💬│general

[tame current]

Wouldn't trust AI with selecting the interesting topics myself

[sleek vortex]

making an automated discover tab is a bit of a long process

youd be looking at all of viral news sources and twitter and whereever

then again all that would still be cheaper than paying a salary to somebody

[tame current]

You could filter Google trends or Slashdot daily, but AI doesn't know definition of "interesting" and probably won't check for duplicates. Moderation would be required

Seems like Discovery is mostly about tech and science news and such,

[sleek vortex]

discovery is like

in my head im thinking like "stories a more advanced sort of perplexity user would want to see"

thats sort of a vague/weak link but

but yeah they havent really generalised discover into general news

[empty wave]

Thanks for the explanation! 👍

[grave coyote]

On the contrary, one of the most common and oldest use cases of AI is suggesting content based on past interest. That is how every social media feed and many news feeds work and have worked for years. But it is a different type of AI than an LLM.

I could see perplexity taking off in an interesting direction if it sort of let people compete to make the most interesting pages and then added a way for people to have pages recommended to them based on interest with some type of like/dislike voting system.

[grave coyote]

So you create a great page with marvelous info, then what? How do you share it on other sites like a personal blog or how do you drive viewers to it? Are people just sending links to the perplexity page or are they importing the pages somewhere else, and if so, where and how?

[south kindle]

I would like to see an additional area like the discovery feed, but for newly created Pages. I think you make a good point, we can get a really nice looking Page, but then what? aside from sharing the link directly, there isn't really a "good" way to share it to the general public. Part of the "joy" of getting stuck in a wikipedia rabbit hole, is being able to link to other topics and pages very easily. I'd like to see something like this built in. A "feeling lucky" option to just pull up a random page. I'd like to see hyperlinks to other pages built into Pages itself, so if we see a term or topic with a created page, we can keep exploring (without having to generate new threads/pages).

Some kind of moderation would need to be done on created Pages, but maybe an "upvote/downvote" system could be implemented so users can rate pages. highly rated pages would get pushed up in the feed/algorithm. highly rated pages would be featured, etc. If a page violates some kind of rule, a notice gets sent to the user asking them to adjust it so it will be "published" again.

For topics that have ongoing updates, something would need to be implemented to add to the page (similar to how wiki pages get updated).

[sleek vortex]

funny how it’s from the same number lmao

and claude too

[south kindle]

that's pretty standard for 2FA/OTP services. not sure how it actuall works but can confirm i've had the same before.

[sleek vortex]

Yeah

But it’s a coincidence all ai model providers use the same sms provider

[grave coyote]

Yeah I popped it over to the ideas channel.

It would be easy to add a "if you liked this, try..." to the bottom for an algorithm to keep you in the rabbit hole. I was suggesting it would be something similar to Medium, but AI generated with human guidance. It would be much higher quality than Medium overall I think. Then it would be up to creators to make pages on interesting topics.

[south kindle]

posted my notes as well. i'm excited to see the direction they take with Pages.

[grave coyote]

Yes I personally find my use of Perplexity going in two directions.

1. I need research to support my profession
2. I enjoy putting together good research on topics that just interest me.

I have already enjoyed sharing threads with people at work and it has been really useful. Pages is a great next step, but honestly If I am going to make it that pretty I am going to want to share it with the world and hopefully get some type of feedback and interaction with it.

They could probably make an interesting monetization model allowing anyone to read generated pages but requiring a subscription to create new pages (or maybe as they plan, more than a few pages).

[tame current]

You need subscription to create pages anyway

[agile jay]

Yep, they just need a button revalidate generations if you want them checked.

[tame current]

Unless you're beta tester, it seems I had access without pro

[grave coyote]

They are planning on releasing pages generally soon.

[agile jay]

Seems like it was automatically using Opus, which made a lot of people run out of opus unknowingly.

Since pages generated multiple times until it's done.

[grave coyote]

Opus is a great model for good writing, but I think it was a bug that you couldn't change it.

[tame current]

It's probably intended, so pages and discovery pages have better quality than ChatGPT and make people buy Pro, thinking it has some magic applied

But the biggest problem is lack of languages other than English, despite the app settings and instructions

There should be "Translate" button somewhere or something

[shrewd hazel]

Not a coincidence, that company cornered the market

Twilio

[south kindle]

I'd assume "each" section of the page was a separate OPUS use as well.

[grave coyote]

From a business perspective Pages is a brilliant move. Perplexity currently leads in a research format, but they aren't using their own model, they just build some good bits of code around using other company's models. Their ability to stay ahead of others will need something more becuase that advantage can easily be lost. If they could take it to the next step and present the research in a finished, polished format it places them that much further away from others. Producing web pages, nicely done PDFs and getting it in a format that can easily be used on other platforms really makes it a much more complete product. I think becoming a hub to host the polished content would also be a good move, but even if they don't pursue that, Pages is pretty cool.

[tame current]

I doubt anyone outside existing Perplexity users will see it. It's pretty hidden and not indexable by Google. It would be better to focus on the main product, which is search. Visualizations using generated images or diagrams would be much better than another unused feature. Actually working search suggestions, etc.

I mean, it's leading on the search quality, but nothing else. Improving Perplexity model would be another important step

[grave coyote]

There is no "Perplexity model".

[fading wedge]

" It's pretty hidden and not indexable by Google."

Well, this is by choice. Not saying that they should go all in on pages.

[tame current]

Sonar begs to differ, it's Llama 3 finetune. But it's worse than the original model

It used to be usable model before they went on Llama hype

[fading moth]

https://discord.com/channels/1047197230748151888/1244758414794428516 Don't forget to post your Opus 50 art here! :3

[south kindle]

if I recall, their planned future features include research paper creation, etc. they've even directly said they want to make an AI Wiki, which would require non PPLX users to be exposed to it. curious how they accomplish that.

[sleek vortex]

diagrams seems like a wanted feature

and pages export

hmm

[south kindle]

AI generated diagrams are just kind random though. usually with mispelled words. we would need them created with data and not as an AI generated image.

[sleek vortex]

yeah

might experiment with that

see if i can make a proper diagram making fine tune

using something like mermaid

[agile jay]

Diagrams aren't that hard to generate accurately.

[sleek vortex]

yeah

it could even just be through prompting

[agile jay]

It's just the method they use are sh*t.

[sleek vortex]

lmao

[agile jay]

Simply go from table to graph.

[grave coyote]

That's really interesting. I don't know if a wiki model would work well, but it would be interesting to clone pages and then publish improved copies. Not sure how much the original contributor would like that but that is basically what we are doing to the rest of the web already so...

[grave coyote]

Yeah this is a totally different kind of AI than an image generator though. None of the current ones are designed to do this.

[sleek vortex]

give me an example

of a prompt

that would require/asks for a diagram

[south kindle]

stock price changes for various leading tech companies?

Or somehow chart product utilization over time via active user counts.

[sleek vortex]

so youre thinking more like actual charts, say from an uploaded file

[south kindle]

that was just the first thing I thought of. ha.

[sleek vortex]

or from web data

[south kindle]

ideally you could do that via pulling updated web data. just using PDFs would be similar to ChatGPTs features I'd think with data analysis,etc.

[sleek vortex]

i think maybe a mix of code interpreter and access to lots of different datapoints could be interesting

say like

yahoo finance

etc

[south kindle]

I don't know they "directly" stated they are on a roadmap, but if I recall, all of that was mentioned in the Pages information they put out with the release.

[umbral pollen]

I'm struggling with code interpreter and the limits of its use

[grave coyote]

https://www.heavy.ai/demos this is designed more to do data visualization. AI has actually been at this much longer than the current LLM craze.

[sleek vortex]

idk im sorta trying to build some form of hybrid of an agent pplx, with chatgpts data features and more

im thinking about how code interpreter could be executed better

chatgpt's new design is good but it still has its limits

[umbral pollen]

as in: can I give it Any code?

[sleek vortex]

e.g no web access

[umbral pollen]

the no web access thing is killing me

[south kindle]

https://www.heavy.ai/heavyiq/overview > their presentation is 10/10. It's clear they've been at this longer than a year (since the AI boom really started).

[umbral pollen]

pplx is great, though I wish they'd take me off waitlist for sources

[sleek vortex]

sources?

in the api?

[umbral pollen]

I feel like we're so early on in these technologies when we can't get the two main players to give out APIs that connect to the f*in internet

how can Google Gemini not be able to Google things? it just seems silly

[south kindle]

do you mean via API, or thru their chatbot? i believe basic search is in there, but it depends on things that have been indexed.

[umbral pollen]

api

[south kindle]

Yeah that's janky. lol. haven't messed with their playground/api's much other than testing out gemini advanced back when they did the 2 month free trial.

[tame current]

ChatGPT does it with Python code interpreter. Excalidraw does it with Mermaid diagrams. Both are better than Perplexity diagrams, which often look like schizo ramblings with giant diagram for small thing, all unrelated

[umbral pollen]

i created a simple but pretty cool app called CompAIr: you enter one search and it returns results from GPT-4o and Gemini Pro so you can compare (get it?)

[south kindle]

relates back to my point. we need them generated via Data and not as an AI generated Image, which has many drawbacks for text.

[sleek vortex]

diagrams from the web would require robust datapoints

[umbral pollen]

Have y'all been using Gemini's API? Its like 90% less expensive

[sleek vortex]

you cant trust an llm to scrape 1000 datapoints off some html page then run code to plot it

youd just have to abstract out the data bit

[south kindle]

https://app.wordware.ai/r/519f528c-2996-4f0e-8815-50a6515f7c27 > tool I threw together on wordware that sends 1 prompt to multiple models and ranks them. no web search though. still nice to test prompts with different models quickly. (hasn't been updated in like a month, so 4o isn't included, plus the last time I went to update it, I broke it and had to rebuild it from scratch, lol).

[umbral pollen]

this is a scary thought, though in my experience, you're right

shouldn't it be able to do that?

[sleek vortex]

well i think you have to entirely avoid bringing the actual dataset into the model's context window

you could get it to manipulate the actual html which would be similiar to what i do myself sometimes

[umbral pollen]

what I mean is I've been losing my religion a bit with ChatGPT because it pretends to do more than it does

[sleek vortex]

since theyre text prediction models
i really doubt its going to accurately reformat like 100 data points if it cant even reformat a simple program into a different language

[umbral pollen]

I give it data to read and tasks to do, i check it work, and its only read 5% of the document

but presented its answer as if it had read the whole thing

[tame current]

If you're generating a plot with like 20 points, it shouldn't be a problem. Any model with web access does it efficiently, if web access is done well

[sleek vortex]

asked it "meow" (my heads full of cats) and it said haiku was best

[sleek vortex]

yeah

but i think large scale analysis on non-uploaded files is something nobodys really done yet

[umbral pollen]

I'm afraid of LLMs leading to too much of this:

[sleek vortex]

this is so interesting

opus being way more strict in it's "over-alignment" than haiku

[grave coyote]

You generally can't make assumptions from a single generation. If you ran that three times in opus you might have gotten each of these results. If you ran it again in Haiku it might have given you one of the other two responses. You have to run a query several times to be sure it will give a mostly consistent answer.

[south kindle]

with the topic we are discussing. I always enjoy seeing the adjusted ranking to see what it caught the 2nd time. Haiku seems to excel anytime it's used in wordware for some reason. ha.

[heavy portal]

I'm on Pro account and over the past month Perplexity isn't remembering my Light/Dark/System setting (defaults back to System when I want it to be Dark), Is this happening for anyone else?

[south kindle]

We could actually build that into wordware. Just have it send the same prompt to a model a few times, and then have it rank it at the end. would be a pain to change the model though. we would want to make a "template" and then duplicate it for the various models in question.

their entire platform is mostly meant to prototype ideas AND to get an API key from the project you can use in your own applications. they haven't introduced any kind of "paid" tier yet, making it pretty open for anyone to build with. pretty cool.

[south kindle]

might be a fun test specifically for Claude models that sometimes don't play along consistently.

[foggy whale]

Could do that with msty

[south kindle]

haven't used that yet. This is a pain in wordware. you can't tell the difference between answer 1, 2, 3, etc. so i just keep picking random "answers" and it just outputs the same reply twice. thats the extent of me messing with it for now until "answers" are designated somehow. from the drop down, I'd assume the original answer stays at the bottom of the list? but that didn't seem to help either. picked both and got the same reply twice.

[south kindle]

From the developer. i need to use a Loop, or adjust the creativity setting. interesting. Very cool being able to just ask them directly and get a reply in less than a few minutes. ha.

[wanton crest]

i dont find the pages enabled on my account, how do I get it enabled?

[sleek vortex]

clear your browser cache first

you need pplx pro as of right now

[sleek vortex]

random, but "opus insight" sounds like a fun company name

[wanton crest]

I thought its available for free version as well

[sleek vortex]

not yet no

[wanton crest]

thanks for the confirmation!

[south kindle]

I actually had a separate thread going on names. Opus came up with that one itself so I stuck with it. It rolls off the tongue well. after I made it and shared it, people were referring to it by name which made it easy. instead of "that wordware template that tests multiple models and then has an instance of OPUS do a full rating/verification".

[south kindle]

Alternatively, I had the thought to transform the scratchpad framework into an online tool and then name it OpusInsight. Alternate name was Second Brain or Brain 2.0. If anyone wants to help make that a thing, shoot me a DM. Idea hasn’t gone away. I think about it at least once every few days whenever I use my scratchpad collection. Ha. I don't have the technical chops to build it.

The tragic life of an ADHD person with unlimited "good" ideas and visions that just float into the ether because they require specific skills/knowledge that I likely don't have the motivation to get thru learning entirely before losing interest in the initial goal/project..../end ADHD life rant...||(fingers crossed the ADHD 2nd brain tool gets made and I can have a BCI to help manage it...)||

[glacial herald]

So what’s the limit on how many pages we can generate?

[south kindle]

I don't believe there is a stated limited, or that any user here has made enough pages to hit it yet. we know initially it was using Opus for the generation, which caused people to run out of Opus uses without realizing it.

[glacial herald]

Oh thank you.

[rustic stream]

Is Copilot currently unavailable?

[tame current]

thru dms or ping?

@harsh stag I use perplexity alot, can i beta test

[harsh stag]

check ur dms

[drifting stone]

Is this a browser extension or nah?

[brazen moss]

is there any possibility to create custom gpt in pro version? i have been using poe for a while. i want to switch to perplexity.

[harsh stag]

you can do that with collections but it only offers system prompt

[opal hollow]

how can I change the title of a page? when I convert a search thread to a page, the system get my main ideas wrong.

[torn ravine]

Neat UX👍

[opal hollow]

will this featured be added in future?

[sleek vortex]

swear this feature used to exist

[south kindle]

I thought so as well. can't seem to do it now. we can "rewrite the intro" but thats it.

[vagrant herald]

how to activate?

But is this approved from Px devs or it's something like 3rd party?

[austere kestrel]

well i don't think they've disapproved so.. we good 🤞

[past drum]

HI all

Anyone knows which AI detector is useful?

Want to check AI plagarism for my assignment.

[stable radish]

Late answer, but I feel like nothing is really difficult to learn (unless you’re 65 yo) It’s just about finding the right way for you to learn.

[austere kestrel]

ha yeah to be concerned about plagiarism is arguably to have already failed one plagiarism test..

[past drum]

Nope. Why?

turnitin

yea, that's why I am asking for useful and high accuracy AI detector, I need to rewrite the part that flagged as written by AI.

Thanks for your sharing bro

[sleek vortex]

just use ai for the ideas

then write it tourself

ai detectors are mostly all a joke

like at some point how is it even going to detect ai or not

on the online detector

or from your actual like marking

yeah

idk i would just write it yourself despite what any online ai detectors say

yeah the real “ai detectors” which some companies like turnitin use aren’t public are they

like actual models which they might have trained or something

i would just trust that over any random online detectors

[royal tangle]

the Perplexity Pages is amazing but how can i download the full page to use it in different platform ?

right now its only to share it in Perplexity

[torn ravine]

What model is used when converting a thread into a page? The default one in user settings, or the last model used in the thread?

[torn ravine]

Let's say I use Opus for a thread. I don't want to exhaust my 50 quota because generating a page can run out of it pretty quickly I suppose.

[scenic walrus]

It's been in this situation since last Friday, do anyone know how to solve it？

[torn ravine]

Weird. I used GPT-4o at the beginning, switched to Opus and then back to GPT-4o at last. Convert the thread into a page and then every time I ask it to write something, it deducts my opus balance.

Also judging from the output, it surely uses Opus.

[austere kestrel]

yeah i don't think it really "converts" the thread so much as builds the Page as a new one, using a handful of key points from the original one

[languid marsh]

Hi everyone, does anyone know if perplexity in response of a question via API request, output the result's sources (references) just like their search engine perplexity.ai?

[lofty quarry]

Hello. Does anyone know how to get ChatGPT 4 free with the Pro version of Perplexity?

[austere kestrel]

you pay for perplexity Pro, so it's not really free? (but it's the same monthly cost as an openai plus sub)

[lofty quarry]

Thanks. I know that part but am I supposed to be able to login to ChatGPT Plus with my subscription to Perplexity?

[austere kestrel]

no, perplexity and chatgpt are not connected at all

[lofty quarry]

Ah. Ok.

[lofty quarry]

Ah ok.

[austere kestrel]

if you have perplexity pro, you can select different models, including GPT-4

but the subscriptions are completely separate

[lofty quarry]

Perfect. Thanks for your help.

[torn ravine]

Yup, it seems one shouldn't have too high an expectation about the automatic conversion. Building from scratch guarantees better results.

[south kindle]

i believe there's a closed alpha going for citations via the API. haven't heard anything about it in a while though. So they are at least working on that and understand why it's important and needed.

[grave coyote]

So playing with Pages yesterday I really like it, but for anyone interested in getting pro for it here are the big limitations. I don't know when or if any of these will be addressed:

1. There is no export function. The pages lives on perplexity's website and you can send a link to others. There is no way to put it on your personal blog or anywhere else without significant work and reformatting.
2. You can't edit the page title. There is a work around where you can edit the title of a thread then convert that thread, but if you start with a page, it picks the title.
3. You can't edit the text. Want to clean up a few words or add a transition sentence, too bad. You can regenerate each section and play with the prompt, but you get what the AI decides you get.
4. There is no interaction with viewers. No likes, no comments. You can see number of views but even this you can't tell what was you viewing the page and what was others.

So yes you can make a nice looking page very quickly. You can share a link to it with others on social media and elsewhere. You cannot export it, add your voice to it, or interact with your audience at this time.

[south kindle]

I would drop this into #1140622008086970420 to hopefully get eyes on it quicker for you. your best option is having one of the PPLX team be chatting here and then bug them at that point. ha.

[south kindle]

regarding Editing text - you have to go into edit mode on the section, then highlight the text you want to change. i don't believe this feature was annouced yet but I know it works on my side, and a few others have confrimed they can do it as well. it's a bit janky, but it does work.

regarding your other points, very spot on. we have a fantastic Page output, but no real way for it to get views outside of manually sharing it with our circles of people. I know they are working on a "feed" for Pages based on the users likes/tags they follow. but no official word on that, just mentions here and there. so they know we either want to export the pages or have a way to drive engagement via the PPLX page. I think a Page Feed for published pages would work well.

[grave coyote]

I will have to give that a try. I saw that but I was thinking it was asking for a prompt to regenerate the section. So can you actually delete or change the text or does it have the AI change it from your "Custom Instructions"?

[south kindle]

If you highlight and just select the trashbin and arrow for done, it removes it. you can use the custom instructions just to specifiy what you want (given the highlighted section is small). I was able to add specific words or change parts of sentences with it. No undue button though, so if it deletes too much, your SOL and you gotta start that section over.

[barren mulch]

Any recommendations on "getting started" guides/videos for Perplexity AI? There are a lot on Youtube, but though I'd ask the community for some favorites.

[grave coyote]

It's a very intuitive interface. You could probably ask Perplexity to write you one. Then you could ask it questions.

[south kindle]

you could also hit us with a few of the queries you'd want to try out, and we can advise on focus modes/settings/collections. without knowing the usecase you have in mind, it's hard to really know what features to focus on. without Pro though, you should be able to get through most of it quickly.

[barren mulch]

Just starting out. Got pro already. Will play around a bit.

[south kindle]

my advice, Collections are very useful. use them to both organize threads AND force a custom prompt specific to the collection.

[grave coyote]

Free tier gets a few pro searches each day. Experiment with the non pro searches then flip the pro toggle when you know what you want an in depth answer on.

[south kindle]

The quick run down of "focus" modes. Reddit sources only reddit, youtube only youtube. Academic forces search to only use published research papers. Writing mode is just chat, no search. Very useful sometimes when you don't want search to be used.

[grave coyote]

Ah! Good for you. Unless the YouTube video is really new there is a major new feature that was just released called Pages we were just discussing, currently only available to Pro users.

[south kindle]

lastly, make sure to check out #1124060874785685554 and #⭐│starred for user created prompts/frameworks/tools/etc. we are a community of curious builders that enjoy collaborating and sharing. you'll find a lot of good system prompts and what not in there.

[grave coyote]

My experience is academic doesn't stick to academic and will include Wikipedia and also some downright looney sources. It also only tends to get access to abstracts of published papers. If you are doing something important, check the sources it finds, exclude ones you don't like, and retrieve the full papers that are relevant to your subject and upload them as attachments. It will work much better.

[warm cave]

You should report that in the bug section, i have not seen that, must be a new-ish bug

[placid coyote]

is my very short proofreading prompt breaking something in pplx or is this happening others too? "External system analysis" is not something I inputted, to me it looks like an agentic search (the pro toggle)

[south kindle]

that is wild. WTF?

I feel like I saw one other person post something similar yesterday. I was confused by what they were showing off, but it makes more sense now. Probably gotta get that into #1140622008086970420

i'm tempted to say use rewrite with a different model, but having a example of whatever the hell is happening is pretty much the main way to get it looked into and fixed. ha.

Heck, I wanna know what system prompt PPLX is amending to force multi-step reasoning. it's gotta be part of the backend API calling that is getting put into the models output.

[placid coyote]

it responded with corrected sentence, so I don't really need to use a rewrite. technically it fulfilled the prompt, but also outputted ton of (for an end user) irrelevant text 😅

[warm cave]

Interesting, wonder if it’s a glitch caused by them testing agent search

Context got applied wrong or something

[placid coyote]

I thought that pro anylsis was a separate generation (possibly by other model), so maybe I was wrong, because how would it leak into final response? it looks like gpt4o either decided to respond in wrong order (but that would mean they do it in one generation and pausing it or something?). or yeah, it integrated context from search in a very weird way

[austere kestrel]

what does it show when you unfold the Researching / Pro pane?

[placid coyote]

I think it understood no search is needed. probably shouldn't have used "pro" toggle for a proofreading of simple english in a first place

[south kindle]

woof. that's helpful. lol. yeah, i suspect it realized it didn't need more info and just moved to "wrapping up". but that somehow amended something it shouldn't have to the output.

out of curiosity, what was your query that lead to that? kind of curious to try it.

[austere kestrel]

based on the response (in the other screenshot) - it completely determined no further search (/"multi step reasoning") was required ha - i'm pretty that gpt-4o has just reguritaged something that one of the other models / agents involved in the new process output somewhere along the way

[warm cave]

Back when I tried you.com, sometimes I would get a telephone game type setup with their agent setup, where one step would just repeat its instructions and that would just pass down the line, haha

[placid coyote]

I think at times it does pretty good job at iterative searching. but then are times when I get asked 4 times in a row to do the work I asked it to do. exactly same step, like "refactor visualizeCircuit function". it has some room for improvement, but showing a lot of potential

[austere kestrel]

yeah i've had a couple of glitchy loops

but totally agree - lot of potential

i'm really liking it

[south kindle]

if you do a bug-report, throw that screenshot in there as well. I honestly think it's related to Pro search being on when it's not needed, and it just moves to wrapping up. Plus the new agent like version of Pro search.

[placid coyote]

well, I didn't want to share work stuff, but it is so generic I guess I can. it is a very simple prompt

proofread following text:

Added link to private NPM registry setup.

but I have filled that AI profile, so something from that could also be affecting it. I know my emoji prompt reliably destroys recommended follow ups at the bottom, but I don't think I ever saw it messing that reasoning/search. I don't think my ai profile is that strong

[austere kestrel]

well it's completely reproducible 😂

[south kindle]

well it's not your AI Profile at least. something with that query. wild.

[austere kestrel]

i think something is seeping out from the new Pro process

[south kindle]

i'm actually leaning towards something with the phrase "Proofread". if you search just the text string, pro search runs with it, and the reply uses the sources as you'd expect.

[limpid silo]

Hello!

[south kindle]

rephrased the text string. same thing.

[austere kestrel]

Yeah but this part:

The query does not require combining information from multiple sources sequentially and does not need multistep reasoning. No
That sounds so specific to the actual system likely working behind the scenes
would be weird if it was just a gratiutious hallucinatino

[placid coyote]

I know a long time ago, when Opus was "free", it rarely leaked similar process reports. but at that time gpt4t was working flawlessly (thousands of responses ok)

[south kindle]

scratchpad got invoked from my AI Profile when I added "and rephrase it for clarity and completeness" given it's in the Key Info area, it's somehow amended SOMETHING to the original query. what a wild AND repeatable bug.

It can't be a hallucination if we can repeat it exactly every time, from different users.

[placid coyote]

I didn't think lean to hallucinating even in my first occurrence. it was strangely reminiscent of how I would expect the agentic "pro" search/reasoning to work. the prompt was very short and my ai profile (+ pplx sys prompt) are quite short as well, plus it didn't do search. IMO you would have to have a very weird things stuffed in there or prompt to force into hallucinating these giant models

[south kindle]

last post on it for now. added an actual paragraph after the NPM registry string. we get more info in the Pro Search area.

The query does not require additional personalization, external knowledge, or multistep reasoning. It is a straightforward proofreading task.

[placid coyote]

so either application passes wrong things in context for the final generation or it is supposed to be passing it, but models somehow stopped understanding they shouldn't be including it in a final response? (I am not a pro, just interested in these things, so take my "analysis" with a grain fo salt)

[south kindle]

We are just users guessing, but you seem to be on the right path. you came across something weird that also seems to "look" behind the curtain a bit at whatever PPLX is doing.

okay, got the external system message on unrelated text. this is a new thread. Pro search on.

[south kindle]

and on that note, I got the pre-prompt PPLX uses (i think). it refers to Pro search as a "separate" system.
https://pastebin.com/DVJjSSqg - was too long to spam here.

[austere kestrel]

oh that's interesting - thanks!

yeah it is a separate system

[gentle dirge]

Interesting

[austere kestrel]

it's why the API isn't the same

[gentle dirge]

ive always wanted to know the sys prompt of pplx

[south kindle]

to be fair, it said NO initially until I used Rewrite / pro search with actual internet search off (with AI Profile set to OFF) so I doubt its meant to be leaked at all. my bad. maybe repeatable in writing mode using Pro search? Edit: Repeatable, but only if you use "rewrite" with Pro Search, within Writing mode.

[warm cave]

https://www.udio.com/songs/rFRxqdPi2XBxB77BHjdN3M

[south kindle]

has to be writing mode, no AI profile, Pro search on, and only via "rewrite".

[icy cloak]

For Pages, any way to create Pages from uploaded content? Such as PDFs or links

[south kindle]

If they are part of a thread first, maybe? I don't think so though yet. We can only add our own images currently. But if you run it through a thread to review it, ideally the information from the review gets included in the Page.

[fast meteor]

true, pplx now becomes not as good as earlier in handling documents. but i have a workaround

1. install xpdf tools: use pdftopng to convert pdf to images
2. (obtain api for gemini 1.5 flash) and run the following quick and dirty python script python ocr.py imgfolder/ output_name
3. ask pplx for parts unclear to you/in interest, this is what pplx is now very good at

"""
Install the Google AI Python SDK

$ pip install google-generativeai

See the getting started guide for more information:
https://ai.google.dev/gemini-api/docs/get-started/python
"""

import time
from pathlib import Path
import sys
import mimetypes

import google.generativeai as genai

# VISIT https://aistudio.google.com/app/apikey
genai.configure(api_key="__REPLACE_WITH_YOUR_KEY__")

def upload_to_gemini(path, mime_type=None):
  """Uploads the given file to Gemini.

  See https://ai.google.dev/gemini-api/docs/prompting_with_media
  """
  file = genai.upload_file(path, mime_type=mime_type)
  print(f"Uploaded file '{file.display_name}' as: {file.uri}")
  return file

def wait_for_files_active(files):
  """Waits for the given files to be active.

  Some files uploaded to the Gemini API need to be processed before they can be
  used as prompt inputs. The status can be seen by querying the file's "state"
  field.

  This implementation uses a simple blocking polling loop. Production code
  should probably employ a more sophisticated approach.
  """
  print("Waiting for file processing...")
  for name in (file.name for file in files):
    file = genai.get_file(name)
    while file.state.name == "PROCESSING":
      print(".", end="", flush=True)
      time.sleep(10)
      file = genai.get_file(name)
    if file.state.name != "ACTIVE":
      raise Exception(f"File {file.name} failed to process")
  print("...all files ready")
  print()

def main(args):
  # Create the model
  # See https://ai.google.dev/api/python/google/generativeai/GenerativeModel
  generation_config = {
    "temperature": 0.8,
    "top_p": 0.9,
    "max_output_tokens": 4096,
    "response_mime_type": "text/plain",
  }
  safety_settings = [
    {
      "category": "HARM_CATEGORY_HARASSMENT",
      "threshold": "BLOCK_NONE",
    },
    {
      "category": "HARM_CATEGORY_HATE_SPEECH",
      "threshold": "BLOCK_NONE",
    },
    {
      "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
      "threshold": "BLOCK_NONE",
    },
    {
      "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
      "threshold": "BLOCK_NONE",
    },
  ]

  model = genai.GenerativeModel(
    model_name="gemini-1.5-flash",
    safety_settings=safety_settings,
    generation_config=generation_config,
  )

  # Path to the folder containing images
  folder_dir = Path(args[0])
  
  files = []

  # Iterate over all image files in the folder
  for image in folder_dir.glob('*'):
    if image.suffix in [
        ".png", ".jpg", ".jpeg", ".gif", ".bmp", ".tiff", ".webp", ".svg", ".ico", 
        ".heif", ".heic", ".avci", ".avif", ".ai", ".psd", ".eps", ".raw", ".tif", ".jfif", ".jpe", ".jif", ".jfi"
    ]:
      files.append(upload_to_gemini(image, mime_type=mimetypes.guess_type(image.__str__())[0]))

  # Some files have a processing delay. Wait for them to be ready.
  wait_for_files_active(files)


  for i, file in enumerate(files):
    response = []
    _count = 0;
    while _count < 2:
      try:
        response.append(model.generate_content([file,"in extended markdown format with katex, describe images with texts. only focus on the contents"]).text)
        _count = 9999
      except:
        _count += 1
        time.sleep(20) # rate limited anyways, no needs for concurrency
    _count = 0;
    while _count < 2:
      try:
        response.append(model.generate_content([file,"provide intuition, insights, explanations, theories etc in details! If it contains questions: answer them!"]).text)
        _count = 9999
      except:
        _count += 1
        time.sleep(20) # rate limited anyways, no needs for concurrency
    with open(f"{args[0]}\\{args[1]}.md","a+", encoding='utf-8') as out:
      out.write(f"=== Slide {i+1} begins ===\n"+response[0]+"\n\n========\n\n"+response[1]+f"\n=== Slide {i+1} ends ===\n\n")

if __name__ == "__main__":
    args = sys.argv[1:]
    main(args)

[fading moth]

Interesting

[fast meteor]

you can choose whatever the cheapest from haiku/gemini 1.5 flash/gemini pro 1.0 vision

[narrow silo]

@fast meteor hello

[fast meteor]

https://nohello.net/

[narrow silo]

😄

[south kindle]

#⚡│ask-community message > there’s a sign up link here.

[narrow silo]

how does ai work?

[fast meteor]

black magic🪄 🪄 🪄

[narrow silo]

is AI something that substitutes writing code?

[warm cave]

https://youtu.be/wjZofJX0v4M?feature=shared

[narrow silo]

i wonder like, if i have program and that program has some script list, can AI replace that somehow or what?

[south kindle]

Also a fun question to actually send to Perplexity. Explain AI like I’m 5.

[warm cave]

depends on what is dose

[narrow silo]

it replaces badly written words with correct ones

[warm cave]

https://www.perplexity.ai/search/Explain-LLMs-like-9GNPk0kHSTywyNeW.sd6hw#0

[warm cave]

Definitely sounds like something AI can do

[narrow silo]

sounds like having Sheldon Cooper for a friend

[south kindle]

It’ll do that even without code. Just use the prompt : review and update this to be more formal and professional.

[warm cave]

talk to it like a huamn and ask it to complete tasks (Of course tasks that i can do over text, lol) you will quickly learn what it is good and bad at, and how it can help you on a daily basis

a lot of people talk to it like its google, dont use key words, use sentences, etc

[narrow silo]

is google not AI

[south kindle]

Negative. It’s a search engine

They have AI features. But they are behind the others.

[warm cave]

google uses AI but its a different type of AI, we are talking about LLM's (Large Language Models) the feild of AI is large and divers with many uses.

LLM's are AI models that you can communicate in natural language

[fast meteor]

one day we only need to define the contract for dev & testing, the code would be ai-generated

[south kindle]

Example. Ask it to build Pong and export the code. You’ll get the entire codebase for a simple Pong game you can actually run on your desktop.

[warm cave]

https://www.perplexity.ai/search/Build-the-game-PEYAkY7lQESZ1N7dr0WXSw#1

[warm cave]

natural language is the new high level coding language, haha

[indigo zealot]

my perp is tweaking

[warm cave]

Bro, what did you do to it, lol🤣

[indigo zealot]

it's like a long thread i've had for months lol

probably starting to lose it

[fast meteor]

the new meta

[warm cave]

You can probably delete the last handful of messages, that might fix it, but idk

[warm cave]

What is your favorite IDE? Me: chatGPT 🗿

[south kindle]

How long out of curiosity? Like more than 30 queries?

[indigo zealot]

oh yeah probably a lil more than that

the scrollbar is pretty small

[warm cave]

You need this, haha

[indigo zealot]

what's that UI

it's a suggestion?

[warm cave]

#😎│cool-projects message

[south kindle]

3rd party app released today in beta for PPLX users

[indigo zealot]

ahh perfect timing

[south kindle]

Once you get it installed. The TOC toggle is in the app itself (click the extension icon) I still wanna know the query count. Ha.

[indigo zealot]

I don't know how to check the query count lol

[south kindle]

You’ll see the long list on the right side. Just count them.

[warm cave]

It wont give you the query count, but it would at least be easier to count with the table of contents

[south kindle]

TOC is just table of contents.

[indigo zealot]

it's still in closed beta? can't find it

[south kindle]

Bounced the screenshot over to chatgpt. Affirmations.

[warm cave]

https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w

[south kindle]

You gotta DM @harsh stag with your email so he grants you access. It’s a closed beta.

[warm cave]

There is a link for the discord at the end of the page i sent

[indigo zealot]

chatgpt 4o is OP

[tame current]

but only 10 messages

per 4 hours

so

[warm cave]

a taste test

[indigo zealot]

ehh enough for me

[warm cave]

Thats good, it would be too little for me, but i am glad they have 10 for free

[indigo zealot]

I use the 10 free messages twice a day

I don't really mind waiting

[warm cave]

Im taking summer classes rn, so i use a lot of uses having it tutor me through the questions i get stuck on

[indigo zealot]

that's useful

I wonder how many people rely on AI to function nowadays

I definitly do

[fast meteor]

why dont you just ask the tutor? it is free xd

[warm cave]

Yeah, well i definitely would not want to go back to how it was when i started College, having a personal tutor is OP

[south kindle]

Probably less than you think. The general public is still confused by what it even does and is just annoyed by seeing it everywhere.

We live in an AI bubble of enthusiasts. Ha. So I’d say the vast majority of us depend on it now.

[warm cave]

Yeah we are in an AI bubble, most people have never tried it

[indigo zealot]

that's crazy

been using chatgpt since like day 1

[south kindle]

Not throwing shade but above is a new user, new to tech, new to AI, and thought Google was just an AI tool. But he didn’t know what AI actually was, and settled on “Sheldon cooper”.

[sleek vortex]

fr been looking at gpt since gpt2

i remember getting codex api access with my 18$ in credit

[indigo zealot]

@south kindle I decided to count each div to see how many queries I did on that thread, try to guess lol

[sleek vortex]

somebody had a 540 query thread

[indigo zealot]

wtf

[sleek vortex]

so idk if ur gonna be beating that

on pplx free asw lmao

[indigo zealot]

I have to restart my threads after like 30 because it's starts tweaking out

[warm cave]

You were to the game way earier than me, I only tried GPT-2 after GPT-3.5 was out. But my first time using an LLM was GPT-3 week 1 with chatGPT

[indigo zealot]

I remember when chatgpt came out I was on my internship automating stuff thinking it was gonna change my life lol

[warm cave]

I used up all of my usage talking to it at midnight lol, my mind was blown

[indigo zealot]

yeah first time using chatgpt hit different

[south kindle]

75?

[indigo zealot]

nah not that much, it was 54

[sleek vortex]

gpt2 was like oh cool i can generate a story on some random website

[indigo zealot]

but i've had to restart this thread like 5 times by now

[south kindle]

Damn. Was gonna guess 50 initially and went higher. Nice.

[grave coyote]

Yeah I just took a moment to try a new thread with Academic focus. It found 4 sources:
Wikipedia
The BBC
Britannica
and another Wikipedia

Only Britannica would I say qualified as published articles and honestly I would not consider an encyclopedia to be a great source. I think they don't really have access to published academic papers, so that focus is a bit of a scam. I will pop it over as a bug and see what they say.

https://www.perplexity.ai/search/I-want-to-ZoV4zN4LRKa2YcbFG52K.Q

[warm cave]

Ohh, its the new agentic search that is causing the problem, or it the old bug that just goes it a search loop. Either way its not normal

Looks more like the old bug actually, but idk, have not gotten the chance to try out agentic search

[shut plaza]

lmfao

[foggy tapir]

Was gpt4o removed?

[warm cave]

Don't think so, why do you ask?

[foggy tapir]

Well, it's not visible when I try to change model in settings

[sleek vortex]

it is for me

[sleek vortex]

is this some old version of the android app

[foggy tapir]

I wasn't able to view it on web as well a few hours ago.

[foggy tapir]

I will try updating

[grave coyote]

I have heard people mentioning a quota on their use of Opus, particularly in relation to Pages draining this. Where are you able to see that?

[sleek vortex]

Network tab in dev tools

Most of the requests include some hydrated state property with your amounts left

[grave coyote]

Well they really hid that. Would be nice if they were a bit more transparent about that.

[south kindle]

Wait till you realize anyone with know how can find your name, email, subscription status, and other specifics to your account, if you share a thread publicly.

[grave coyote]

Yeah I already noticed that. Not terribly worried in general, but it's kind of sloppy.

All these AI companies operate like first year startups...because they are.

[south kindle]

A painfully accurate statement. Ha.

[smoky spade]

Is there now a daily limit on Claude 3 opus? I cannot use the rewrite feature with opus selected by it works for sonnet and gpt.

[sleek vortex]

This worries me too as someone who’s trying to make a good product myself

[sleek vortex]

Have you used pages by any chance

[thorny canopy]

When this 'soon' fix for opus gonna happen actually? 👀

[smoky spade]

no

[south kindle]

when is the OPUS 50 discord reaction/emote coming

[sour bough]

hey guys, does anyone know whats the model behind the free default model in perplexity?

when i ask it it says that its from anthropic

so i assume its a modified claude 3 haiku? (unless its just a hallucination)

[smoky spade]

i need claude3 to write exams and answer sheets and mark them.

teacher's aid

[warm cave]

Are you asking for advice to get good results or just letting us know about your plan?🤔

[smoky spade]

I need claude to mark and write exams and answer sheets.

[warm cave]

Oh, I did not see your message above

Yeha its 50 per day

What topic is it, if its stem then you might find GPT4o more accurate

[warm cave]

Yes, a mod confirmed that it is haiku

[smoky spade]

law, filipino

[warm cave]

GPT4o and Opus should both do good imo, but give it lots of course material and example problem if you can, it will improve its accuracy

[granite kelp]

hi everyone! is there a channel I can post in for support type questions?

[agile jay]

#👉│feedback-general
#1140622008086970420

Probably one of those.

[granite kelp]

thank you!

[umbral pollen]

What good APIs other than Perplexity have internet access?

[agile jay]

I don't know of a public API that offers a similar experience.

Your best bet is probably making your own.

Using searXNG, for making searches, and then a scraper framework to make http/headless requests, and then preprocess the html to markdown before passing it to your model.

[fading moth]

https://discord.com/channels/1047197230748151888/1244758414794428516 Don't forget to post your Opus 50 art here! :3

[rigid drum]

whats happened to the referals? have they been removed?

[teal bluff]

As a newbie, what is this? Is it art we generate in perplexity?

[fading moth]

Uhh, it's... How would this be best phrased... Opus used to be 600 messages a day and now it's 50. As a result, a little meme started involving random AI generated pictures with the words "Opus 50" in them.

Like that

But the genre and style vary wildly

I wouldn't say it's protest, but it's not a meaningless meme either

[fading moth]

I made this one tdoay, and it's just too awesome not to share it. I make 2-3 a day :p

Michael Bay felt a disturbance in the universe when I generated that.

[kindred vessel]

Is there a way to get perplexity to summarize a youtbe video? Seems odd that for everything it can do, it still requires you to pull a transcript to summarize.

[half venture]

Is chatgpt down

[stable radish]

Yup

[half venture]

Billion dollar company

[stable radish]

You’d think

That’s why I’m laughing when people are talking about AGI, GPT-5 and everything else “coming soon”

[quartz totem]

I am pretty new to Perplexity and pretty impressed by most things it does good.
One thing I do not understand is: AFAIK the "Pro" switch never makes sense with anything except "All" mode, i.e., web search. Why the heck do they keep the switch displayed when we switch to another mode? I mean, that is 1 minute coding time to put a little if(mode == ALL) show() else hide() into their code.
Is there any reason to use "Pro" mode at any point in time when using anything else than "All", especially "Writing" (have not really found that much use for the other modes yet except for probably Reddit)?

[tame current]

Subliminar message.

What do you think about Pages function?

[south kindle]

https://tenor.com/view/larry-david-seinfeld-pretty-good-enthusiasm-curb-gif-3938316

[fathom pollen]

how do i enable the perplexity enhanced + as in the above image?

is this browser specfic??

[quartz totem]

@fathom pollen It is this Chrome extension.

[pine valve]

My account briefly gained access to the new Pro Search feature, but now it's no longer available. I think the new Pro Search was quite interesting.

[tidal sail]

Is there a way to reply to this ai chatbots while am interacting with them?

[south kindle]

#😎│cool-projects message - Chrome based browsers only.

[obtuse pagoda]

is there any way to enter a beta program to get access to the new PRO search, for example?

as I understand it, its activation and deactivation are related to A/B testing or something like that

i see, i'm already pro user

[harsh stag]

hey guys, is perplexity down?

[pine valve]

so slow

[harsh stag]

i got stuck here

[pine valve]

metoo

[rocky moss]

Something is happening. Chat GPT having issues to.

[thick tulip]

Me too, stucked after asking.
On laptop and smartphone.

[granite grail]

same here, tried changing to claude opus as default and doesn't work either

[dawn lantern]

Me three, have no idea what's happening

[thick tulip]

Just with pro search. The normal one is working.

[granite grail]

ok they're aware

[dawn lantern]

Probably little boi at full capacity and went bruh

[thick tulip]

Ddos attack or underestimation of future requests?

[swift terrace]

is perplexity down or nah?

[dawn lantern]

Happened multiple times, hope not so

[proven crest]

I see I'm not the only one

[solar bridge]

down?

[dawn lantern]

Yup

[dense goblet]

Perprlexity off?

[solar bridge]

not on status page?

[thick tulip]

Back to stackoverflow🤣

[rustic gate]

Ouch. Couldn’t happen at a worse time

[covert maple]

What happened?

[thick tulip]

Still no response from the official team

[warm cave]

We are the status page, the channel only updates after 20 min of is complaining

[shadow ginkgo]

Stone age 🪨🔨

[opaque stump]

mmm

capacity huh

[covert maple]

😂

[granite grail]

sorry, after so long making dumb questions to AI, I couldn't handle being lambasted by the community there

[dense goblet]

I have a paid 1-year code from Perplexity. Can I sell here? 50$

[rocky moss]

Looks like all the AI tools are having issues.

[granite grail]

Nvidia used backdoor to bring all AI servers down

[warm cave]

Wow, it's is the end of the world

[granite grail]

they hit the nuke button

[dawn lantern]

Holy 🦆 gpt literally went offline?!

[molten yoke]

Same issues here. I'm not so sure about first time Claude is down. I've had plenty of issues with Claude.

[undone frost]

I just signed up to Pro and found all the AIs are no out of service. Typical.

[warm cave]

So all 3 of my subs are down, might as well give up

[livid aurora]

https://tenor.com/view/server-is-fine-burn-fire-spongebob-blow-gif-14178373

[shadow ginkgo]

What is this page? 🤔

[warm cave]

We are under attack 😟

[livid aurora]

https://downforeveryoneorjustme.com/perplexity

[vernal remnant]

can you share the link of this?

[atomic mulch]

any update from team regarding with the current situation

[livid aurora]

there are tons of others sites

[coarse nimbus]

How am I supposed to name my variables now? "varvar" is good, right?

[austere kestrel]

is it still down? seems to be working here

[livid aurora]

still down

[peak yoke]

Well the labs is down

[rapid pebble]

https://downdetector.com/

random things seem to be exploding rn

lmao

[peak yoke]

It's down

Do you have a dark theme extension?

Or something like that

Well not sure why it's all white then

[calm compass]

Isn’t working for me

[warm cave]

Might want to update status channel @halcyon coral @serene sand @wraith crow

[maiden swan]

Is this possibly a cloud service provider issue? (AWS?)

[peak yoke]

Well people overused today probably

Lmao

Lemme check pi

[normal heron]

Dang I thought only GPT is down

[half escarp]

yea GPT and Perplex down for me

[halcyon coral]

Yep, thx, done

[zinc shale]

wow AI got too smart and somebody finally pull the plug?

[molten yoke]

AWS does appear to have a spike this AM in issues...

[dawn lantern]

Any one checked Copilot?

[mighty topaz]

the end is coming

[calm compass]

Midjourney still working so not quite lol

[maiden swan]

all my api requests are going through promptly (gpt, pplx)

[peak yoke]

So all services supported by AWS are down then

[dawn lantern]

Guys little boi going good again

[nocturne crown]

ChatGPT not working, copilot very slow

[livid aurora]

6 Hours ago it was down

[molten yoke]

Which I think includes OpenAI, Perplexity, and Claude. Copilot is in Azure

[sullen ether]

Works for me

[normal heron]

Has anyone checked Llama?

[peak yoke]

Perplexity is back

✨

[wide mesa]

Its telling me I need a pro subscription when I try to use ChatGTP 4o

[wide mesa]

even though I have one

So it seems like its just using the default pro search atm

[molten yoke]

I think the issue was at least partially AWS since Copilot (on Azure) and Gemini (on GCP) were fine.

[upbeat island]

forced to use gemini 🤮

[vernal remnant]

guys does anyone know why perplexity goes down with chatgpt? i mean why doesn’t it work with other models?

[peak yoke]

@vernal remnant

[vernal remnant]

thanks

[maiden swan]

I suspect they are using similar underlying cloud services which are experiencing outages / issues

[peak yoke]

Didn't openai get Azure credits to run their models?

[wide mesa]

in short, this isnt on Perplexity

[maiden swan]

I'm using that chat completions api and it did not go down, so maybe on the front end they are using aws etc

[atomic mulch]

is it working again for anybody? I still have the same issue

[vernal remnant]

yeah me too, doesn’t work at all

[rustic stream]

#🔔│status message
me too😭

[livid aurora]

Its down again

[solar bridge]

not working at all.

[harsh stag]

if opus limit gets increased by 10 every time perplexity's down, we would get 600 again by the end of the year

[rustic gate]

Is anyone else’s entire collection missing?

[pliant ridge]

wow

wtf

[sick hill]

Any update from pplx on the infra status?

[worldly summit]

not working

[silver sinew]

Cute sticker

[livid aurora]

@halcyon coral It's down again

[stray tiger]

So openai down = perplexity down?

[fossil sentinel]

Seems like copilot still works but super slow

[fathom pollen]

this is odd.. never got this before

[rustic stream]

I see.

[fossil sentinel]

There's a major outage ongoing

[solar bridge]

chatgpt is also down for me!

[fossil sentinel]

AI outage in general

[solar bridge]

https://status.openai.com/

China, Russia, North Korea or Iran????

[atomic mulch]

somehow i get a reply for "hi" prompt but when i gave a proper prompt it has an issue

[harsh stag]

same

[visual barn]

But they don't have to use openai, so why are they down?

[lyric hill]

so I just subscribed to perplexity and it seems it's in a very bad time 😂
I wanted to know if the claude 3 opus has the same context length it has in the claude webpage

[warm cave]

No, Perplexity has 32k and Claude Pro has 200k

But keep in mind the more context you use on Claude Pro, the after it runs out of usage

[visual barn]

https://status.anthropic.com/ also Partial Outage

[past zealot]

@halcyon coral hello there. I am sorry to bother you. But how does one apply to become a beta tester on iOS. I couldn’t find the link. Thank you

[supple sleet]

Gemini down too? What happened?

[lyric hill]

ok that makes sense thank you

[halcyon coral]

Hey @past zealot! The beta is closed, we'll announce when we need more users. Thank you 🙏

[agile jay]

Oh, why so much chaos? I'm assuming cloudflare went down, or something.

[visual barn]

Nope: https://www.cloudflarestatus.com/

Damn, i need to speak to a bot, ppl are annoying xD

[foggy pecan]

Bro acting like a crackhead

[agile jay]

Oh, then which service is down that has taken down all these sites?

[foggy pecan]

gib me ai 🤣

[agile jay]

Normally it's been cloudflare, but I guess it could also be as simple as a random AWS service being down.

[foggy pecan]

maybe one of the nvidea servers

[visual barn]

Some cleaning lady pulled a plug to put her vaccum in and now AI is gone until she has finished this office -.-

[winged moon]

How do I think without AI?

[visual barn]

Thinking is overrated, just do ... something!

[foggy pecan]

go and read some tweets lots of thinkers are there

[livid aurora]

Gemini now down too

[winged moon]

No. I don't want to actively lower my IQ

[foggy pecan]

aye yo league too?

Bro is this because of the hacker thingy

[visual barn]

Where you got that stat from?

[livid aurora]

https://downforeveryoneorjustme.com/perplexity

[winged moon]

Damn. How are we supposed to keep those salty neckbeards away now?

[regal nest]

is perplexity still down?

[foggy pecan]

https://www.youtube.com/watch?v=OkJnY8bpXAM&t=744s&pp=ygUaaGFja2VycyBzb21lb3JkaW5hcnlnYW1lcnM%3D

[thorny rain]

it does appear to still be down even though status says its good

[quiet plover]

how is everything down at once

[regal nest]

yes, i can not use that

[pastel karma]

Confirmed, ChatGPT and Copilot down as well

[quiet plover]

GPT, Perplexity, Claude, Copilot

[winged moon]

At least Llama 3 isn't down 🙂

[copper willow]

Nothing like a paid for service

[junior adder]

i lost so many messages on opus thanks to this screen bruh

[south aspen]

Can you imagine a paid service going down temporarily? It's never happened in the history of earth

[agile jay]

Obviously we haven't paid enough, which is why it's down.

[graceful grove]

I beleive lots of services are out. Gemini, ChatGPT, Prelplexity. Is this some kind of an attack from Anti-AI hackers?

[tribal ridge]

wtf

[thorny rain]

trying to fact check this statement, but perplexity is down

[fossil sentinel]

Widespread AI outage yippie

[agile jay]

No it's a service that they rely on that is down.

Since a lot of sites are currently down.

[graceful grove]

What about Grok?

[thorny rain]

how would multiple services go down at once if not for a coordinated attack?

[agile jay]

No idea, I only use Groq.

[thorny rain]

and don't say solar flares

[craggy current]

If they have dependencies upstream that would also explain it

[agile jay]

Just a random other service going down.

[lyric hill]

it might be some service they all use, league of legends is out too

[thorny rain]

aws outage?

[graceful grove]

What service is common between all of them?

AWS?

[thorny rain]

ah, copilot would be azure

[agile jay]

It could be a specific AWS service.

[craggy current]

It's dns lol

[lyric hill]

openai might use azure since they are so close to microsoft

[normal heron]

App still doesn't work

[cinder leaf]

the same for me 😭

[spice zodiac]

gemini working

[graceful grove]

May be the servers near my region aren't working...Hmmm

[foggy pecan]

They gonna touch grass staph dem

[wary trout]

Gemini working

[austere kestrel]

eh? who's who there

[thorny rain]

perplexity is back

[sleek vortex]

coincidence that a combo of all these have gone down?

aws bedrock
gemini
azure openai
anthropics own api

[wary trout]

Yep

[chilly helm]

i'm seeing the normal Perplexity loading page now, about to see if it works

[wary trout]

Cloudflare again probably

[foggy pecan]

waiting for discord servers to go down as well

[sleek vortex]

why would aws or azure use cloudflare

[lyric hill]

perplexity is working now

I can use it

[sleek vortex]

they're their own big clouds

[tribal ridge]

hi welcome back

[wary trout]

No idea. Is what perplex web uses

[austere kestrel]

safe to say not a cooincidence i think ha

[foggy pecan]

DRG servers down as well

we aint rock and stone bois

[thorny rain]

lol

[cinder leaf]

Cool, That's it for me too

[chilly helm]

it's working for me

[solemn cedar]

the question is what's related between all those websites ? It's weird really

[chilly helm]

it's a bit slow still, but it's working

[stoic wave]

nah, chatgpt ain't working

[sleek vortex]

openai api is still up

i think possibly a PTU outage

wouldnt explain gemini tho...

[wary trout]

My guess

[sleek vortex]

gemini

doesnt use cloudflare

aws and azure

dont use cloudflare

[wary trout]

shodan it

[sleek vortex]

openai, uses cloudflare
but api and chatgpt.com is still up

[stoic wave]

Is this Shodan?

[sleek vortex]

only their backend inference for chatgpt.com is down

[wary trout]

yes

[wide mesa]

The ZeroGPT site seems to be really slow as well

[sleek vortex]

api latency was really high for me

on openai dashboard rn

[austere kestrel]

dunno but something right? i mean, it's an insane coocidence otherwise

[wide mesa]

They have the same cloud hosting service

(AWS)

[cinder leaf]

nice 😭

[rocky rose]

lol that's pretty funny

[woven cosmos]

Use the application on macos, it currently works well

[pine valve]

down again😢

[wide stratus]

Wish all the best to Prplxty team to get the service back working again ASAP 🙏

[junior adder]

I refreshed my page every ten minutes to get the update on this whole situation, and this is the unfunny result i've got?

[opaque stump]

Well that’s just insane to disallow VPNs

[mortal eagle]

died again what....chat GPT also not working

[junior adder]

I don't use any. That's the funny

[hardy willow]

Perplexity, OpenAI, and Anthropic Claude are ALL down simultaneously right now as I type this.

[winged moon]

I was happy when I had like 20 minutes of usage

[mortal eagle]

seems like its time for gemini to shine

[royal quail]

wow it's been down for quite awhile now, whats happening

[winged moon]

Time to use Llama 3

[dark ruin]

why perplexity can't use?

[hardy willow]

No idea, it's a separate third party LLM platform... and there isn't anything about it in the Announcements thread here on their Discord

[vagrant frigate]

Well, time to switch to GPT. Useless

[hardy willow]

Anthropic (Claude) just came back online... but twitter and various news sites have started to report these bizarre synchronized outages. AI is taking over...

[dark ruin]

Nooooooooooo:(

[hardy willow]

How odd is it that 2 of the largest AI companies on earth are BOTH down................

[vagrant frigate]

Time to switch to Yahoo

[dark ruin]

gpt can use too:(

[hardy willow]

Time to switch to Speak-n-Spell

[wide stratus]

chatGPT? You mean on chat.openai.com ?

[wide stratus]

WHAT 😳 The AI has started to fight back!!! 😱

https://tenor.com/view/skinet-skinet-remember-maquinas-revolucion-gif-7115195292076587720

That is crazy

[hardy willow]

https://poe.com/login and https://copilot.microsoft.com/ are the only viable alternatives

[fossil sentinel]

Chatgpt back up now for me

[wide stratus]

Ah, so Skynet didn't manage to take down those 2 yet, nice.

[hardy willow]

Same

[fossil sentinel]

Nvm responses don't work

[wide stratus]

Yahoo has a LLM?

[fossil sentinel]

Oh it does

Just high demand

[hardy willow]

I was using Ask Jeeves in the meantime......... LOLOLOLOLOL

[devout geyser]

The AI models protesting or something today? 😂

Feels like they've gone on strike

[wide stratus]

No, it's just Skynet... #💬│general message

[devout geyser]

😂

[wide stratus]

https://tenor.com/view/robot-tech-skynet-ai-arnold-gif-13186419

[regal nest]

is perplexity down again?

[chilly helm]

down again >.<

[cinder leaf]

nooo down again

[wide stratus]

https://tenor.com/view/terminator-terminator-robot-looking-flex-cool-robot-gif-978532213316794273

[severe prairie]

Claude became self aware. Only the Golden Gate Bridge will survive.

[wide stratus]

https://tenor.com/view/terminator-smiling-arnold-schwarzenegger-gif-16197002

https://tenor.com/view/down-town-la-bomb-dtlablowup-gif-21604415
Hopefully Perplexity team will be able to avoid the LLMs from doing this... 🤭

back online in germany and US it seems.

Humanity has won again!!! 🥳

https://tenor.com/view/kristanna-loken-tx-melt-down-terminator-rise-of-the-machines-gif-24408281

https://tenor.com/view/terminator-gif-21649154

[spiral mesa]

How is perplexity AI premium better than Chatgpt apart from being cheaper and annual membership?

[tame current]

It has superior web search, completely nukes ChatGPT web search in every aspect. On other hand, GPT has plugins. Decide what you need more, functionality or access to new informations

[cinder leaf]

It depends on your usage:

• ChatGPT Plus allows the creation of files and the upload of more files than Perplexity thanks to its Python interpreter.
• Perplexity is better for information retrieval; it allows the use of multiple AI models depending on the usage and is especially less limited in the number of prompts per day compared to ChatGPT Plus.

Personal opinion: Having tested both recently, I find ChatGPT Plus very practical for programming with the "Continue Generating" button and its Python interpreter, but it is too limited in its usage. The limits imposed by OpenAI are very restrictive and can be dynamic, which bothered me. This means that in case of high demand, OpenAI can lower your limit. This happened to me once when I hadn't used ChatGPT 4 at all during the day, but as soon as I entered my first prompt and after ChatGPT's response, a pop-up appeared indicating that I had reached the limit even though I had only sent one message to GPT 4. However, this is rare. I had to wait 3 hours to use it again. 🤬

[hardy willow]

I've tried both, extensively, and truly, OpenAI by itself is reeeeally bad, without the robust internet scouring of Perplexity. It's really bad, comparatively, I mean.

[warm cave]

Also there is more model selection here, so you can use Claude and etc. So more pros and cons. But the new voice stuff will be on chatGPT first

[devout kestrel]

also I think perplexity hallucinations are more rare does't make stuff as much cause it has internet access

[warm cave]

If you can try both, you can get your first month for $10 at perplexity, see if you like it or not

If you like it keep it, if you like chatGPT more, go with that

[cinder leaf]

And now ChatGPT is down, lol, but Perplexity is on.

[warm cave]

Hard to say what pros and cons will matter more for you, until youn live with them

[spiral mesa]

Hmm. Perplexity we can use for $10 for a few months if we have multiple email ids

[devout kestrel]

bro just admitted it

[cinder leaf]

And with Perplexity, you have $5 free to use the API (if you are a developer).

[spiral mesa]

I am mostly into coding using Python Azure Databricks and Azure Data Factory right now I am using chatgpt specific plugins for adf,adb and python

[warm cave]

That is nice, I have been using it with msty for RAG

Local rag I should say

[wide mesa]

Im shocked with one of ChatGPT's custom gpt

[warm cave]

Why is that

You discovered the power of system prompts?

Or some other feature?

[wide mesa]

Decided to use a "Humanize AI" custom GPT, ran a fully AI generated essay through it and said essay is basically undetectable by AI detector sites

[fading moth]

What I miss

[wide mesa]

Thats all without adding ANY prompt aside from "make it sound academic"

[warm cave]

Until then humanize ai detector comes out lol

[fading moth]

Well, I'd run it through actual checkers before taking GPT's word for it

GPT lies more often than I'd like

[wide mesa]

I didnt use GPT as a detector lol

Thats silly

[fading moth]

Confused

But okie

[warm cave]

Humanize ai made a paper that is supposedly undetectable, and them it was likely put in some online AI detector and said it is human or something

I don't trust that stuff but idk, could be hard to detect

[wide mesa]

Did not mean to say it was fully undetectable

[warm cave]

No, humanize says that, on the web site

100% success rate

[wide mesa]

Just that it passed through some popular AI detector sites

[fading moth]

Oki

[warm cave]

If it works for you great, I just don't trust it, its an arms race between detectors and writers, and the particular one a school uses may be better than the online one

[cinder leaf]

Online detectors are not great. I once used Microsoft's Copilot to generate a text, and at the beginning, it said, "Hi, this is Copilot." I copied it like that into several online checkers, and they all said less than 20% AI. 😂

[lilac tundra]

Hello, I have a question. When I use the Perplexity API, the response I get seems to be generated using the standard Perplexity model, not the Pro version. I cannot specify in make.com that I want the responses to be generated using the Pro version. How can I solve this problem?

[hardy willow]

The voice-to-voice interaction, along with vision, if it works and isn't a parlor trick gimmick that becomes tired after 3 days of usage, will be game changing

[spiral mesa]

Why is Chatgpt not offering annual subscription? Any specific reason?

[warm cave]

With pplx api you get the llama models

[warm cave]

Yeah, I hope its amazing

[lilac tundra]

so there is no solution?

[wide mesa]

Schools that use Turnitin arent too common from what ive seen

for now

[warm cave]

So I guess technicaly the llama model is a pro model, are you wanting GPT4o or something?

You get the option between 4 models, llama chat small and large and llama search small and large

[lilac tundra]

The problem is that I am automating a research process, and when I use the Perplexity interface to research a company, the model's response is very good and detailed. However, when I perform the same procedure using the API (automation in make.com), the response is poor and simplistic, as if it is not using the Pro model.

[warm cave]

Yeah, the pro search is better on the web, you can't get the same quality over a pi afaik

Web Pro gives 20 sources and uses a model like GPT4o

I bet the api uses less than 5, but idk

I just using the chat variant

[lilac tundra]

This is disappointing, as without the ability to implement it with automations, the use cases for Perplexity are significantly reduced. Tjhanks for youre response!

[warm cave]

No problem!

[fading moth]

Hmm

So, when I turn on Pro... Does it keep Opus on or does it switch to the perplexity llama?

[warm cave]

No, it uses Opus or whatever mode you have selected

I was just talking about api

[fading moth]

Mmkay

[severe prairie]

My understanding is that perplexity does the search and then whichever ai you have selected writes the response. Is that correct?

[fading moth]

It noticed that if I tell the pro search to like... Only search "derp.com" it doesn't exclusively search derp.com

Is this a common issue?

[warm cave]

I don't know how there web retrieval/rag works, but at the end the sources are given to the AI modle you chose, and that is what gives the final answer

[fading moth]

Also, this msty thing, daily... What's the dealio with it?

[warm cave]

Its an app I like, I saw it on the Ollama github under community projects, you can run local models or you can use API’s, I like to use my free $5 api credit in msty. I can lnk local files on my computer and it will use those as sources

https://msty.app/

[severe prairie]

I know you can remove sources using the three dots menu below each response. But I don't know if there's a way to focus a specific page.

[warm cave]

There use to be a way, but it seems to be broken now

You use to be able to put site:example.com

But I can't ever get it to work now days

[fading moth]

Hmm, good to know ty yall

[warm cave]

I'm pretty sure Alex is the one that shared the feature a while back

[fading moth]

But like... I couldn't link it to perplexity's (not their fine-tuned Llama3) 4o or Opus?

[warm cave]

No, perplexity only gives you llama over api, there are other options, but those use compute power or cost money over api

[fading moth]

[fading moth]

I'm a cheapskate and want a way to get as much Opus as I can and as optimized Opus as I can

[warm cave]

Opus 50 😭

[sleek vortex]

lol

[sleek vortex]

do you think i should just release "fast llm readable web scraping/data" as an api

[fading moth]

I've yet to find a service that gives me more than 50 a day

[sleek vortex]

opus?

[fading moth]

Ye

[sleek vortex]

well yeah opus is bankruptcy expensive

[fading moth]

I know Q_Q

[sleek vortex]

opus isnt even the like best

only for a few things

4o and gemini are still really good imo

idk

[fading moth]

I do creative writing for almost everything in my LLM stuff

Opus does better than 4/4o imo for creative writing

[warm cave]

Maybe, that dounds useful

[fading moth]

Gemini Ultra 1.0 was my jam, but then google kinda abandoned it

[warm cave]

Btw, have you tried Gemini 1.5 in the developer console?

[fading moth]

Yeah, but it's missing that spark Ultra 1.0 had... Google even admitted to it

[warm cave]

I guess there is nothing to do but remember the days of opus 600

[fading moth]

Tis true, daily, tis true

And it isn't even full Opus

But I'll take it all the same

[torn sierra]

This is super lame; we've been messing around with Opus 50 for two months now without any info on whether Opus 600 will come back or not. The lack of communication from Perplexity feels like they don't care and makes it seem like the abuse claim was just a bluff

[sleek vortex]

batch scraping + parsing with my tool at least is insanely fast

[warm cave]

I am guessing that abuse was a problem, but it was more of an excuse

[sleek vortex]

99% was a bluff

they realised in like 3 days theyd go broke

and then opus 50...

[torn sierra]

I understand that it might have been too expensive, and the 600 was only available for a limited time. However, such information should be communicated directly

[warm cave]

Yeah, handled in cowardly way, made a small problem out to be the reason, and then ignored everyone

[fading moth]

I don't want to get banned, but my suspicion is that when the limit drop happened it was a legitimate attack... But it was what perplexity needed to drop the limit without being blamed for a business decision

[warm cave]

Is that 53 sources???

[sleek vortex]

if they said opus 50 from the start people already wouldnt care as much

[sleek vortex]

53 links scraped and parsed into llm readable data in 4 seconds

meanwhile other services take 4 seconds for one link

so um

[warm cave]

Dang, that is fast

[sleek vortex]

might have a usp at my hands here?

[sleek vortex]

thats all hot, no cache

from cache its like 16ms

well disk cache at least

dont exactly have a cloud cache

also this includes my own wifi's network latency

let me run this on a cloud server rq

183 github urls in 4.5 seconds @warm cave

im sorry what

the cakes same 53 urls ran in 1 second

[warm cave]

What 😮

The Groq of ai search

[sleek vortex]

ai scraping*

ai search i could do maybe

[warm cave]

Well yeah, depends on if you have it as an api or put it in search platform

[sleek vortex]

my plan is to do both

[warm cave]

Nice

[sleek vortex]

have the exact same quality available in api

as consumer

[warm cave]

Great!

[sleek vortex]

serper.dev is awfully expensive

so is tavilly

surely i could make my own

make it a bit cheaper than them

profit?

like even before i get to offering actual ai models

i could offer apis for every part of the chain

[warm cave]

Yeah, that is true, have many products, and then use those to build up you main project 💡

And you have a product you can sell very soon

[sleek vortex]

plus it would be a way to raise attention and interested developers before i finish my ai thing

plus i think i can ship a nicer landing page than all of these other websites...

[warm cave]

I like it!

[sleek vortex]

i was rewriting sidestore.io for the 4th time to test solidjs

actually go dm lol

[gentle dirge]

😮

[elder magnet]

How does ChatGPT-4o compare to Perplexity now?

[steel hawk]

Hello, I should do school presentation about AI (even Perplexity.ai), could you please help me what should be definitely there?

[dark ginkgo]

Here check out this link from perplexity that explains the differences: https://www.perplexity.ai/search/how-does-perplexity-Qm71LBYBSkOApKNFCISS0Q

[sleek vortex]

what like a speech

i would do like advantages and risks

[steel hawk]

PowerPoint Presentation about how it works

[steel hawk]

I need to have, but I can't find anything about how does it work

[warm cave]

Is it a technical presentation, I know a good video

If you are trying understanding how GPT works

[steel hawk]

Yeah technically

[steel hawk]

Well it should be AI generally

[warm cave]

AI is a massive topic, do you have a focus like LLM’s (Ai models that communicate with human language)

[steel hawk]

No, I don't, it's just for school presentation complexly how ai works

[warm cave]

I will share the video I have about GPT’s, and btw GPT is an architecture in this context, not the name chatGPT

https://youtu.be/wjZofJX0v4M?feature=shared

[warm cave]

Ok, so you are doing more of a large overview

[steel hawk]

Yes

[warm cave]

Ok 👍

[steel hawk]

Well what if I don't really understand it? I don't really study IT just have it in school

[serene sand]

https://www.youtube.com/watch?v=OFS90-FX6pg&t=612s

[warm cave]

Great video, that would be good for the natural language processing part

[warm cave]

https://www.perplexity.ai/search/Please-provide-a-3D5Vd5XbRl25cylq.8XofQ#0

Here is a big picture overview of AI

@steel hawk Obviously for a presentations for school you wont need to be in depth on the various fields, but you can touch in each one, and go into detail on a few particular interests fields and the real-world applications, etc

Obviously depending on the expectations

chatGPT type ai would fit under NLP in the one I sent

[lethal bobcat]

I opened the site today and almost panicked that it was down.

[warm cave]

https://discord.com/channels/1047197230748151888/1244758414794428516

[elder magnet]

But that is 4 not 4o before they made it multi modal and internet search for all etc.

[dark ginkgo]

I’m not sure what you mean by that. Isn’t it just a comparison of the two different AI engines?

[crude drum]

Use AI to create your presentation. Tell it what you are looking for, ask it for an outline, then take that outline a chunk at a time and ask it to write your talking points, suggest visuals, etc.

[elder magnet]

Openainew model is gpt-4o which is much different from gpt-4

[hardy willow]

I've been using AI for ~30+ hours a week, every single week, for just under 2 years now. Some weeks I've leveraged 8-10 LLMs on various paid models, for my specific work... and I have to say something very unpopular I'm guessing, but objectively true nonetheless form my personal perspective of hundreds of hours of use: Claude 3 is the single worst AI I've ever used. Opus was touted as something special, but after a few months on the market, it's fallen so flat it would be hard to describe.

It is not only useless due to having no native internet connectivity, immediately relegating and consigning it to the trash bin of bygone AIs, but aside from Google Gemini (of course), it is the single most handicapped stifled overly censorious and needlessly limited LLM by a landslide.

Even typical daily prompts, 100% non "controversial" will yield massive gatekeeping triggers and information firewalls that truly seem far more aligned with corporations and "health" organizations, than any semblance of reality and useability.

Other than using it rarely as a pure "rewrite" tool, it is the digital equivalent of a pet rock used as a paperweight.

[blissful pendant]

GPT4o, Turbo, and Claude 3 Opus & Sonnet are my go to models. What are yours then?

[steel hawk]

I would need just something simple I guess

[warm cave]

If you want you can just pick a few topics and ask it to explain the topic like I'm 5. Its a great starting point

Then you can work from there

[cinder comet]

the page feature is nice, but one thing I noticed a lot is that it fails to follow instructions when you select an output and send your custom instruction to edit

I guess it only refers to the sources per section, but hopefully it will be improved and extended to :

• search all sources on the page.
• if not found, use the no-internet option (you may need to enable it first).

also ive noticed it rewrites the same output in the same section a lot

why isnt the tts available on the desktop ver?

[quiet monolith]

Hi everyone, I have a relatively long thread on perplexity but I wanna go to the top of it. Scrolling only gets me so far. Is there a page button I am missing somewhere? How do I get to the very top? Or, are old messages simply deleted?
Chatgpt seems to save everything, no matter how old

[cloud sigil]

What ai does perplexity ai use? I’m free

[warm cave]

Claude 3 Haiku

[halcyon glacier]

Hi why Mistral Large has disappeared from Perplexity Pro ?

[warm cave]

They did not say, I'm assuming they thought it was outdated, if you want it back there is an extension you can use

#😎│cool-projects message

[halcyon glacier]

i go on the official mistral website for it

but it looks very different

[warm cave]

What looks different? the Mistral site looks the same for me

[halcyon glacier]

I asked a question about necklace

and the chatbot from the official website acted silly

like it cant check for you stores

[warm cave]

Perplexity has web search, the official website does not, do you think this could be the difference?

[halcyon glacier]

there's no image

no links showed

it says in advance it's sorry to not be able to help me

[warm cave]

The official website dose not support images or links

[cloud sigil]

never heard of them. How do they compare to ChatGPT 4o or others?

Also does anyone know if I use my 5 daily pro questions, which ai they use?

[warm cave]

Its better than GPT 3.5, but not nearly as good as GPT4, Claude haiku is the budget model from Claude, Claude has Haiku, Sonnet and Opus. With Opus being the most powerful. But haiku is solid for most tasks

[warm cave]

Your 5 daily pro dose not change the model used, it just makes it use 20 ish sources instead or 5 ish

[warm cave]

If you want Mistral back you can grab the other extension, otherwise you can keep using the official website, but you will lack the features that perplexity gives

But you can always paste in web search content, a little more manual but gets the job done

[cloud sigil]

Ic so would normal ppx be ok for school work/mcq

[trail pebble]

#💬│general message regarding this. I would happily pay on a per-article basis, say something like $0.50 to export a page (I know I could do this for free using some simple scraping), but for a blog it is important to have as much control as possible, so having export options to markdown or html would be helpful (possibly using pandoc). Ideally re-exporting the same page would be free if there were updates, as long as it wasn't being clearly abused

[cursive lodge]

What happened?

Does anyone know?

[sleek vortex]

they have too many customers and not enough ai units on azure

so often downtimes for the last like month since enterprise pro dropped

this usually goes away in like 30 mins

[cursive lodge]

Can you charge him and solve this problem?

[sleek vortex]

wha

i can’t do anything about this

i don’t work at perplexity lol

you’ll just have to wait for their servers to come back

[cursive lodge]

I've been waiting for 8 hours. I'm going crazy.

[sleek vortex]

reload your page

ctrl + shift + r

or on mac;

cmd + shift + r

[cursive lodge]

Is there any official notice when it can be resolved?

[cursive lodge]

i try a lot

[sleek vortex]

usually in #🔔│status but unfortunately i think this isn’t perplexity’s actual timezone so they might be off work right now not sure

[sleek vortex]

then it’s their servers

you’ll just have to come back in a bit

not much you can do :(

[cursive lodge]

ok.

😂

@frank rivet Bro, fix him quickly. I need your tools.

[acoustic void]

Can I get an invoice if I buy with a bank card from China?

[rancid dome]

What do you use in place of Claude 3 Opus?

[sullen mural]

hi

[dense goblet]

Perplexiry is not reliable at all. Doing stupid things. I asked the same question in Poe (GPT4) and the answer was correct. 👎

[fading moth]

what was the question

some LLMs are better at some things than others, calling one bad because it is less efficient in one category than another is innacurate

[shut plaza]

Are you using the same model in Perplexity?

[shut plaza]

Ah

[fading moth]

GPT4 and GPT4o are not the same

[shut plaza]

Poe also has GPT-4o

I think

[fading moth]

Yes, but GPT4o is listed as the model used there, and they said GPT4 got it right

[dense goblet]

I switched to Claude opus and the answer was worse

[fading moth]

again, you haven't told us the question

it's hard to say what is "worse" when we don't know what was asked

[grave coyote]

If you ask an AI model the same question twice you will get 2 different answers. Both Poe and Perplexity just pass your query to OpenAI's servers. They may add some extra magic to your prompt but saying one is better than the other is just illogical.

[somber geode]

Is Perplexity and ChatGPT run on the same servers because both services were facing trouble last night

[agile jay]

Not just them, a lot of sites were down.

Most likely a key service in their infra was down.

[somber geode]

Claude was up! But this can't be going on for longer... Like I can't be paying for 5 AI chatbots each month just to have access! This is crazy! 😂

[agile jay]

Lol, you do it just for claude?

[somber geode]

Yeah... Claude's long context window helps! Perplexity you have search! ChatGPT you have memories. Midjourney the images. and Udio for music. This is getting out of hand! 💀

[agile jay]

Have good job and it's no problem.

[somber geode]

And they used to say that just get an internet connection and everything else is free

[somber geode]

Btw, I see you are a writer. Where do you write?

[agile jay]

I own a publishing company so I don't actually write, I write so I know how it is to write for the writers we work with.

But back in uni, I was the editor for a big author.

[somber geode]

Ah cool... what company? Like paperbacks or online?

https://tenor.com/view/scientist-myself-youknow-science-gif-18288578

I am also something of a writer myself...😅

https://www.howtogeek.com/perplexity-is-the-only-paid-ai-chatbot-you-need-heres-why/

[agile jay]

I don't dox myself on socials, but it's both digital and physical publishing. Most of the money is made from digital nowadays though.

[agile jay]

What genre you write?

[somber geode]

Ah cool! No problem! I focus on Generative AI, Linux, Productivity, and Marketing (write for Authority Hacker)... So somewhat all over the place

[agile jay]

Oh, so more article etc rather than actual books.

[somber geode]

Yeah... no books.. yet! Hope to dive into novel writing but it requires so much patient.. like writing a good novel is at least 2-3 months, and then waiting on the returns!

[agile jay]

Yep, even longer if you are going through traditional publishing.

Normally you have to wait for over 12 months before any returns.

[somber geode]

Yeah... online articles are faster and more safer! The idea is to save up enough money doing this that I can explore novel writing later on

[agile jay]

Yep, but the good parts about publishing articles can also be the bad part, since there is a lot more competition.

Writing a book is pretty easy. Just takes a lot more patience.

Write 30mins a day, and you'll have a book pretty fast.