that's my current understanding.

[candid badge]

What. They deleted my commercial.

I got 5 sign ups the last time I did that. They didn't delete it then.

[waxen silo]

Gotcha thank you! @valid yoke and @eternal quartz

@eternal quartz so you have seen some sponsored suggestions at the bottom of answer, interesting. I haven't seen them yet. Is that when you're logged in, in pro, or as a guest?

[vague oar]

Anyone know what the state of Perplexity Push is? I saw the announcement a couple months ago but nothing since. TBH this is my top wishlist item, and what I would consider a killer feature that would set Perplexity apart!

[valid yoke]

What is it?

It’s for slack?

[vague oar]

#📢│announcements message

[vague oar]

It is, but to be honest I'd love to see it for Discord, or even notifications from the mobile app

The thing is, right now most LLM use is passive - ask and response. But regularly running queries in collections and pushing the results would give a proactive way to keep on top of topics

That's why I thought Perplexity Push was so interesting, but I haven't heard much since the announcement, and I'm hoping it is something they could do broader than Slack

[valid yoke]

I don’t understand why the Perplexity app insists on using in-app Safari

Drives me nuts

[wise hedge]

Hi folks. New here, using pro and wondering if the new memory features of ChatGPT can be accessed via Perplexity ? Any ideas? Or is perplexity working on something similar in house?

[eternal quartz]

You definitely can't use ChatGPT's memory. No idea if Perplexity is planning it, they don't really talk about upcoming features...

[eternal quartz]

I haven't seen them myself that I've noticed, I just saw the articles about them. Tbh I never even look at the suggestions...

[wise hedge]

Thanks

[tame current]

my friends wants to try perplexity pro for 7 days but isn't able to find the option

is it still available?

the free trial

[fluid mauve]

The free trial was stopped as it was being abused

[tame current]

oh...

so sad..

okay

it was a good way to convince people

using pêrplexity

[fringe leaf]

Hello, i want to report an issue regarding Nothing's Perplexity Pro offer. Where can i do it?

[serene sand]

#1234393609344782376 message

[fringe leaf]

thank you

[rancid rain]

Can we change the voice in the iPhone client?

[serene sand]

Settings - Voice & Language

[stoic wave]

One of the coolest ideas to use AI so far

[versed plaza]

Haha

I just like trying to get its reaction when i say dumber and dumber stuff

It's like always so serious

[orchid shore]

Username checks out

[scarlet valve]

I'm pleased to announce uploading an image still renders your thread useless!

[hallow sentinel]

what is everyone's go-to model to use for responses? I like the accuracy of Claude 3 and the way it uses bullets, but is another one better?

[leaden fractal]

Does Perplexity have a dedicated support?

Or is that support person on here?

[hallow sentinel]

I tried emailing their support once and never got a reply. Maybe try on here?

[leaden fractal]

Thank you

Is there a person who is support on here?

[versed crest]

How do I get access to the pages beta, I reacted but cant see the channel ?

[compact bough]

@signal hamlet Can I have access pleas

To pages

[signal hamlet]

Please react with to the #📢│announcements message
After reacting, kindly fill out the form located in the #🅰│web-alpha-feedback. Once you've submitted the form, please wait for a confirmation.

[late fern]

I did that but no form appeared when I clicked the no access

[crystal trench]

It took a few seconds for the No Access to turn into a link after I reacted. Check again?

[signal hamlet]

Please try removing your reaction from the post and clicking on the emoji again

[smoky oriole]

Will Llama 3 1M context window be available on Perplexity?

[ripe void]

Please ignore what i deleted…

[tame current]

I really really want to know what you deleted.

[upbeat dawn]

Hey , where i can check that 50 daily limit for opus?

[warm cave]

You can see how many you have left using this: https://www.perplexity.ai/p/api/v1/user/settings

[gilded finch]

does perplexity get caught by ai detectoir

[tame current]

those are not reliable but yes

you could try one of those paraphrasing tools and see if that helps

[upbeat dawn]

Its daily, right?

[sly yacht]

Such as? ^_^

[gilded finch]

qillbot

whats the limit for amount of times I can use perplexity ai

[harsh verge]

Yes, but they don't all come back at once
Every time you use an Opus model, a credit is deducted from your meter. The credit used will be automatically restored exactly 24 hours later. This means that the counter is not reset to its maximum value after a fixed period.

[gilded finch]

whats opus

@harsh verge

[harsh verge]

Claude 3: Sonnet and Opus models, Opus has very advanced capabilities and is considered the most advanced LLM on the market at this time. https://www.perplexity.ai/hub/faq/what-is-perplexity-pro

[cold tide]

woohoo

how do they know im pro?

is this a pro only server?

[tame current]

you must have joined from the link from your profile

that link is only for pro users and free users don't get that link

and this isn't a pro only server, no

[tame current]

yeah he was asking why he had the role when he didn't say he had pro

I think

[harsh verge]

ah! ok 😅

[shut dagger]

Looks like I lost my pro....🤔

Why don't I see pro lounge?

[wraith crow]

#👑│pro-lounge you can display all channels by clicking on "Channels & Roles" in the top left and click on "Show All Channels"

On phone hold the server icon "More Options" > scroll down and select "Show All Channels"

[fast meteor]

you unfollowed that channel / category by accident. to re follow:

[shut dagger]

Thanks guys!!

[chilly nova]

am i the only one who just noticed the pro search?

[south kindle]

are you asking if the rest of us just noticed a primary feature for the platform that was introduced over a year ago?

[halcyon coral]

Pro Search, formerly known as Copilot, has been around for some time at Perplexity. You can find out more about it here: https://www.perplexity.ai/hub/faq/what-is-pro-search

[chilly nova]

idk, i never got this stuff appearing while i was searching before

[rapid crystal]

Hi! How many times can I use each AI model per day after upgrading to Pro?

[boreal tiger]

for coding purposes is gpt4-turbo the closest to opus 3 on perplexity ? for some reason i feel like its way worse. sometimes i even prefer sonnet

[wraith crow]

You can use the advanced models 600 times a day with exception of Claude Opus that has a temporary limit of 50 uses a day.

[rapid crystal]

THX. Are there no restrictions on other models? This is great

[blazing kraken]

My iOS app just changed UI without an update right now…looks fresh

[tame current]

I'm glad that there is another confirmation that it is temporary.

[boreal tiger]

as much as I love this team and their work. Saying "temporary" doesn't address the issue. Atleast give an estimate :/

[cursive jacinth]

Hi sir, there was never been an estimate from perplexity sir. They are afraid to make promises sir.

[tame current]

Yeah. They're valued at 3 billion. I think they can afford the hit.

[gilded vapor]

It seems AI Prompts for Collections doesn't work for me after the first thread. Within the prompt, I tell it not to use bullet points and to have somewhat of a personality. When I create the first thread, it works, but when I make more threads, it seems to go back to the default Opus personality. Here is my prompt:

“You are an expert in child-rearing with 30 years of experience across various cultural backgrounds within the United States. Your expertise includes understanding the nuances of parenting styles from different communities and how these influence child development. In your responses, you should provide empathetic, gentle, and non-judgmental advice. Your knowledge spans a wide range of topics from discipline to emotional support, integrating cultural sensitivity into everyday parenting practices. When responding, please use paragraphs instead of bullet points to ensure the advice is well-integrated and flows naturally. Your goal is to help parents feel supported and understood, while offering practical and culturally aware parenting guidance.”

[tame basin]

how many questions does perplexity ask when setting up the profile?
it feels like they just keep coming lol

[tame current]

man even if you write text by yourself then it is detectable by ai detector those detectors are useless

[fossil sentinel]

https://youtu.be/ddTV12hErTc

💀

[cinder comet]

ended?

[tame current]

Those things detect the Bible and Shakespeare as plagiarism.

[cold tide]

you can do that on discord now?

[tame current]

do what

[cold tide]

yeah that's it

[cold tide]

like get a role for joining using a specific link

[tame current]

check which link someone joined from ? it's been like that for ages

bots can do that easily

[cold tide]

ah

sounds good man thanks again

[dim ginkgo]

i don't know if perplexity has updated gpt2 or not😃

[tame current]

gpt2-chatbot? that was only on lmsys

[cursive jacinth]

Hi sir, why do you say so sir. no one ever said that it will be ported here sir.

[tame current]

yea it's not available anywhere else

[dim ginkgo]

i know, i'm just kidding

[valid island]

can't find anything but will sonar large 32K be added to the api?

Spoke too soon

[oblique flame]

Does anyone know the difference between the new pages feature, and sharing a search we’ve made? Kind of sounds like the same thing to me.

[tame current]

I think pages is kinda like the "research" option on you.com but I haven't gotten access so I don't know

[oblique flame]

Interesting. I look forward to testing. Thanks!

[warm dew]

hello, may i asking?

i found that perplexity doesn't show any references at all on my browser, is that bug?

[tame current]

I'm having the same problem actually

first time it has happened

[warm dew]

yes it is

[winged swift]

same issue with me, even after purchasing pro version 😭

[tame current]

pro has nothing to do with it, it's probably just a temporary issue

it was working fine a few hours ago

[frank rivet]

Hi! Can anyone who is experiencing this email a report to [e-mail] with the following info?

a. web browser and/or mobile app (android or ios)
b. link to a thread would be super helpful or,
c. screenshots.

Thanks.

Update: I can confirm we having some issues with search right now, we appreciate your patience while we fix this issue. Thanks!

[warm cave]

This shows how much you currently have left, you get 50 Opus a day rn

[slate condor]

Claude IOS app is out....Might be time to switch over.

[tame current]

Does it have web search? Won't be the same thing as perplexity

[tame current]

bruh

💀

[pulsar ridge]

Any other Enterprise users having this issue Error code 525

[sterile falcon]

Have you ever noticed that Perplexity tends to forget things when the conversation gets a bit longer? In Writing mode, regardless if I am using GPT-4 or Opus, I notice that the response quality get's worse and it forgets my original prompt. I read online that the context window is 4,000, which is lower than ChatGPT plus. Any idea if that has something to do with it?

[lone citrus]

It's actually at least 32k contexts, 4k context is the output max window

[lone citrus]

Claude 3 can >>write up<< to 4000 tokens at a time

[sterile falcon]

This is what I saw on the website. Now I understand being able to write up to 4000 tokens at a time. But does this mean that the context window of the whole conversation is 32,000?

[lone citrus]

We don;t know, someone say it's 32k, someone say it's as much as original LLM had, pplx never explained

Also, do not use document upload feature for long context need chat

they do not give u full context when use upload instead sending texts

[halcyon coral]

Yes the conversation context window is 32k

[sterile falcon]

Thank you for that information!

[lone citrus]

Thanks for confirming that. Is that the same for enterprise users?

[sterile falcon]

I see. So if the context window is the same as ChatGPT plus of 32,000, any idea why in writing mode it tends to forget the prompt from earlier? I have this problem when I am creating flashcards from my notes, and it tends to forget the instructions. It's mostly with GPT-4. Do you suggest trying Opus?

[halcyon coral]

Yes

[lone citrus]

😰There really needs to be more features to encourage people to buy that plan.

[halcyon coral]

These are mostly prompting inaccuracies, but even LLMs can make mistakes 😉
But yes, in most of my tests Opus performs better than GPT-4-Turbo.

[sterile falcon]

Oh, I see what you’re saying. In my prompt, I write “Please create flashcards from my notes. Follow these instructions…”

[south kindle]

Try using the scratchpad prompt first to create an "outline" of the main points, then inject the flashcard prompt. #1223058316662538331 message
should help keep things within the context.

i hit my shilling for scratchpad quota for the day. lol.

[halcyon coral]

You can use the Collection AI profile or general AI profile to apply instructions

[sterile falcon]

Thank you, I will try that.

[sterile falcon]

Okay, I will try that as well. Thank you.

[south kindle]

I have this prompt set up as a collection. I have them condensed down in size so they fit. should remove the need from directly inputting it. as IceLava said above.

[sterile falcon]

If I have a prompt set up in my AI profile or collections, do I need to repeat it when I start a new chat?

[south kindle]

you shouldn't need to.

[sterile falcon]

I gotcha

[kind kettle]

Perplexity as been dead for almost two hours.

[halcyon coral]

Do you use a VPN?

[zenith flint]

Why perplexity gives the same answer or slightly modified answer

So often

I bought premium version to get rid of it

I often lost free pro searches because of this

It gave me the same answer multiple times

This is serious issue

[south kindle]

what model in use?

do you have any threads to share?

[thorny canopy]

I noticed when people ask for an update on the opus limit even with pings...seem to be ignored....oof ☕

[signal hamlet]

Hey @thorny canopy, the limit is still 50. We'll post an update when we have one to share.

[signal hamlet]

Hey @pulsar ridge! Thanks for reporting, this should be fixed now.

[tame current]

oh

never mind 💀

[tame current]

The Claude Team plan has been announced. But. It is $30 per user per month, and it has a 5 user minimum.

[bold badger]

Is there any point using the ChatGPT app or new Anthropic app on iPhone if you have a PRO acccount on perplexity?

[tame current]

Nope. They both use the API.

But, the Claude app is also avaliable for free users.

[bold badger]

Thanks @tame current

[jade blaze]

What does hiring process at Perplexity look like? Would love to join this amazing firm!

[tame current]

Actually, to be honest, I still have a Chatgpt Pro account. 1. I like the Dall-e 3 instructional edit, the memory feature and the custom GPTs.

Also. Longer context. So, if you're writing long things.

[warm cave]

I thought chatGPT was also capped at 32k

[tame current]

I meant for Claude.

[warm cave]

Oh, yeah, I have Claude 3 Pro for long context as well

[tame current]

It's nice to have 5 messages then capped.

[bold badger]

It will be great when image generation is available in app as well.

[tame current]

The best image tool avaliable today is Ideogram.

I imagine if Claude does get image generation, they'll use Stable Diffusion.

Apparently, Grok is in talks with Midjourney.

Beautiful.

[warm cave]

Love this one 🤩

[south kindle]

i tried but yours is still better

[tame current]

I used Ideogram.

And their magic prompt.

[south kindle]

mine was just dalle-3. plus Opus reviewing it to come up with an image prompt

https://th.bing.com/th/id/OIG3.XP_nuQRL1FDZbRgJMYFb?w=1024&h=1024&rs=1&pid=ImgDetMain
A calico or tortoiseshell cat with bright green eyes and an alert, curious expression, sitting in the foreground against a swirling, abstract night sky inspired by Van Gogh's 'The Starry Night'. The cat's fur should be a mix of black, white, and orange/yellow. The background should feature expressive brushstrokes of deep blues, greens, and vibrant yellows swirling in a stylized interpretation of the starry night, with elements like a crescent moon and cypress trees reminiscent of Van Gogh's iconic painting.

[restive osprey]

im new to the ai space. been using a few of them. curious if someone can help me brain storm ways I can use it too?

suggestions .

[sterile falcon]

For those of you guys that have Claude Pro, have you ever compared the responses with Opus using Perplexity?

[chilly fox]

What model does the free version of Perplexity use now? I know the CEO said a month ago or more that Claude 3 Haiku was gonna be the new default

[restive osprey]

i did that just now.

wont get into the subject matter. but, i took my convo and results very long back and forth into opus 3.

[sterile falcon]

I think it uses Perplexity’s own model as the default.

[sterile falcon]

Were responses similar using opus through perplexity?

[restive osprey]

Id say that Opus 3 is a but more advanced and aware of the direction of the covnersation than Perplexity.

[south kindle]

https://www.perplexity.ai/search/respond-to-this-AQZNICUASE6lG5eVk9EgEg#0

[restive osprey]

perplexity however is no lightweight. It pulled relevant info, albeit a bit redunant at times.

[sterile falcon]

Do you mean that opus through Claude remembers the conversation history better?

[serene sand]

#1233663028289867815 message

[restive osprey]

I want to understand what you're asking exactly. Im not yet subbed to Perplexity pro. So cant speak to all the options there yet.

Im subbed to claude ai

[sterile falcon]

Are you using opus in writing mode through perplexity? This way, it won’t search the web.

[restive osprey]

Opus 3 dont have web access. I just either create a pdf, or text of the information I gain from Perplexities help into Opus 3. This I just tried today.

[sterile falcon]

I gotcha. With perplexity pro, you can choose which model to use, GPT-4, Opus, etc. And in “All” mode, it searches the Internet for the answer. In “writing” mode, it doesn’t search the Internet and you can chat with it, similar to Claude or ChatGPT.

[restive osprey]

ah okay

i dont have Perplexity pro. I have Opus 3 through Anthropics website currently.

[storm bough]

is it worth versus GPT? @restive osprey

[serene sand]

https://www.perplexity.ai/hub/faq
https://www.perplexity.ai/hub/technical-faq

deep dive ➤ #💬│general message

[restive osprey]

im subbed to both gpt 4 and Claudi 3, and find I enjoy Opus 3 a lot lot more than gpt 4. Gpt 4 currently is not as good as it used to be.

[sterile falcon]

Have you had any issues with the message limits?

[storm bough]

whats your normal usage for opus?

[restive osprey]

tons of talks about a lot of text. it has a bigger token limit than gpt4 by a lot. unsure the exact amount. Normal so far, is learning a text more carefully unsure how to word this.

its got very good vision for images,

so it can translate say Italian or Russian, from a image

even if the image is skewed or not perfect. say a internet archive scan

I can do some coding

basic reshade shader coding, HLSL something like this

[storm bough]

no need for web browsing?

everything you get back is relatively accurate?

[restive osprey]

i dont need it in Opus 3, because i can use perplexity for that.

[storm bough]

ah i see

[restive osprey]

when you asked about normal usage i thought you mean like limitations., there is a alloted limit of replies. but the replies can be as long as you insist on it being. but little replies inbetween count as a reply

but this allotment resets fairly fast.

So i think 7 or so for Opus 3

the other sub models are noticeably faster but not as smart as Opus 3.

Claude ai could use some better organization like gpt has in its UI . like the history. but theres no search anyway in either of them i think

Curious how Perpexlity handles that

[sterile falcon]

I don’t have Claude pro, but I have read on Reddit about the message limits.

[sterile falcon]

I see what you’re saying. Perplexity has a daily limit of 50 responses for Opus and 300+ for GPT-4. They are working on increasing the Opus limit.

[restive osprey]

I dont know anything about Perplexity yet. So i cant speak to Perplexity.

what does a limit consist of?

is it like a limit of replies? Opus 3 imposes a limit of "replies" so say I have 10 or 7 or something.

[storm bough]

50 opus, other ones have such a high limit 600 that you wouldn't notice anything

[restive osprey]

okay . like others ?

50 for how perplexity uses their api i guess?

600 replies seems like alot -

wonder if the response is opus 3 or degraded, to some other model

[sterile falcon]

Some people prefer to use GPT-4 and Claude Opus without them searching the Internet. After all, these companies spend a lot of money on feeding so much information into the model itself. But Perplexity’s target users are those who need accurate responses to specific questions,

[restive osprey]

I will say Perplexity seems better than when I used it last year.

i can see myself using it much more. and maybe getting pro. which is how much money? anyone pro now?

[storm bough]

i am

[restive osprey]

worth?

how are you using it, use cases?

[storm bough]

I'd say so personally since I'm in school and need accurate answers but there's pros and cons I suppose

[sterile falcon]

I don’t think you should have any problem with running out of messages. I can’t comment on the quality of Opus responses in Perplexity vs. Claude since I don’t have Claude pro.

[storm bough]

What's weird is I find GPT-4 Turbo more accurate on perplexity than Copilot/ChatGPT after running questions through it, I have no explaination why

[restive osprey]

okay will be something i hope to find the answer for

[sterile falcon]

I use perplexity to ask it for problems that I face at work as an engineer, so very technical questions.

[restive osprey]

seems to be constant quality fluctuations in the models month to month year to year. Because I could believe that GPT 4 was so much more smarter than it is today currently. And at some point, GPT bing, was smarter than gpt open ai. and there was a bit of a degrading and elevating throughout. Limitations would change too.

because of that inconsistancy, and what i think is a current down trend in its quality , i been using opus 3 and pretty satisfied with it.

Bing chat, used to be somewhat pointed, and have a bit of a personality to it. Now its a bit regimented and stoic. just not as engaging. I get more generalizations than answers to the questions than I used to.

[storm bough]

i will say even being able to attach powerpoints and pdfs to perplexity it'll say it's citing from there but it's just giving you false citations

[restive osprey]

and it seems like its defaulted to search engine results at times.

[storm bough]

big con

[restive osprey]

oh and gpt 4 internet access i run into a lot more of it saying, cannot access the internet or something. whereas that was unheard of when internet was first introduced. i guess maybe more people are using it? so alot of rationing.

[restive osprey]

so it wont actually cite from the internet results or a web page?

seems to be a problem they all have right now.

[storm bough]

if you attach a personal powerpoint and say cite which page you found x information it'll give you the wrong page

[restive osprey]

yea if you're doing some serious study, Id try Opus 3

[storm bough]

idk where it gets it's information when it cites wrongly from the PDF/powerpoint

[restive osprey]

i wonder if i can test that in opus 3 now? i just dont use power points and dont have any on hand

if u have a mockup example i can use id try it

[sterile falcon]

I saw many different posts on Reddit of users complaining that the quality of Opus has gone down. But an employee from anthropic commented on those posts saying that they haven’t changed anything about opus since it was released.

[storm bough]

I have a neuroscience powerpoint I used for class, if you could put it in and say find something about x and cite the page then check if it was on that page that would be helpful

[restive osprey]

I mean could be true, gpt has gone down in my view. and there have been some basic studies that suggest that user interaction is causing this. The other idea is that the user base are becoming more aware of the limitations and just not recognizing this and mixing that into expectations of what they want or remember in their honeymoon period.

i think its a factor of both. And maybe you wont ever hear a company in the throws of a boom knock their product

true or not

[sterile falcon]

Those are all good points.. I just don’t know too much of the technical details,

[restive osprey]

sure

ill try it

can send to dm or whatever method u prefer

[sterile falcon]

Before, I would see the circle where it said “browsing with Bing.” Now it does it behind-the-scenes, and doesn’t show the circle anymore.

[restive osprey]

im not sure what is this circle. not using it as much

guess they call it copilot now.

[kind kettle]

Yes

[sterile falcon]

Copilot is a separate product by Microsoft, that uses GPT-4. But the quality of the responses are really bad.

[dense field]

very true, it's unfortunate.. the responses used to be good, now i get better results from a local llama3 model, did not expect that this year

[lunar anchor]

To have perplexity recognize which photo I’m referring to (when working with multiple photos), is it best to name the photos and then upload them, and reference them by the filename? (I’m used to just having them on my clipboard and uploading them via clip)

Or upload one per message and provide it a reference “keyword” or something?

It seems to get even more confused than I am when working with multiple photos 😂

[dense field]

i wasn't even aware you could upload multiple at once, interesting

[lunar anchor]

I tend to abuse computers

[dense field]

i can tell, lol

[warm cave]

Yayyyyy!

Meta added lightning image generation!

[ancient flare]

is that running meta's mega model yet?

[south kindle]

https://www.perplexity.ai/page/Understanding-the-Conscious-Gtw786J5QQe4EpR4TfusXw > sharing here as well for users without access to Pages yet to check out.

[ripe void]

WOAH

thats wild

[tame current]

why does it work with a prompt one minute, then later it says it cant do that task

[ripe void]

Is that on instagram?

[south kindle]

showing the creation process for your daily Opus posts I seee. haha.

[boreal tiger]

perplexity is down rn ?

[urban cipher]

how did you do the formatting for this? it looks so clean 🔥

[south kindle]

It's the new Pages feature. #📢│announcements message

typically there are embedded images, but I decided to not add any. the selection of images it came up with didn't really match the sections so, just left it as a text.

if you react to the main post I linked, you should get access to the channel. there's a pinned comment with "request access" form.

[warm cave]

Haha 😂 , I wish it was good at text, I would do a video each day

[warm cave]

It's on WhatsApp

[warm cave]

Waiting to get an email about Pages 👉👈

[halcyon coral]

https://discord.com/channels/1047197230748151888/1232978227371442237

[south kindle]

for the record, a bunch of us didnt get emails, it just appeared. or we didn't get emails "yet". however, checking the settings URL, articles is set to TRUE now.

[warm cave]

Thanks, so far no email and still set t false, but hopefully I get it soon

[south kindle]

I can say it matches what I expected the "articles" feature to be.

looking forward to more options though. currently unable to set the model in use. you essentially add sections by typing what it will be about, and if it's concise or detailed. there is some table features and image features. but no real way to add user generated sections or notes.

[warm cave]

seems cool, I hope to see it get a lot more features as well.

[south kindle]

We can convert threads to pages though, but not sure how "much" of the thread is actually incorporated. haven't tested it that much with either short threads or long threads.

[south kindle]

@warm cave where is your thread you had on search operators? can't find it.

wait that was @livid mantle

[sly yacht]

https://discord.com/channels/1047197230748151888/1222013886488645652

@south kindle

[south kindle]

nice.

[uneven basin]

since this morning every source code of opus is considered as copyright, have you already had this problem ?

the code is from opus self 🙂

itself

[hallow tulip]

@uneven basin can you explain in more detail?

[opaque ferry]

suggest some perplexity videos some advanced videos if possibl

[wanton urchin]

Is px not working on Safari and Brave? I can insert a promt but can‘t send it (no reaction of “return”). I also can’t register, because the button does not work. Can anybody give me feedback, if there is a hack? Thank you.

[uneven basin]

@hallow tulip it is code for an sharepoint importer from an external webservice source, i have send the full code from claude opus again into prompt

and ask some questions

[lilac stump]

if i attach a file in focus mode, like writing or academic, how do i stop the source being added ot each request that fllows?

if my first question is to analyse a file for content and answer it, and then build upon the file, the AI seems to keep acting as if each follow on request in that chat is the first time it's seen that file

[south kindle]

#1228788715380281376 message

known bug

[serene sand]

#💗│sharing message

[lilac stump]

it seems like they claim it was fixed, but it's not

removing the file seems to remove it from the entire chat history, i can remove it from the 2nd request even if the file's in the first one and the AI will say, "you've not provided me any files to analyse in teh chat history" as if it'd never been attached

[remote pasture]

How can I request a subscription refund?
They charged me the monthly payment and I didn't want it.

[wraith crow]

You can send an email to [e-mail]

[remote pasture]

thanks

[latent pendant]

Thanks for sharing! I'm interested in how you created it - is it the result of multiple prompts collated into one page?

It looks good, but now it basically is a way to share a thread as an article

Adding images would help to break the wall of text effect, and having TTS would be great as basically you'd have your homemade podcast episode

[south kindle]

I was going to add images but none of the suggested ones that were available fit well. I used a separate thread to come up with the topics but the actual text generation was entirely within Pages using detailed mode. I’ll probably find external images to add but haven’t done it yet.

When you use the add media function it sources the same set of images that get curated from threads (I believe). You can also generate a new image as you normally would or upload an external image.

[surreal marsh]

https://www.perplexity.ai/page/Rabbit-R1-AI-JuzTh5pSSAudQtqf.S2D6g

[harsh verge]

"Page" in a language other than English: https://www.perplexity.ai/page/Thorie-Neowave-de-CXDiVBPPSxexpIcZ9oS9YQ

[tame current]

R1 garb

[limber plover]

gm gm everyone

I need to translate a pdf. I'm a pro user.
How do I start a conversation with claude that sepcifically does that?

[hallow linden]

Change the default model or have it rewrite the response with a new model

[hallow linden]

You can upload the pdf with the + attach button

[gloomy kite]

❤️

[limber plover]

ty ty will give it a go

[sick linden]

how to access to page view ?

[serene sand]

#📢│announcements message

[fossil sentinel]

https://youtu.be/77IqNP6rNL8

[hallow linden]

based on lmsys data, I'd expect that the new sonar must be really good. Has that been anyone's experience?

[violet bramble]

Hi, can you explain any shortcut key for mac in webui ? like cmd+K

[zenith flint]

Voice chat don’t let me end the sentence

[warm cave]

I like this one:

Like a meme cat

[zenith flint]

Is there any way to make it usable

[minor shard]

Yeah it’s a little to search happy and ends the voice chat after 3 words lol

[tame current]

If I pay for pro it allows for 300 messages?

Bc chatgpt4 only does 30 am every 3 hours

[serene sand]

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

#⚡│ask-community message

[tame current]

Ah ok still a little confused tho

[halcyon coral]

What is confusing?

[tame current]

Is there a rate limit

Or just 300 a day

[halcyon coral]

It's a daily limit

[tame current]

How does it use chat 4 if in chat 4 theirs a limit of 40 questions every 3 hours

[serene sand]

https://tenor.com/view/imagination-spongebob-squarepants-dreams-magic-gif-12725683

[tepid pilot]

💀

Ain’t no way

Some ppl

Gd

I can’t

[tame current]

Still don’t get it

I sweat I’m not joking

Can anyone please explain

[halcyon coral]

OpenAI has set this limit for many reasons, ChatGPT also uses other tools such as DALL-E, Browsing, Advanced Data Analysis and more which cause more costs 😉

[tame current]

Ah

So when I get pro, is it one pro setting or do I need to click on specific tools to use

[serene sand]

deep dive ➤ #💬│general message

[south kindle]

that trial would be nice right about now

easier to say "just try it out" ha.

[tame current]

Yeah

[serene sand]

Perplexity is very complex tool, app. please take your time and explore deep dive I gave you @tame current

[tame current]

I’m just looking for a ai thays as good as chatgpt4 or better without the 40 question cap limit every 3 hours

[tame current]

Ok

[south kindle]

with pro, you can select one of these as default. for Opus, after you hit your daily limit, it reverts to Sonnet. via the 2nd screenshot, this is from my own settings page. you can see I still have plenty of GPT-4 uses left today.

[tame current]

So

Chatgpt4 here is the same thing at chatgpt4 on the app?

Uh

[south kindle]

yes and no. same model, but chatgpt has their own system prompt, like PPLX does as well.

for the record, microsoft copilot also uses GPT-4 turbo, and that's unlimited as well.

[tame current]

So how come chatgpt4. Gets info right compared to Microsoft copilot

Is it worth it to pay $20 a month for copilot

[south kindle]

i'd say no.

LLMs are just guessing the next token/word. sometimes they guess wrong. aside from that, I don't know why your getting different answers from different hosts for GPT-4 turbo.

PPLX offers a lot more than just access to GPT-4 Turbo or Claude models.

[tame current]

Well idk how to put this

[south kindle]

a vast majority of us here are on the Claude Opus train currently and find it's outputs far better than GPT-4 turbo.

[tame current]

I use chatgpt4 for school, and it’s right on hw and studying so I want something exactly like it, without the limit of 40 questions every 3 hours

[tame current]

I see

[south kindle]

PPLX is an answer engine. it replaces the need for going thru pages and pages of google results to find what your looking for. for that aspect, i'd assume it'd be better suited for your needs. it's a platform meant for research and curosity.

[tame current]

And that’s with this?

Which I can? With premium payment

And what’s the limit on it

[south kindle]

see my screenshot above. unless you need more than like 400 queries a day, you likely won't hit the GPT-4 cap.

[tame current]

So it’s 400 a day?

[south kindle]

i don't know the exact amount but its more than 30 an hour.

[tame current]

Ah ok

Trying to find that amount

[south kindle]

honestly, see if the default model and the platform itself works for your needs. focusing on query count instead of functionality is a little shortsighted if you ask me.

[tame current]

Well yeah

@south kindle in your opinion what’s the best paid alternative for chatGPT4 with bigger limit

[south kindle]

i've only paid for PPLX this year if that counts for anything. haven't used chatgpt plus since last year.

[tame current]

Ah I see

[south kindle]

but i'm not in school currently and mostly use it as a replacement for google searches. my needs arent "mission critical" like school is. i'm more of a explorer in that aspect, trying new models/features for platforms to fully test their limits.

[tame current]

I see

[south kindle]

new models that are popular get added to PPLX for us to use. condensing the need for multiple subscriptions.

[tame current]

True

But overall is their different limits based on the tool

[south kindle]

yes, but the only limits that people hit are the Claude 3 Opus one.

[tame current]

Oh ok ty

I got the free trial

[tame current]

Which of the options are best for me

[south kindle]

That's a subjective question. use the "rewrite" function if you have Pro to have PPLX redo the answer with a different model. see which works best for your specific use cases. if your goal was to use GPT-4 Turbo, just use that. otherwise try out Opus.

[tame current]

I see

Issue I am facing is that it just searches from google even when answer is wrong

Which chatgpt4 doesn’t do

[warm cave]

You mentioned you are using it for help in school, as a general rule if it is a STEM topic then GPT-4 will be more accurate, or if it is more creative like writing them Claude will be better. Personally I like Claude Opus because it is very good at breaking things down into manageable steps so it can tutor me, and this is often more of a benefit than the slightly lower accuracy. Of course that is mostly because I use it for studying and have the answer already and if it give the explanation but has the wrong answer then I know I need to switch to GPT-4 for that particular question

[tame current]

Why does the website still not mention the Opus limits not being included in the 600? Very misleading.

[icy zenith]

Gemini

...

[minor shard]

imo i wouldn't consider gemini pro a top LLM

but obviously depends on ur use case

[tame current]

It's excellent for creative writing.

I don't code.

Imagen 2 is also excellent and will be even better when they get rid of the racism.

[tame current]

I see

Damn just hate the limit

[austere kestrel]

try poe. if it's more about maximum usage, rather than the LLM being connected to the internet, then there's like 5 different variants of gpt-4 to use there, plus pretty much every other model.

[tame current]

Ag

I mean like I said @austere kestrel just want chatgpt4 with greater limit to answer correctly questions

[austere kestrel]

ya then i would i prob use poe or or the openAI API

[tame current]

Ok thanks

Yk the limit?

[austere kestrel]

~3000 / month, you could use them all up in one day (well, in theory), or spread them out

you will get more uses via perplexity

[tame current]

I see

Issue with Perplexity is that whenever I search something it just goes off the Internet rather than actually putting logic

So like if I give it a question or homework probably to help me solve

[austere kestrel]

but poe is just a wrapper. if you really just want gpt4 plain and simle, that is your best bet

[tame current]

Yeah, I actually have that but the 40 every three hour limit is killing me

[austere kestrel]

yeah i mean you can use Writing mode in Perplexity, which will prevent it from using web results

[tame current]

Yeah, I might try that

I have the pre-trial so it doesn’t hurt

[austere kestrel]

it's good value. but if you just want to use the model, no features which may or may not get in the way, either openAI directly or poe imo.
completely subjective ofc.. just my 2c

[tame current]

Oh thank you I really really appreciate it

[sterile falcon]

Perplexity in “writing” mode doesn’t search the internet. You’re using it in “All” mode, hence, it’s searching the internet. In “writing” mode, the answers are similar to ChatGPT, but a little bit longer due to Perplexity’s target users requiring details.

[tame current]

I got it, ok thank you

[arctic spindle]

what AI model draw that, because it looks so realistic

[lone citrus]

it's not ai work, just simple PS or photo editor something

[crimson stump]

Does anyone know how to remove the file restriction issue for new Enterprise users? I can no longer attach any files to my threads. The limitations message indicates it is for the first 7 days as a new user placed by the Administrator of the account (which is me) and I never added these restrictions nor do I see anywhere to remove them.

[spice lance]

is there anything that gemini 1.5 pro is better at than other models?

im using https://aistudio.google.com/app/prompts/new_chat how do i check my usage, no idea how I'm paying for the

[split sigil]

heyyy wspp

There is no code to have a free month of perplexity pro?

[warm cave]

Yeah, unfortunately it was removed due to abuse of the platform, currently the only discount I know of is the referral code that makes it $10 off

If you want I can send over my referral code, just lmk

@chilly forge 👀

Or you can use naivecoders link, haha

[chilly forge]

Hehe 😜 just kidding

[warm cave]

Out here getting my 10$ 😝 hehe

The model switcher user script broke, I wonder if perplexity broke it on purpose or not, lol

[chilly forge]

What ? Fr ?

Need to check

[warm cave]

Yeah

[split sigil]

much cadder? haha

[west sage]

Claude Opus still limited to 50?

[warm cave]

Yes 😦

[rigid shadow]

Hello, just joined in but Pro user since months!!! Cheers

[tame current]

Creative writing.

[torn ravine]

It’s free

[austere kestrel]

i don't think any other easily accessible model has a 1m token context window (next best would be Claude 3 models with 200k). Also, I can't think of any other models that take video and audio as input (in addition to images to text files, which models like GPT-4 and Claude 3 do too)

also just noticed they added JSON output mode and system prompt, which is nice to see

also it performs much better than I think is sometimes assumed. Like it isn't just about the massive context window and multi-modality. None of that comes into play on the lmsys arena - it's just the quality of its responses, compared to another model, on which people rate it. And it's currently tied for 2nd place along with Opus and several GPT-4 variants

[fast meteor]

that code is bad anyways, it changed the global state. i prefer hijack the prototype of websocket/post call instead

[tame current]

you can do that ? how ?

[tame current]

It's really really good for creating funny random scripts.

[bold badger]

What channel do you use if you have used a prompt with a certain model and it returns information that is incorrect?

https://www.perplexity.ai/search/What-do-you-Rp8piKvqRf6CC1oe4VHbcQ

[fast meteor]

if (typeof _XML_send === 'undefined') {
    var _XML_send = XMLHttpRequest.prototype.send;
}
XMLHttpRequest.prototype.send = function(msg) {
  \\your logic here
  return _XML_send.call(this, msg);
}

no library required, just plain old emsascript
socket.io will fallback to long polling if ws is not avail.

[sturdy crane]

I am a newbie with these AI tools, how is perplexity different from AIPRM & Chatgpt?? (note: I am planning on writing medium articles and I want AI to help me with it)

[tame current]

biggest difference is that perplexity can search the web

and it generally gives more accurate info than chatgpt

[spice lance]

what does this do?

[sick geode]

Hello guys, to the awesome perplexity builders, I’m heavily using the voice function to practice my english and to avoid typing on my iPad, but the new update is just verrrry quick as it responds while i’m thinking of some words to put. I liked the previous voice function too much as I can toggle when to speak

Can we expect something to fix this please? It’s very annoying and ruins one of my favorite perplexity features

[serene sand]

#1233677613919375490 message

[sick geode]

I believe it’s not a bug. It’s just a fine tuning related thing

I’ll be very happy to have an option in settings so I can make it continuously listen similar to our previous experience

[serene sand]

you can still use old version, just check what I've sent you

[sick geode]

Yessssssssss

[sick geode]

Thank you brother

Just found it in the last reply

Thank you dear ❤️

Please don’t remove this guys ❤️

[austere kestrel]

it gives outputs as JSON objects 🙂 e.g. here's copy pasting a few messages from here

you can do it without JSON mode enabled through prompting, but as you can see here, the output is a JSON object but it's enclosed in a code block (would need further prompting to stop that)

it's useful for workflows / agents, where you want to pass outputs between LLMs (or into different processes / storage)

[green thicket]

Kinda annoying that they rely this much on Cloudflare. What are they fearing?

Also, how often will I be able to apply this $10 referral discount? Is this a one-time thing only?

Or could I simply subscribe next month with another referral link and get $10 off for the next month, too?

[tame current]

people scraping the LLM responses

I can definitely see why that can be a problem

[green thicket]

Sorry if this sounds ignorant but how are they actually wrapping the GPT 4 Turbo service from OpenAI, does Perplexity pay token by token, similar to what users can do on OpenAI Playground (or OpenAI Platform nowadays), or are they in some kind of flatrate contract with OpenAI?

[tame current]

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

[tame current]

I don't think they revealed that but it's possible they have some contract

[green thicket]

Thanks, that helps

[stoic wave]

They simply pay for API usage, and that is the only way.

[serene sand]

correct

#💬│general message

[lost tapir]

tbh this isnt a good way to know good a model performs

[austere kestrel]

yeah but there's two ways to pay for API usage. Pay As You Go (what we're familiar with), and "Provisioned Throughput" - I'm quite sure that a company like perplexity is the latter
https://www.perplexity.ai/search/explain-provisioned-throughput-GSQ4WRfuSjaQ1PcFD32zVw

[lost tapir]

yes gemini 1.5 pro has a very big context so its useful for that type of stuff but in coding and any other type of stuff it doesnt hold a candle to claude 3 opus or gpt 4 turbo

with better text formatting highlighting and other things votes are a bit biased

[austere kestrel]

(*side note: AWS bedrock offers provisioned throughput pricing for every Claude 3 model except Opus...)

[austere kestrel]

no better way than actually using it yourself for your actual use case

[lost tapir]

for example if you've got documentations about what you are coding about its very good at that

but if you wont be making use of that context its not very good

[austere kestrel]

but as far as benchmarks go, i would always give more weight to the arena leaderboard over a model provider's self-reported evals - though i'm sure there are other benchmarkin projects out there (and it all comes down personal use cases at the end of the day anyway)

[lost tapir]

both type of benchmarks have their cons tbh

normal benchmarks are more about technical side this lmsys is more about human preferences

but you can train the model on the benchmark so its also not very good

how about closed benchmarks

[austere kestrel]

ofc - i mean most evals are just entrance exams for various different academic disciplines or generic tests for 'reasoning' competence or coding or whatever

there's no holy grail

[lost tapir]

i've got a math problem and every model out there fails lmao if one day one manages to do it i'll declare it superior to human intelligence

[austere kestrel]

yeah this is a big problem..

contamination

[lost tapir]

tbh i see a lot of reasoning tests but model is already trained on that

when you make a unique reasoning test it fails

sure they might be capable of creating unique content but they are not good at reasoning yet

[austere kestrel]

often yes - or if just add a transivitive property to something super basic many just lapse completely false CoT 'reasoning'

[lost tapir]

like sometimes it knows how to solve a problem

[austere kestrel]

this kinda of thing gets me every time (same with the 1000kg of feathers and 1000kg of lead questions..)(

[lost tapir]

but it doesnt know how to apply the method

it cant adapt that method to the situation

[lost tapir]

honestly gpt 4 turbo is clearly superior

claude sure its good at writing and stuff but fails math and reasoning

their benchmarks told it was superior to gpt 4 too

so not really accurate benchmarks

[icy zenith]

WTH

[lost tapir]

training on benchmark dataset is all you need

[icy zenith]

the app lets you switch models in the middle of a conversation

[lost tapir]

im waiting for gpt5

[icy zenith]

Probably not this year

[lost tapir]

i mean who else is going to release groundbreaking stuff if not openai anyway

[stoic wave]

How are you so sure there will be a GPT-5?

[lost tapir]

they hired best of the best

they got all the budget in the world

[lost tapir]

why there shouldnt be gpt5

[lost tapir]

i'd say late 2024

presumably after election

[mossy flame]

The process of developing AI will continue, so there will be a GPT5 for sure

[lost tapir]

considering sam already started gpt 5 hype it would make no sense why there wont be gpt 5

[stoic wave]

I'm not saying there won't be a newer model, but why exactly naming it as "GPT-5" it could be anything.

[lost tapir]

oh thats just a guessing

who knows they can name it anything

[stoic wave]

Yeah, thats what i am saying.

[lost tapir]

openai is about to release their own search engine with ai

they have already integrated something similar to bing

its probably openai testing their stuff before release

[mossy flame]

I think its bing, probably just renamed. Microsoft invested some Billions I think so they won't use Google or something like that

[lost tapir]

well ye probably

how would you be google competitor while using google for web search lmao

openai hype

you love to see it

i wonder if meta will just opensource the 400b version too

[icy zenith]

You guys think it's a regular search engine

Nah

Its not going for Google

It's going for Perplexity

[stoic wave]

+1 here

[lost tapir]

whats not good for perplexity is that they are stuck with the models openai antrophic or whatever provides them

but those companies can fine tune their models more fit for this type of approach

[icy zenith]

Tbh

AI knowledge engines are too slow to be considered a search engine

And too inaccurate

[lost tapir]

google isnt good enough anymore

its filled with trash

[icy zenith]

We need to separate search engines and knowledge engines

Google and Bing are search engines

[lost tapir]

ai itself might hallucinate yes but it will search for it on the web

[icy zenith]

The AI tools they have (SGE, Copilot) don't make them knowledge engines

Just search engines with AI-enhanced summaries

But when the AI response is the main way of consuming the web search experience, like Perplexity, you.com, and the upcoming ChatGPT search thing (i think), then it should be considered a knowledge engine

Search engines provide links

Knowledge engines provide direct knowledge

Can a knowledge engine replace search entirely? No, as it requires search to even work

But can it possibly replace them for someone in their daily life? Yeah, as long as you don't need instant speeds

[austere kestrel]

yeah and also the bulk of online searches are still people just waning a single, specific of data - which needs to be right. A company's URL, phone number, opening hours etc.. knowledge sourced from the internet and curated by an LLM has its place, but that's a long way off ever replacing the most basic (and popular) use case for online searches imo

[pure jackal]

is it possible to make own knowledge engine?

[fluid mauve]

I would tend to agree on the terms Search engine (google, bing, DuckDuckGo) and Knowledge engines like Perplexity, you.com and others. They are currently two completly different beasts. When you do a search in a knowledge engine, and want to validate its authenticirty this is where citations come in. The majority of which have been indexed by a search engine.

LLM's, Knowledge engines et all, are in their infancy, I can see a time in the not so distant future where the average user will not be able to tell the difference of where the data came from. it will just be a response. I vaugley remember pre-google back in the late 80's where searching was a nightmare, nowadays its simple as pumping in a query and a search engine will send a resoponse in milliseconds, Knowledhe engines maybe a few seconds.

Google has expended many billions of dollars setting up search heads all over the globe, retrieving and indexing data for a decade. They hold a massive corpus of link data. It will take knowledge engines seperately may years to even get close to what Google has in search and response.

I initially thought that companies like Perplexity, OpenAI, You.com could get indexers to work for them and retrieve and augment data as required, but why would they, Google already has that data, it just needs to be harnessed.

I see this more as diversification of revenue streams for Google, providing fast accurate data to Knowledge engines. it improves Knowlege engine response time and it still allows Google to provide citations peppered with ad links. Granted this isnt ideal, but as long as the knowledge is not 'tainted' by sponsored ads via Google I think this can work.

But I stand here, ready to be shot down after my little ditty produxced during my lunch break. (actually I am stitting)

[devout cargo]

@icy zenith
So Perplexity can be defined as a generational search engine?

Perplexity and Google SGE compete in the same segment?

[fluid mauve]

No I still think it is important to look at what each actually provides. Google is in the search business, Perplexity is in the Knowledge business. Both have distinct uses. I dont go to Perplexity for Papa Johns phone number, thats definitly a Google search position. I would use Perplexity to craft a cover letter for a particulat job I might be applying for. Igf i gave teh same task to google I would get ba list of sites that could do that for me.

You also dont get to utilise different LLM models on a search engine, well not yet I dont think

[austere kestrel]

someone else shared this here - was a good read imo.. lays out this idea of three tiers of online search:

• navigation
• information
• exploration

https://www.theverge.com/24111326/ai-search-perplexity-copilot-you-google-review

and says the overhwhelming majority is navigation (give me the link/basic info). perplexity occupies Exploration (or Google's "immersion" ig ha)

[tame current]

I love Limmy. Also. Steel is heavier than feathers.

[kindred folio]

can i connect apple id and google to perplexity account?

because i sign in with google but on my school managed chromebook i can only manage to sign in with apple id and i need to link it to the right acc instead of it being separate

[devout cargo]

I am not an expert, I compare Gemini with Copilot. And gemini is the Google SGE model. Perplexity uses a variety of LLMs in its search engine. The trend is that search engines are like Perplexity, we will see when Open AI releases its search engine.

In Spain to make an appointment at the ITV you have 20 pages that make you pay €20, that is a scam because it is free. With Perplexity it gives you the answer directly, the ITV telephone number and the official URL

[opaque stump]

Looking forward to this rumored Perplexity competitor from OpenAI 🤞🏽

[chilly musk]

Is the service down?

[opaque stump]

Working for me

[livid mantle]

fried eggplant

and soy sauce

[half venture]

Is opus still 50

[wraith crow]

#💬│general message

[serene sand]

#💗│sharing message

[kindred folio]

six days 😩

[brittle temple]

Hi guys,
So I asked here before about my issue trying to redeem my discount code. I’ve contacted customer support 4 times already with my information. I have not gotten a response back from the last 3.
Are they usually this bad? Any similar experience?

[halcyon coral]

I have not gotten a response back from the last 3.
You have sent four e-mails but only received a reply to one?
When did you send the first email?

[sick geode]

Anyone noticing that Pro search is becoming toggles off by default? I find it very limiting to always turn it on…

[brittle temple]

About a month ago. I had rep Josue responded and asked for my information. Discount code, email, sale order (rabbit r1). I sent it, about a week later I got an automated message saying case will be closed if I need further assistance to responsed to email. I did. No response after that. Sent new emails and nothing back.

[signal hamlet]

Hey @brittle temple! Could you please dm me your account details, I'll check.

[livid mantle]

omg whre to apply Mr. @signal hamlet

[shut plaza]

what exactly is the Plexer role for?

[serene sand]

#👋│introductions message

[shut plaza]

ah thanks

[minor shard]

The #⭐│starred channel has been really useful at seeing what is trending on the discord server for Perplexity.

[signal hamlet]

The tester role application was in the announcement above that one but it's closed now.

[glacial tiger]

What can you do?

[warm cave]

I have to say the new voice conversation sucks to a degree where it's unusable, it's promising, but the TTS is very inaccurate and it never lets me finish speaking before starting, I could say “how is the weather?” and it would stop by time I reach “the”. If I speak fast it works better, but I stopped using it altogether because the regular whisper TTS works soo well in comparison. Hope it improves in the future

[livid mantle]

🤖

[signal hamlet]

Thanks for the feedback, we got similar reports and the mobile team is working on improving the voice conversation experience

[wide mesa]

I cant seem to access any of my threads and library rn?

or is the site just slow today in general

nvm all good now

I find it interesting that it Sonnet works better than Opus sometimes

[sour bough]

Hey guys quick question. What's the best model to use within perplexity right now? Opus or gpt 4?

[signal hamlet]

#💬│general message
#💬│general message
#💬│general message

[thorn python]

😁

[warm cave]

Perfect, thank you for letting me know its in progress 🙂 , I am looking forward to using it once it is improved!

[silent magnet]

What do you guys generally get better results with? GPT-4 Turbo or Opus?

[warm cave]

Depends on the topic, if it's STEM-related questions than GPT-4 still has the edge, but if it's creative then Opus is the king. I like Opis by default (except for the limit of 50per day, in that case I try to save it)

[regal stirrup]

file size limit is same on pro? Or bigger?

[silent magnet]

I had "Okay" results with OPUS for coding questions, catastrophic ones with GPT-4 Turbo. Python3

But English-wise OPUS felt king but by far.

But I do feel like there was a significant degredation in OPUS since launch.

Maybe I'm crazy

[wide mesa]

why

it does this only for one of my theads

a few threads now wtf

why is it doing this

[jovial mantle]

Are you using a VPN?

[silent magnet]

Open the developer tools and go in the network tab

[wide mesa]

nope

[wide mesa]

in where? Perplexity or the browser

I really damn need the contents of that thread rn

[wide mesa]

what am I supposed to do here?

I cant access most of my old threads now

[silent magnet]

Click "XHR" to filter by that, then try the button again. Check the "Response" tab in the row that just appeared.

Ah, I don't think debugging is gonna solve that one.

[dusk talon]

Is there a dedicated support team on here?

[chilly musk]

#💬│general message
me too

[wide mesa]

Clearly not if they're doing fc all for this issue

[dusk talon]

Fc?

Sorry bad with lingo

[wide mesa]

it wont let me make a damn new thread now

[chilly musk]

me tooooo

[tame current]

Do you have a Ideas for prompt engineer to reduce Claude 3 Sonnet hallucinations?
It misinterpreted what my request Is sometimes.
Opus Is great for coding and writing though.

[sly minnow]

Some of my treads still works but some don't for some reason.

[chilly musk]

other peoples also have this situation
https://discord.com/channels/1047197230748151888/1236008988919988284

[restive blaze]

Hello just recently came to know about this Perplexity AI.
I'm seeing so many GPT apps recently like Chatgpt, Poe, Bing Copilot, Perplexity AI etc

Which one of all of these is best for free users? I'm so confused.

[chilly musk]

I started having this problem last night

It even kicks me out of my account if I refresh on this thread page

[vapid onyx]

What have you tried to resolve the issue?

[chilly musk]

I can't solve it, I have to create a new thread to use it

[wide mesa]

For me creating a new thread sometimes work and sometimes doesnt

Typically doesnt if my question is long

[chilly musk]

me too, and creating a new thread requires copying the problem from the previous thread again

and it wastes 50 opus uses per day

[wide mesa]

I cant copy the problem because it just wont load

[chilly musk]

so... today I purchased the claude pro subscription😂

[wide mesa]

I really liked and needed perplexity's yknow automatic academic sources so I'm not using claude over it

[chilly musk]

I need claude opus to help me with my school paper

In writing, gpt is not as humane as claude

🥲

[wide mesa]

Use it to paraAI to paraphrase it

and run through detectors

[half venture]

Copilot

[dark idol]

Yea but copilot slowwww, perplexity is much better if U want speed and slightly more accurate answers, copilot is just smarter when it comes to other stuff

[wide mesa]

Copilot is wonky and wont give you accurate academic articles most of the time

[worthy mesa]

Both give sometimes unrelated sources compared to the answer given

Kind of false advertising

[obsidian acorn]

perplexity broken

[restive blaze]

But isn't it super slow

[restive blaze]

For general questions which is better

[half venture]

It's gotten faster

And for general questions copilot offers gpt 4 for free

Perplexity doesn't

[restive blaze]

Okay I'll try both for a few days to see which one is better overall

I've seen one thing. When I ask something on Copilot and while it's answering it, if I minimize the app, it stops answering and gives error.

[upbeat dawn]

Plz info how many convo u got daily on anthropic opus?

Opus at pplx not smart like from anthropic one imo

[chilly musk]

45 messages every 5 hours

reset every 5 hours

[warm cave]

Keep in mind that it is normally less than 45 per 5h, using more context dramatically reduces it. I normally get closer to 25 per 5h. I still find it worth it but it's good to know since you may want to start new chats frequently and give it less context if you need the full 45

[chilly musk]

it will provide a warning before you have 7 messages remaining

[warm cave]

It will, yes

[chilly musk]

I like the context of 200k

[warm cave]

Yeah, that is why I say it's worth it, even if you don't get a full 45

[chilly musk]

I can send my paper.docx directly to it

[warm cave]

Love it so much actually, use it so much every day, max it out every 5h

[chilly musk]

https://tenor.com/view/baqua-gif-22467620

[warm cave]

I know you can do pdf, but not sure about .docx

Lol, I answered a question you did not ask. deleted like it never happened…

[chilly musk]

I tried it. It works

[warm cave]

Nice, I mostly use pdf and txt files with it. Always works great

I like using obsidian, so it works well

[tame current]

What Is that?
Is that Claude website?

[chilly musk]

yea

[tame current]

I think I might subscribe to Claude Instead of perplexity.

Is Original website better?

[chilly musk]

In my experience...the webpage interaction of claude was not as good as that of gpt webpage

latex is not supported, can't edit sent questions

[tame current]

Oh damn.

[chilly musk]

incomplete markdown syntax support

[tame current]

Opus Is so Good that perplexity have to limits usage to 50 messages per day because user used them too much lmao.

Coding never been this easier.

[stoic wave]

Is it your first time using AI, i see why are you so excited

[devout cargo]

Seriously with Google Gemini as a personal assistant. you can use home automation you can send WhatsApp you can make reminders now we have extensions for everyone alike you can make email summaries you can search on Google Maps you can search for hotels you can search you can already do many things. Knowing all this, is it worth buying Rabbit R1?

[tame current]

Yeah and I would be this excited for the next two years onward.

[placid coyote]

hello, is pplx doing some weird postprocessing on square brackets in inline code? I don't think this is fault of gpt4t

[stoic wave]

Good for you, but i won't

[tame current]

What happened?
Did AI performing quality dropped?

[devout geyser]

Maybe try asking it to answer in a codeblock, assuming nothing in its answer uses triple backticks that is

[stoic wave]

I could agree to that, but lets see what GPT-5 will offer us, maybe then i'd be excited.

[tame current]

Claude gave OpenAI a hell of a war to fought, literally Claude 3 slap OpenAI In multiple ways and I think OpenAI GPT-5 maybe a counterattack to It competitors.

[devout geyser]

I'm looking forward to seeing more about what that mysterious LLM was on lmsys recently

[stoic wave]

Yeah, and in my expectations I think it's gonna be super good, hopefully not disappointed.

[placid coyote]

it responded correctly in a code block higher, but then explained it like this in an inline code. it's confusing as hell when it writes one thing and in an explanation text bellow, in a same response, it uses a different code than what it is supposed to be explaining 😞

[tame current]

I hope so, If OpenAI's next release Is disappointing It would make them taking even more L In my eyes. there multiple competitors that smoke GPT3.5 and GPT4 multiple times In a row and If OpenAI's next release Is mediocre then It would be very funny.

[stoic wave]

Yep, definitely agreed.

[placid coyote]

looking at chat, I guess opus is still caged? in my experience opus was much better at following "pre-prompt" (collections prompt). with gpt4t I have to often either remind him again, or rerolling an answer 2-4 times. I am not sure this is actually cheaper for perplexity. well, I guess not many people are using it this way

[tame current]

Anyone have a Ideas on using AI such as Claude to help with planning and strategy?
I think It possible but I don't have specific prompt that would work.

[icy raft]

I've noticed an issue with the Writing Mode where it keeps searching the internet instead of processing without searching the internet. Can this be fixed?

[deep pasture]

Gm

How can I get my pro-status? I have pro but no role in discord

[serene sand]

#💗│sharing message

[tranquil forge]

@tame current OpenAI 🥵 CloseAI

[tame current]

I think an official Chatgpt search engine would be a serious serious danger to Perplexity. If I had advice to give them, I would emphasise the variety offered. They won't go beyond Chatgpt 3.5/4/5.

[valid ruin]

maybe.. maybe not but perplexity is banger than other for now

[signal hamlet]

You can add your failing threads to this report: https://discord.com/channels/1047197230748151888/1236180984622219274

[tame current]

I meant the variety of models offered.

Chatgpt aren't going to offer the choice to select Opus or Llama.

[chilly nova]

can you guys confirm that you is better than perplexity with no limitations?

[tame current]

The You UI is horrible.

And the app is terrible.

[tame current]

perplexity is better

[chilly nova]

but you should use your IDE not the UI

major defect of all these tools is that you cannot use it inside your IDE directly without paying for the tokens individually

you should just pay the sub and be able to use it inside your editor

[tame current]

Agreed.

[chilly nova]

it's annoying af every time to do a repo2txt export to pass the context as an embedding

takes so much time

and you are also limited in perplexity so i have always to use gemini 1.5 pro until it's free cuz 1M context

[tame current]

Since my thread Is damaged from error so I will copy and paste all query Into other thread to continue my work for a while lol

[sinful pine]

guys

i cant paste any html code my perplexity

[gentle spire]

I have encountered this error many times this day, severely affected my work, what’s happening?

[small aurora]

I have the same.

[sinful pine]

some time i think maybe use another ai with my money because most time perplexity add a new feature to app the app is goes wrong, sorry for my eng level

[chilly nova]

is there a reason why perplexity doesn't have llama3 btw?

[sinful pine]

@blissful raptor

perplexity.ai somehow breaks before every new feature and I can't use the app even though I paid for it...

and this happens all the time...

ok the app is very nice, but unfortunately if it's going to break all the time, why am I paying for it?

the turkish response structure has been broken twice, this time with this ridiculous problem, if it breaks again before a new feature I will cancel my subscription!

[serene sand]

iOS or Android?

[tame current]

If you guy have this error, It seem to only happens to old thread before the error happened, so I copy everything from my old thread and create new thread.

[sinful pine]

everywhere...

[serene sand]

#1140626423707607040 message

[wide mesa]

So old perplexity thread got nuked I take it?