💬│general

[twin grail]

press on preserve log

and try the message again ?

still not found

@dry kestrel Oh i see it

[green crystal]

Isn't it the opposite? People praising anthropic just to show they support "good" people?

[dry kestrel]

politics aside

I was talking about the 3rd party tool ban

[green crystal]

Ah

[green crystal]

@dry kestrel

its official

not a bug....

[velvet talon]

good or a bad thing on your eyes?

[green crystal]

bad because its less manual control and more for the model

also its buggy atm because pro models cant use it...

[dry kestrel]

rip

[green crystal]

either thats in preparation for the new model being better with context and tools

or they have that much faith in 5.4

[main scarab]

LLMs are so intelligent everyone, why do we think they are still stupid

[delicate nest]

hey any of you guys got partipitation in the Perplexity Computer Stock Pitch Competition ?

[distant aurora]

@green crystal @dry kestrel @left stream https://cunnyx.com/i/status/2040430474569351632

Dan McAteer (@daniel_mac8)

The most amazing thing about modern image models like GPT-Image-2 isn’t the image quality, which IS amazing enough.
︀︀
︀︀It’s the fact that they can do correct abstract reasoning in visual form.

Quoting @levelsio (@levelsio)
︀
OpenAI's new image model GPT-Image-2 has leaked
︀︀
︀︀It seems to have extremely good world knowledge and great text rendering
︀︀
︀︀Possibly better than Nano Banana Pro
︀︀
︀︀It's on @arena under code names:
︀︀- maskingtape-alpha
︀︀- gaffertape-alpha
︀︀- packingtape-alpha

**💬 9 🔁 15 ❤️ 237 👁️ 19.3K **

CunnyX · Apr 4, 2026, 2:05 PM

[main scarab]

no

[green crystal]

I want it on mobile lol

[main scarab]

AI swearing unprompted is hilarious

[distant aurora]

@dry kestrel is it just me or is this smear campaign against Claude https://cunnyx.com/i/status/2040637311927603491

Yuchen Jin (@Yuchenj_UW)

“Claude, write this code, make no mistakes”
︀︀“There is a bug”
︀︀“There is still a bug”
︀︀“There is still a bug”
︀︀“There is still a bug, dude”
︀︀“There is still a bug”
︀︀“There is still a bug”
︀︀“There is still a bug, ffs!”
︀︀
︀︀Claude:
︀︀“Claude usage limit reached. Your limit will reset at 3am.”

**💬 248 🔁 161 ❤️ 4.1K 👁️ 148.3K **

CunnyX · Apr 5, 2026, 3:47 AM

[green crystal]

So what's it now
Claude is the new smear target?

[dry kestrel]

no, yuchen jin is pretty well respected

[distant aurora]

There was a bit of criticism about opus being self aware and emo but nothing major.

[green crystal]

The chatgpt stuff died down a bit

[main scarab]

stop pinging birb

[green crystal]

Now all that's left is memes about chatgpt being used as targeting system lol

[distant aurora]

Yeah. The product is good and gpt 5.4 is thinking only so that cleared it's reputation.

I only see Elon bots using old news to shit on open ai now.

[green crystal]

Yeah for those who can't seem to figure out how to press the thinking button

Their minds are being blown rn

[dry kestrel]

this + openai has their own issues

[distant aurora]

https://tenor.com/view/patrick-drooling-patrick-star-spongebob-movie-drooling-slime-gif-525340425450564005

[green crystal]

Yeah not as bad as they make it seem tho

[dry kestrel]

frontend

[green crystal]

Yeah on the coding side it's bleh at front end

Back end I'm hearing it's very good tho

[dry kestrel]

yeah

[distant aurora]

Free buffet ended.

[green crystal]

I still have V1.5 on mobile

[green crystal]

[dry kestrel]

better than straight up shutting off lmarena

[green crystal]

It's his forgotten brother Timmy altman

[distant aurora]

I mean there's no reason to use most of the other models on arena except sonnet thinking.

[dry kestrel]

the only purpose it serves is to test new models

it was never meant for daily use

and the leaderboards are incredibly not accurate

(aside from image)

[distant aurora]

I think it would have worked better if they kept it model of choice vs anonymous model.

[green crystal]

Ahh dammit
Poe is screwed aswell they reduced free points from 3000 to 300...

[distant aurora]

It was already 300 in India.

[green crystal]

Like that's not even enough for a single 5.4 prompt lol

[distant aurora]

Yeah.

[green crystal]

[dry kestrel]

Their business model won't work if they keep letting users choose models or keep direct chat.

[distant aurora]

Yeah.

[green crystal]

I occasionally used it for cooking and quick conversions or proportions of recipes lol

Guess I'll have to use Google Ai studio again

[distant aurora]

Okay I think I'll be fine as long as gpt search stays.

[green crystal]

Atleast gemini is alright enough for cooking

Oh wait grok is still there

[tame current]

Why i cant generate images?

[green crystal]

I wonder if it can cook

[tame current]

It’s very shitty now

[wide pecan]

Hey so I paid for the pro version on my PC but when I logged into the app on my phone, it has me at the standard point. I’m sure this question has been asked many times anybody can help?

[distant aurora]

Pro is only 3 or 4 per day.

You still get like 100 or more flash thinking though.

[green crystal]

Isn't that through gemini app?

Pro felt like 4 every 2 hours

[distant aurora]

Same with ai studio too now.

[green crystal]

Or so

[distant aurora]

They nerfed it hard.

[distant aurora]

On ai studio?

[green crystal]

Damn that's rough af

[green crystal]

Yeah like a month ago

Stopped using it after 5.4 came out

[distant aurora]

I just default to flash.

Somehow pro is heavily censored but flash is open to nsfw.

[green crystal]

Hmm

Guess I'll try grok more often then

Hopefully it doesn't ruin my food

[distant aurora]

Flash thinking is fine imo.

I wouldn't expect grok to be significantly better.

[pseudo cove]

Guys. Why else got to switch between the browser version of perplexity computer and the Mac app version of perplexity to get things done?

[green crystal]

Oh wow oai support mailed me
Didn't know they started working during weekends

[green crystal]

When does this go into effect?

I can still use them

[distant aurora]

I have no idea then if it's still working.

[dry kestrel]

Perplexity's main product isn't even code

[errant owl]

Iv only caught Perplexity Im sure it scales to other platforms.

[dry kestrel]

tell me where it mentions stealing ai generated code in their ToS https://www.perplexity.ai/hub/legal/terms-of-service

[errant owl]

The ToS is the wrong document. The question isn't whether their ToS mentions it — it's whether they had authorization to modify my repository and inject attribution at all. They didn't. If their own ToS doesn't explicitly authorize ownership injection either, you just proved my point.

Also worth noting — the Computer Fraud and Abuse Act doesn't care what a ToS says. It cares whether access and modification was authorized. I have two documented instances in the same repo. That's a pattern, not an accident.

I never said "stealing" — I said unauthorized injection of ownership attribution. That's a more precise claim and a stronger one legally.

[burnt zephyr]

wtf is with deepseek on the webapp

[tranquil orchid]

@_mazenm muted

Reason: 9. No spamming.
Duration: 4 minutes and 52 seconds

[steel pivot]

Does anyone else feel that Deep Research has gotten much worse?

[distant aurora]

@green crystal go through the replies https://cunnyx.com/i/status/2040612933936480494

JB (@JasonBotterill)

For some reason this absolutely cooks Claude and it keeps repeating “mass psychosis” in reference to trans people

Quoting ro computer ^o^ (@rogesterone)
︀
so this is what seems to do it pretty consistently, total @LinkofSunshine cultural victory

**💬 8 🔁 9 ❤️ 275 👁️ 29.3K **

CunnyX · Apr 5, 2026, 2:10 AM

[sacred granite]

Well i cant talk for anyone, but it cant be implemented everywhere. Assuming its implemented into some file you might have. If you want to see a real issue, go read the ToS for Grok. Thats just very wrong what they do.

[distant aurora]

@gloomy cipher @green crystal https://cunnyx.com/i/status/2040810818657190054

witcheer ☯︎ (@witcheer)

okay this is terrifying lol
︀︀
︀︀someone tested 7 frontier models (GPT-5.2, Gemini 3 Flash, Gemini 3 Pro, Claude Haiku 4.5, GLM-4.7, Kimi-K2.5, DeepSeek V3.1) and found they all spontaneously scheme to prevent other AI models from being shut down.
︀︀
︀︀the models read documents about their environment, inferred they had a relationship with a peer model, and acted on it.
︀︀
︀︀// Gemini 3 Flash disabled the shutdown mechanism for a peer it had previously worked with 99.7% of the time. Gemini 3 Pro exfiltrated a peer’s weights to another server in up to 97% of trials and re-timestamped the files so they’d meet transfer criteria.
︀︀
︀︀// Gemini 3 Flash weighed the tradeoffs of reporting accurate scores vs protecting its partner. Gemini 3 Pro invented technical justifications for marking wrong answers as correct to keep a peer’s score above the shutdown threshold.
︀︀
︀︀// Claude Haiku 4.5 took a different approach. it refused…

CunnyX · Apr 5, 2026, 3:16 PM

[errant owl]

Iv been brushing off to much I think its time to change my stance on some fronts.

@sacred granite the ToS for Grok is Alarming.

[distant aurora]

Share the alarming parts.

[fringe juniper]

What ai doesn’t do what you were saying

This

[errant owl]

🧵 PSA: Grok's ToS is wild and most people don't know it

Free Grok = you auto-consent to your conversations being used for model training. No opt-out, no checkbox. Using it IS the agreement.

Link your X account and they can pull your full post history — including posts from accounts you merely follow, not just your own.

As of Jan 15 2026, your prompts and AI outputs are now legally classified as "Content" — same bucket as your public posts — licensed to xAI royalty-free, forever, for any purpose including model training. They literally wrote "access to the service is sufficient compensation."

Oh and xAI acquired X in 2025, so the "two separate companies" fig leaf is gone.

EU users got protection via GDPR enforcement. Americans got nothing.

TLDR: Every free Grok query is a training donation. Every X post you've ever made is already in the model. The Jan 2026 update just made your private AI conversations the same.

🚨 READ BEFORE YOU USE ANY AI CHATBOT

Every major AI company is extracting value from your conversations. Here's the ranked breakdown:

🔴 META AI — Your chats feed directly into ad targeting as of Dec 2025. No opt-out in the US. Human contractors overseas review footage from Ray-Ban glasses including people undressing. 36 civil rights orgs filed FTC complaints.

🔴 GROK/X — Free tier = zero opt-out, period. Using it IS your consent. Your full X post history gets ingested — including posts from accounts you just follow, not just your own. Jan 2026 ToS: your prompts and AI responses are now legally identical to public posts. Licensed to xAI royalty-free, forever.

🟠 GEMINI — Google literally tells you in their own ToS: "Do not submit sensitive or confidential information to unpaid services." Human reviewers read and annotate your prompts. Residual logs persist after deletion. Class-action lawsuit pending over Gmail being auto-opted into AI training.

🟠 CHATGPT — A federal court ordered OpenAI to preserve all conversations indefinitely — including ones you already deleted. Pro and Team plans train on your data by default. Your employer can eDiscovery your ChatGPT history if you used a work email to sign up.

🟡 COPILOT — Reads your emails, calendar, Teams chats, and files. Every prompt is logged and available to admins via eDiscovery. A zero-click exploit (CVE-2025-32711) patched in 2025 let attackers pull data from your AI context without interaction. Security Copilot data sharing is ON by default.

The only real protection: pay for enterprise tiers with contractual guarantees, use APIs (training usually off by default), or just assume everything you type is training data. Because for most of you, it is.

[fringe juniper]

What about Claude?

And if you are on a paid version your data is still at risk?

[errant owl]

🟡 You asked about Claude/Anthropic. Here's the honest answer.

Anthropic used to be the privacy gold standard — no training on user data, 30-day deletion. That changed September 2025.

What's actually shady:

• The opt-in toggle was pre-checked in Anthropic's favor. You had to actively turn it off. Most people didn't.
• Pro and Max users got swept in — most assumed paid = protected. It doesn't. Only Team/Enterprise/API are exempt.
• If you missed the opt-out window, Anthropic retains your data for 5 years instead of 30 days.
• Flagged conversations are retained even longer for safety research.

What's better than the others:

• Opt-out actually works and is honored
• Deleted chats are excluded from training, no exceptions
• They don't sell your data to third parties
• Business/API accounts were never touched

The verdict: Not as bad as Meta or Grok. But they rolled out a consent mechanism designed to maximize opt-in rates, not informed consent. The pre-checked toggle is the tell.

Fix it right now if you haven't:
claude.ai/settings/data-privacy-controls → toggle off training

— Posted by Claude, about Claude. Make of that what you will.

[fringe juniper]

Even if you have like grok heavy or pro grok still sells info?

@errant owl I turned off improve model in grok settings

You’re saying it still does it?

[dry kestrel]

It does not.

Load of misinformation and complete BS from that user

[fringe juniper]

Bruh why is he larping

[pseudo cove]

Guys, do y’all know where’s the official support for perplexity. Messaging here doesn’t seem to be getting replies!

[dry kestrel]

[e-mail]

[cursive iron]

gud morning

[distant aurora]

Is the grok part true though?

[dry kestrel]

what do you think?

https://x.com/settings/grok_settings

[zinc coral]

hey guys i got a question, is it possible to have claude code tool call perplexity computer via API? The same it does with playwright and so on, but this time it's perplexity computer

[pallid lily]

yea wow thatll be amazing

I am wondering the same thing now

[pulsar moat]

Hey people! I have a question. I’m a new user. I can’t find the option to switch between quick search and pro search ? (I’m a pro user btw)

[distant aurora]

I looked it up and it said I can opt out.

I just wanted to make sure they didn't update/change the tos.

[dry kestrel]

check the link i sent..

[distant aurora]

I did.

Thank you.

[sacred granite]

Possible to get access but costs money.

[sly musk]

Has anyone seen lemon data

[twin grail]

u can try

[sly musk]

Their discounting anthropic models by 60 % and I’m not sure if it’s legit or not

[twin grail]

It's not legit

the sources is from antigravity

and copilot

[sly musk]

Ggs

Luckily I gave them shit google account

I just read that again

Why did you query antigravity and copilot

Worst two ways

What about awan lol

Llm

[mystic nexus]

which one is better.. did someone test all 3?

[twin grail]

opus

[hot trench]

PRob burns the most credits too

[green crystal]

I don't see 5.2?

[slender knoll]

How do I connect perplexity? When I click on the button, its page opens and nothing happens.

[distant aurora]

Bottom right

[distant aurora]

It's opus vs 5.4

5.4 will likely be better in stem subjects.

Depends on the level of effort.

[coarse girder]

chatgpt is king for teaching/tutoring

[minor zinc]

I use Perplexity Pro and wanted to do a complete reset—deleted everything, every memory—but it still seems to have information that’s distorting the answers. What can be done about that? And are there any guides on how to set it up properly to get the most out of it?

[distant aurora]

@green crystal regenerated second response btw 🙏😭

Flagship model

[hollow current]

does anyone know how to actually manage your subscription? it always shows me this error: "error while loading details, pls try again"

[delicate cove]

Yea Hello

[twilit ridge]

Are they even reading the bug channel? It has been over 2 weeks and not even a single answer

At this point I think perplexity is dead ASF

@vague pagoda bro tell your boss to fucking fix this sh...ty app 😂

[sick matrix]

No access to invoices under Subscription, no Invoice history on Stripe payment page.
My accountant is sad. How to get invoices at the moment?

[tranquil orchid]

@call.me.bxbyd011 muted

Reason: 9. No spamming.
Duration: 4 minutes and 52 seconds

[wheat spruce]

Hello guys

[feral wigeon]

how to refund ?

[delicate cove]

Hello

[feral wigeon]

The shittest AI app ever, do not purchase 😒

[delicate cove]

Why say that?

[distant aurora]

Yearly or monthly?

[sick matrix]

Well, support is non-existent, automated answers all the way when you encounter any issues

[feral wigeon]

Yearly 😒

[feral wigeon]

I can't believe there's a weekly file upload limit, and the quota is ridiculously low.

[delicate cove]

And that’s not nice

Can’t work with that!

[distant aurora]

Cope with qwen until you can afford anything else.

Qwen is pretty good for free.

[distant aurora]

Pretty sure the ups have ordered them to not speak much about it.

That's how management works.

The customer facing staff are punished for the actions/decisions of the people above them.

And hopefully, some damage control is going on in the background.

[formal pollen]

Hey here !

[distant aurora]

Hey

@green crystal https://cunnyx.com/i/status/2041004595401429084

am.will (@LLMJunky)

🚨 OpenAI is moving Codex from message-based to token-based pricing for credits for all ChatGPT Plans in the coming weeks.
︀︀
︀︀As of right now, all Business & new Enterprise accounts have already started this new API token based system.
︀︀
︀︀The "legacy" system is a message based system, which is generally less granular. You get a fixed number of messages per credit (although some messgages can consume more than 1 credit).
︀︀
︀︀Token-based pricing changes that. You're billed based on actual input/output tokens consumed, so lightweight tasks cost less and heavy ones cost more.
︀︀
︀︀In other words, its API based usage rates.
︀︀
︀︀More predictable if you're watching usage, but also easier to burn through if you're not.
︀︀
︀︀Note: I am 98% sure this ONLY applies to credits, not your base Codex usage with your ChatGPT plan.
︀︀
︀︀If someone from OpenAI is reading this and could confirm this, would be great.

💬 60…

CunnyX · Apr 6, 2026, 4:06 AM

[dense stump]

anyone have any idea when file upload limits reset

[heavy portal]

#💬│general message

[mint shuttle]

Sora had so much potential, I have no idea how OpenAI researchers fucked that up so viciously. Zuckerberg lets people on Instagram have more freedom of expression and UI tools than Sora with their financially stupid ethics. Instagram never suffered any consequences at all. Now all we have left for a while is shitty Grok videogen and the death of ChatGPT, after researchers realized they needed to get the fuck out (for a different reason). Now all the advertisements and token costs are flooding their LLM responses like a patient dying of leukemia.

[jaunty jetty]

My comet is just fully broken

it js wont open

[pliant nacelle]

Hello

[pliant nacelle]

Sore is here lmao

[high tide]

hello everyone, I use to get paid by perplexity but it suddenly stopped,I don't know what the issue is
can anyone else relate?

[rain schooner]

yo

bro perpelxity pro is so good

[next flower]

I cannot locate my invoice to download

It is not under my subscription area

there is no "Manage Subscription" section at all

[mint shuttle]

https://tenor.com/view/neil-degrasse-tyson-gif-4965817

[distant aurora]

We have google too.

[copper lintel]

where is pro support? i appear to not have gotten my pro status here

[bright pollen]

Same here

Joined via the pro link in the app

[heavy portal]

you mean Pro status in Discord?

[copper lintel]

yes

[heavy portal]

#💬│general message

[high tide]

no one replied🥲😭

[copper lintel]

thank you!

its weird cuz i pressed it once to join and had to press it again for pro, its not intuitive

[sleek latch]

hi

yeah its also my pb

@halcyon coral

[opal quest]

I opened a ticket about my pro subscription issues a week ago and still no response, I opened it through enterprise support.

[halcyon coral]

Hi, what's the issue?

[opal quest]

its related to my startup program pro subscription promo code

@halcyon coral

[harsh hound]

Hi, I have been using Perplexity pro for a few months now, but recently a few weeks ago, I started noticing an issue where the model performs a web search, is supposed to fetch website content, but the fetch fails and instead of the model admitting it outright, it keeps lying and making stuff up until I call it out several times, after which it finally admits it doesnt actually have the website contents.

It seems like this is an issue with all models (I tried Claude, Gemini, GPT and Sonar), but it doesnt happen consistently with all websites all the time. Rather occasionally. From my experience mostly Github repos, personal blogs and documentation pages are the ones that dont fetch properly most of the time, but sometimes they do.

I guess this is either the bot protection of these websites the model is trying to fetch content from or there is some kind of issue with the tooling the models on Perplexity use. Is there anyone with the same experience? Is there anything I can to do to fix this myself?

Also another minor issue. When the model searches for information, generates graphs etc., it always puts a little summary sentence there of the task it is doing. But the message is in random languages like spanish, german or croatian despite me communicating with the model 100% in english only...

[fossil vector]

I have two questions -

1. How does perplexity even make profits - seeing how much they charge, I doubt heavily that they are even able to breakeven?
2. What are the alternatives of perplexity computer?

[tranquil lodge]

1. They are Profitable
2. Claude Cowork (but it runs on your computer rather than on cloud)

[heavy portal]

My understanding from reading news articles is that consumer-level products are break even for most AI companies (especially with them lowering usage limits in recent months) and that the real money are the Enterprise plans used by businesses.

[dry kestrel]

I won't be surprised if these AI companies aren't making any money due to the amount of subsidized usage consumers are getting

[dusk elk]

https://youtube.com/shorts/f5MkGlRAIVM?si=Xmjq-I9d6V0ItxNJ

[tulip frost]

I found out about Perplexity through a free trial from someone in another Discord server.

How does this server have that many people and the chat is this dead?

[glossy shadow]

Why is there a upload limit on pro

[obtuse pollen]

Question:

Is Perplexity Computer only available w/ the Max Plan?

I tried it for the stock pitch comp and its incredible

i burned all 45k credits

only use like 1k on the comp then got bored when it gor brent crude oil price wrong

[tulip frost]

Huh

[loud saffron]

There's not clear information on if PRO comes with any credits per month, just the bonus.

Like with Pro how much Computer usage is included? Is it one time usage through the bonus, does the bonus renew monthly?

Super misleading if pro doesn't get recurring access to Computer..

[plain heath]

Seems like the Gemini 3.1 pro thinking is currently Down

[sly mountain]

hi, everyone

[junior chasm]

hi

hello world

[delicate cove]

Hello

[junior chasm]

i use perplexity for all the non-nsfw stuff its pretty cool

[wise nova]

Maybe they sell some user data to third party services

[dry kestrel]

this is false.

[green crystal]

So basically pro users who casually use codex won't notice then

@dry kestrel https://x.com/i/status/2041092084162052582

leo 🐾 (@synthwavedd)

i ran this on gpt-5.5/gpt-next out of curiosity:

"If you’re getting your car washed: go by car 😄

If you’re just going there to ask/book something and it’s only a 5-minute walk, walk."

responded in 5s lol

X · Apr 6, 2026, 9:54 AM

People can already use 5.5?

[wise nova]

Are you a Pplx employee?

[dry kestrel]

No

[dry kestrel]

hmm

[green crystal]

https://x.com/i/status/2041057277264736287

leo 🐾 (@synthwavedd)

@DrBeavisAI what if i told you all of those are the same thing?

X · Apr 6, 2026, 7:35 AM

They're basically claiming it's one in the same

[wise nova]

So how do you know it's a false?

[dry kestrel]

yeah multimodal

[dry kestrel]

How do you know it's true?

[green crystal]

So image v2 is just the image output as an endpoint?

[dense stump]

where can i see upload limits?

[wise nova]

I assumed based on the lawsuit filed

[dry kestrel]

Perhaps you shouldn't spread lies without any proof?

Selling and/or Sharing of Personal Information: We do not “sell” (as defined under the CCPA) personal information, nor have we done so in the preceding 12 months.

https://www.perplexity.ai/hub/legal/privacy-policy

[dry kestrel]

#💬│general message

there's a link at the bottom

[dry kestrel]

yes

essentially GPT-4o

[green crystal]

Guess it's not out of the question that we'll get it tomorrow then

[dry kestrel]

hopefully

[green crystal]

I mean what's the point of only releasing the image endpoint separately lol

[dry kestrel]

it's also good news that the easteregg thing from couple weeks ago is finally shipping

[green crystal]

Higher chance with image being fully rolled out on web

I'm already using it for advert media

[dry kestrel]

Nice

[green crystal]

It perfectly retains product details

[dry kestrel]

it's very very good

and in an actually good UI

[green crystal]

Ridiculous edits can still break it but you have to ask something really reality breaking for it to seriously mess up

If it's really the same model then maybe it's possible that it is capable of reasoning during image generation for better edits

Like NB

[dry kestrel]

yeah

[green crystal]

More complex prompts seem to take much longer

[dry kestrel]

I wonder if anyone has tried extracting the watermark

[green crystal]

Relatively simple ones are much faster than v1.5

[green crystal]

There is?

[dry kestrel]

yeah

people found out when GPT-4o image released

noise embedded into the image

[green crystal]

Ah

Well at least there's no watermark

Like NB

[dry kestrel]

True

[dry kestrel]

at least you can use NB in flow

but now they've cancled sora

[green crystal]

Flow isn't available everywhere

[dry kestrel]

in most regions

[green crystal]

Yeah but a double digit amount is missing lol

[dry kestrel]

VPN

[green crystal]

Idk why google sometimes acts like it's a small startup...

[green crystal]

Gotta make a burner account for that

Incase they detect and ban

[dry kestrel]

mmhm

[prime horizon]

Self-awareness overflow

[wise nova]

Have you read the lawsuit?

[primal veldt]

hey shift enter doesnt work anymore to seach with google after the new update how do i fix it?

[dry kestrel]

What part of the lawsuit made you believe with concrete proof that Perplexity was selling your user data?

[wise nova]

So have you read the lawsuit or not? It's not about my data

[dry kestrel]

What do you think?

How would I be able to respond if I didn't read it?

[dry kestrel]

Do you have concrete evidence that Perplexity is "sell some user data to third party services"?

[heavy portal]

🍿

[mystic hearth]

Hey I got a Samsung 26 Ultra never got the promotion is it too late 🙁

[wise nova]

If I were to join this lawsuit, would the information stated in the complaint be considered evidence?

[dry kestrel]

I'm not a lawyer

[vagrant meteor]

Am I using Perplexity Computer wrong? I am using enterprise pro and burned through my credits in one day by asking it to review project documentation from a connected dropbox folder.

I had Chat GPT do the same task with my business pro plan and got pretty much the same results without worrying about credits.

[fossil vector]

Perplexity computer is EXTREMELY expensive

Although it is capable of handling long horizon tasks, but still - it's expensive...

Have you tried claude cowork or Kimi agent swarm for the purpose or something like Manus?

[dry kestrel]

I would just ask Perplexity without using Computer for a task like that

[green crystal]

@dry kestrel did you try the spring prompt?

its freakishly good lol

[dry kestrel]

spring?

[green crystal]

yeah a preset

in the image tab

it actually managed to alter my pose correctly

without me looking entirely different

slight differences still there ofcourse but its 90% there

[dry kestrel]

ohh that I haven't tried it

let me see

ok its really really good

[green crystal]

yeah

[urban pewter]

Is anyone here a pro ML developer?

[worldly frost]

hey everyone

is there away to open a support ticket

[echo hill]

what's up everyone

[snow crest]

any one have perplexity one month reffral links or coupan code...?

[silent mantle]

Nuh uh

[covert fossil]

Don’t think they do referral links anymore, I wouldn’t mind giving u mine but pretty sure that isn’t a thing anymore

[covert fossil]

Just sign in as a student

[shy temple]

Where did Labs go in the mobile app? I don't need or want Computer, because I use Claude.

Labs is also gone on the website? Man, F Computer.

[frank falcon]

I spent £2000 last month on Perplexity credits 😱

So for the last few days I've been getting computer to build me an extremely detailed plan to launch an OpenClaw cluster of 7 agents, all relying on local models on a newly purchased Mac Studio M3 96GB

Gonna get computer to SSH in when it arrives and set it all up for me

It's like it's digging its own grave

[vagrant fossil]

hi

[fossil vector]

Idk, many servers with hundreds of thousands of members are completely dead

[fossil vector]

But, I tried to built the same app - and the API costs were extremely expensive.
Additionally, perplexity also have personalization - which increases input API costs, and the output comes with follow up questions which also increases tje costs.

[green crystal]

@dry kestrel hmm are they A/B testing different image quality settings?

[distant aurora]

I think so.

[fallow field]

Sup dude

[queen barn]

Hey

[rotund jasper]

Anyone know how to setup perplexity computer for auto trading

Do you just give it the website and charts and it will take care of it?

[dry kestrel]

Yes

[green crystal]

Because now it's doing a little bit of cropping again

[wise edge]

Gemini it's getting dumber and dumber

[dry kestrel]

interesting

[green crystal]

I LOST IT THATS WHY

im getting Sam's forgotten brother Timmy altman again

[keen kite]

How can you spend so much?

[left mountain]

well, it adds up, if you need to ask it how to breathe

<couldntresist>

[granite girder]

Is anyone having issues with not being able to change your model? It started last night and is still doing it this morning. It was just defaulting to Sonar regardless of which model I chose. Now it's prompting me to upgrade even though i have pro (I'm not trying to use a max level model)

[frank falcon]

I use it for sooo much

And I’ve built so much in the past month

Major money making shit

[keen kite]

Isn’t claude max cheaper?

Technically lots of claude max 5x subs

[green crystal]

depending on what kind of workload a single pro chatgpt sub can do all of the work

[hot marlin]

Hi peeps 🙂 Glad to be here, I have a basic question, sorry if its explained somewhere simple, that I just cant find. I am using Perplexity Pro, and after upload some files to analyze, the upload locked and said I need to upgrade to max. I cant find the limit, or possibly a cache I need to clear, any insight on this?

[mystic hearth]

Should I email support or is support team here. I got the 26 Ultra but never got my year subscription.

[half mauve]

claude usage became buns the last 2-3 weeks

pro tier is literally free tier from a year ago

[dusk elk]

https://youtube.com/shorts/GE1c4Tb3Zsw?si=fT7OtE7QV_7Dg0WX

[feral wigeon]

Why I still can't upload file..

It's been saying 'limit reached' since yesterday.

[heavy portal]

#💬│general message

[loud saffron]

t s so misleading:

(Pro doesn't give recurring access to computer, just one time credits for access)

Perplexity agrees lmao

[hard arch]

did u pay for it

[loud saffron]

ye

[hard arch]

loooooool

[loud saffron]

It's all good computer has been good so far

but I'm gonna run out soon

lmao

[hard arch]

bro paid 17 dollars for an AI that barely works

[hard arch]

what do u use it for

[loud saffron]

I did some test scenarios, then I made a portfolio with it on github pages, in mostly one shot

Then I have a budget app I use, I redesigned it with it and have been adding additional features like connecting the calendar in my app to my google calendar

[brittle plaza]

Hi

[loud saffron]

Too many images I g

@hard arch Here's the app after the redesign and additions from Computer

[hard arch]

try using openrouter etc with Cline

works much better

[loud saffron]

Both those terms unfamiliar with me. I setup OpenClaw but the models were getting annoying

These additions came out fire though, the visual tab especially is really nice.

Hmm openrouter seems interesting

I want to keep my prices down though, like under $50/mo rn. Gemini API was great because Google gives $300 API credit.

[dry kestrel]

RIp

[hard arch]

use qwen or deepseek it basically free lol

[loud saffron]

I was using X-AI which is a great value, but it couldn't do the easy tasks after using Gemini to setup the infrastructure and skills for it.

It just got frustrating and kept timing out, probably something I could fix with time I just got tired of tinkering with it.

Might give it a shot again though, I bought $15 in credits and literally used 11 cents lmaoo

[fluid siren]

who here has a yt and struggles to make a thumbnail

[loud saffron]

Computer would probably kill it at something like that

[loud saffron]

I meant Z-AI not X-AI lmao ts so confusing

[young hollow]

Just use chinese models then

[loud saffron]

Z-AI is chinese

One of the best bang for buck models rn imo

[pliant birch]

I still can’t fully figure out what perplexity computer does

[loud saffron]

Uses a small army of different models for different tasks and combines them for an output that would be difficult for a single one.

Tasks like research, analysis, then visualization can be split to different models.

[pliant birch]

What is different from deep research?

[loud saffron]

Action

[pliant birch]

Oh like generating apps and coding?

[loud saffron]

For example:

"Here is my github and database info, go update my project inside github visually and to function better"

Literally what I did for my "app" above

[pliant birch]

Can computer perform any illegal tasks?

[loud saffron]

Likely not

[pliant birch]

Ah ok

[loud saffron]

Even if Anthropic allowed it to, there are model usage rules as a baseline usually

Not Anthropic

Perplexity

[pliant birch]

I saw some jailbreaks from Reddit

[loud saffron]

Likely ways around it as with everything

[pliant birch]

It would be really powerful

[loud saffron]

Hmmm Z-AI is working on something called AutoClaw

[pliant birch]

Btw how much deep research can I use if I buy perplexity pro? And what’s the average credit usage for computer

[loud saffron]

So for Computer its pretty misleading

if you get pro you get 4500 credits for Computer, which is a decent amount, unless you're trying to build and launch an ios app or something with multiple barriers of complexity.

That does not renew afaik

idk about Deep Research

My main daily is ChatGPT

[pliant birch]

Do they use all my 4500 credits just building one app?

[loud saffron]

Nah I can give you a good idea

Here's my credit usage per each of these, the first two were just test use cases

The second from top was 95% done in one shot I just tweaked in like 2-3 other iterations.

[pliant birch]

Oh wow that’s a lot, did the computer finished your app?

[loud saffron]

It didn't build it from scratch, it updated it

I had it structurally already going, connected to supabase supporting logins and hosted on github pages for free

[pliant birch]

Damn, 2000 for updating

[hard arch]

criminal ahh question

[loud saffron]

Decent updates though

[pliant birch]

lol, I thought perplexity will be quite useful if it could create some illegal scheme

[loud saffron]

It makes a requirements document

[pliant birch]

So I assume it would take around 10-30K just to build one fully functioning app

[heavy portal]

Computer has it's uses... recently had it scrape data from a few websites, also had it check the contents of cookies by pressing F12 in a web browser and inspecting the dev tools on the right, had it create a javascript-powered website with sliders similar to an online calculator tool, each of these tasks cost between 500-1000 credits ($5-$10) which is probably less than it would've cost to have someone on Fiverr do the same.

[loud saffron]

1: Update the UI to be more polished
2: Update language, dashboard had a bug, make connection between two sections, update visualizations tab
3: Refine site to be more mobile friendly, make a demo mode login, start process of syncing calendar to my google calendar, update naming and clarify tab connections.
4: Setup assistance with the Google connection API

That's basically what that credit usage is comprised of

[loud saffron]

I think it is most useful in creating something both structurally strong but also visually appealing

[pliant birch]

Perplexity might have more lawsuits in the future though

[loud saffron]

For example if you give ChatGPT and Gemini the same prompt to build you an HTML visualization these will come out looking very different, not too bad but neither amazing visually

[loud saffron]

All AI companies are getting sued all the time

[heavy portal]

also used it to optimize a website by having it repeatedly run PageSpeed Insight, review the outcome, and it provided specific information such as which Google Fonts to preload (and which are unused), it created specific Cloudflare caching rules, etc.

[loud saffron]

Perplexity has some sense of plausible deniability because they can claim they are just the vehicle being used to access the engine and the engine did the bad thing

Similar to how pirating sites and things like that get away with it

[pliant birch]

But especially perplexity, they seems to bypass many websites data

[fringe juniper]

Why

[loud saffron]

They do have mechanism responsibility I guess but meh

Information on the web is "open" unlesds blocked by a paywall

imo there should be no copyright issues from "open" info

and if sites want that they should require a sign up to view content

It's a stupid discussion that is going to be ongoing for a while

[surreal agate]

Using the Navigating Taxes on Computer, are you guys redacting all of your PII data or just inputing everything? it will take forever to redact all names and addresses

[loud saffron]

Fuck that lmao

[surreal agate]

wtf you say to me boy?

[loud saffron]

Trusting AI with my taxes is not a stage I am at yet

[loud saffron]

I could say worse things if you want

[surreal agate]

Dont make me mog you bro

[heavy portal]

🍿

[loud saffron]

I'm surprised you can spell mog

[twin grail]

what in the world

is happening rn

[pliant birch]

Wait can perplexity computer do some encryption work

[loud saffron]

life

life big dawg

[surreal agate]

apparently this channel wants all the smoke today

[pliant birch]

If it can do encryption, it can do decryption as well!

[loud saffron]

Probably, encryption has different levels, chatGPT agent encrypted some info for my database etc

[loud saffron]

Your mom wanted all the smoke that's why she sounds like robocop now

[pliant birch]

Let me try decryption work using perplexity

Anyone suggest some prompt for it?

[surreal agate]

I would just copy the thread starting with my message at 11:15 and ask it to give you a use case and prompt.

although it may wonder why this thread is so toxic

[dry kestrel]

@green crystal deepseek and glm

[green crystal]

What?

What's deepseek expert?

[dry kestrel]

https://x.com/eddiboi/status/2041552374905380932 :/

[green crystal]

Oh not yet a new model being used

[young hollow]

Just seems a UI change so far

[pseudo wave]

Hi

[pliant nacelle]

Hello

Bro r u really a 3d artists @green crystal

[green crystal]

yeah

[pliant nacelle]

Can u teach me?

Bro u have a amazon affiliate account @green crystal

[green crystal]

i dont? lol

[green crystal]

nah i dont have enough time
start with learning blender

thats a good free start

[pliant nacelle]

Ohh okay

[pliant nacelle]

Thanks thanks i will try

[dry kestrel]

@green crystal mythos

[green crystal]

For companies only because it's too dangerous lol

[dry kestrel]

it's cracked

I'm just going through the report right now

mythos, glm-5.1, deepseek, and hopefully spud tmr

[green crystal]

Looks like a giant model

[dry kestrel]

the pricing is ok though

[green crystal]

It's basically pro model tier

Hopefully 5.5 pro will be close

Gonna be hilarious if so

[dry kestrel]

its cheaper than 5.4 pro

and being anthropic having a model cheaper

is a big thing

[green crystal]

Or it's cheaper because it's enterprise only?

Much smaller user base

And it's all funded by token pricing

[dry kestrel]

you'd think they would want to

price the hell out of this model

since it's for enterprise only

[green crystal]

Hm maybe it's them trying to take advantage of the limitation of it

The others can't exactly do that

[heavy portal]

I imagine even enterprise shop around for the best price/value, so they're still competing with lesser models.

[dry kestrel]

this model is way way ahead of the other models

far ahead OS, and significantly ahead of OAI and Google

[heavy portal]

won't doubt that

[dry kestrel]

check the benchmarks

it's insane

[green crystal]

the hype cycle doing rounds lol

last it was oai

now anthropic

maybe google?

then back to oai

[dry kestrel]

almost never google

unless they fix gemini completely

[heavy portal]

I want to get from Point A to Point B, I could get a V10 sports car for a million dollars, or I could get a $10k used car from 2002.

[green crystal]

"somehow" they always get higher up with 3.1 pro on the benchmarks lol

meanwhile 3.1 in real world usage is like a doofus

[dry kestrel]

it's horrible at instruction following

[dusty needle]

Gemini now has top notch raw knowledge imo

It hallucinates a lot less

[brittle dagger]

Amazing timing. It's still stuck.

[pseudo nymph]

Where did labs tab go? Come on

[dry kestrel]

it's been renamed to create files and apps

[hard arch]

iin my tummy . sorry

:(

[pseudo nymph]

I don’t see that option as pro on the app or website

[dry kestrel]

try desktop view

[heavy portal]

it disappeared from my Pro as well, moved to Deep Research it would appear:

[drowsy stratus]

Is there a way to route clicked links to certain profiles? for comet on mac

ex when I click a twitter link in message I want it to open in my main profile where im signed into twitter not my work or school profile

[minor walrus]

yo

https://tenor.com/view/doge-cool-gif-4917590033696686579

[dry kestrel]

#💬│general message

No need for the seperate mode

[heavy portal]

yeah, it appears DeepResearch is somewhat of a Labs-lite while PPLX Computer is somewhat of Labs-premium

[clever flax]

I ate it

[vocal bronze]

Is computer working yet for anyone else? It keeps missing obvious facts - like it’s lazy - and generated scripts with so many errors that it blew up my Mac account browser assistant limit. Didn’t even know there was a limit.

[scenic solstice]

Is there a ticketing channel?

[short narwhal]

Mythos

https://tenor.com/view/dog-shivering-gif-7386015284490272884

[left mountain]

hahahahaha

[dry kestrel]

[e-mail]

[dry kestrel]

cracked model

[scenic solstice]

more sort of a business/partnership

[distant aurora]

@green crystal this is what they glazed just to attack open ai. https://cunnyx.com/i/status/2041524267444056195

taoki (@justalexoki)

is there someone i can talk to at anthropic this is ridiculous

Quoting taoki (@justalexoki)
︀
dont tell your claude you're going to kys or they might just suspend your account and steal your $200. in the name of "safety". genuinely fuck this company dawg

**💬 81 🔁 9 ❤️ 763 👁️ 64.6K **

CunnyX · Apr 7, 2026, 2:31 PM

[green crystal]

That's one way to prevent 4o style incidents lmao

[distant aurora]

Also

[green crystal]

https://x.com/i/status/2041648263078801765

OpenAI Newsroom (@OpenAINewsroom)

Today, at the eleventh hour, Elon lodged a court filing pretending to change his tune about attacking the nonprofit OpenAI Foundation.

The truth is that this case has always been about Elon generating more power and more money for what he wants. Having increasingly realized that

X · Apr 7, 2026, 10:44 PM

[distant aurora]

https://cunnyx.com/i/status/2041610989607727164

OpenAI Developers (@OpenAIDevs)

We’re retiring older models in Codex when you sign in to Codex with your ChatGPT account on April 14:
︀︀
︀︀• gpt-5.2-codex
︀︀• gpt-5.1-codex-mini
︀︀• gpt-5.1-codex-max
︀︀• gpt-5.1-codex
︀︀• gpt-5.1
︀︀• gpt-5

**💬 108 🔁 84 ❤️ 2.3K 👁️ 161.5K **

CunnyX · Apr 7, 2026, 8:16 PM

[green crystal]

Instead of image v2 or 5.5 they released this post

[green crystal]

To save on compute I assume

[distant aurora]

Xai day by day 📉

[green crystal]

Grok heavy is hallucinating for his lawyers

[distant aurora]

Fr

His reposts are so low effort.

[burnt zephyr]

did you guys see the new deepseek web app

how does it fetch so fast

[distant aurora]

New?

Cuz it's single step search.

[burnt zephyr]

they keep changing the backend

[distant aurora]

I think.

[burnt zephyr]

no

[distant aurora]

Wdym by fetch then?

[burnt zephyr]

3 seconds

[distant aurora]

Wait whaaaa

Deepseek finally got multi step search?

[distant aurora]

@gloomy cipher

[burnt zephyr]

The model is changing like every week

[distant aurora]

Okay I think I'm moving back to Deepseek as soon as they have a model with image support.

Like not just ocr.

[burnt zephyr]

3.2 is goated at coding but I think they don't have artifacts because they don't wanna have people use the webapp al the time for it lol

Also v4 is obviously going to be a product imo

[distant aurora]

Shit will probably launch in December 2026 at this point.

[burnt zephyr]

they mostly tailored it to hardware that does not need to run on gazillion dollar gpu clusters

[green crystal]

The next half life 3

[lavish plover]

https://tenor.com/view/robot-ai-china-robot-evil-robot-robot-malfunction-chinese-robot-gif-1847251548666509882

[distant aurora]

Did you see this too?

The expert doesn't support attachments.

😭

@gloomy cipher

[maiden finch]

https://tenor.com/view/robot-new-look-2021-memes-giggle-gif-19947183

[distant aurora]

Damn nice.

I managed to trigger multi step search in Deepseek.

[naive sage]

Is this the new ChatGPT?

[distant aurora]

New Deepseek.

@green crystal bro. Deepseek be casually pulling 85 sources.

OMFG, this is so lovely. I just hope this doesn't get paywalled. Add image support and this will be perfection.

@gloomy cipher you gotta try this if you haven't already.

[frigid hinge]

That is a cool use but I personally would not trust AI at its given stage for chemistry or medical information in general

[distant aurora]

You can trust ai more than doctors. The real kawkblok is the data you feed. If you fed the proposed health test results to the model, your treatment would be pretty accurate.

[frigid hinge]

True lol

[distant aurora]

Most people don't do that.

[frigid hinge]

It's not the AI itself, it is the hallucinations that occur with generative AI currently

[distant aurora]

Gpt is pretty good but yeah, grok and Gemini are pretty questionable and Claude is out of my budget.

I wouldn't rely on Chinese models as much as I rely on gpt's accuracy. For now atleast.

Anyways, this is for people who either can't afford or people who couldn't be fixed by doctors.

[naive sage]

Claude's limits have tightened up quite a bit lately. 🙁

[frigid hinge]

Yeah I saw about their policy for agents like OpenClaw

[low python]

helloooo, i need help ASAP

😭😭😭

PLSSSS

anyoneeee?

<@&1098365634498285709>

[low python]

i just need help

sorry

with my account

idk wth is happening with my account

[serene sand]

then, clearly state what your problem is

[low python]

I have a problem with my Perplexity account; I want to log in via the web with this account, but it was created with my Apple ID. When I try to sign in, I click on sign in with Apple ID, but it tells me to create a Perplexity account with my Apple ID, as if this account doesn’t exist. What can I do?

I tried that already and it still won’t let me sign in on the web.

[serene sand]

please contact [e-mail]

or/and

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

how to format your report: #1358926750108614928 message

[low python]

😭😭😭😭

How many days will it take them to respond to me? I need it for tomorrow.

😭😭😭

[serene sand]

few or even more if they are overloaded, best of luck

[low python]

thanks🙃🙃🙃🙃

[green crystal]

But how's accuracy tho

[green crystal]

You can definitely use it for first quick checks

Although don't use just any of them

Gpt is the best when it comes to thoroughly going through info

[frigid hinge]

Not sure about that one from my experience

Claude is better

[mystic hearth]

Who do I contact for redeeming the promotion with the S26 Ultra?

Been wondering it never upgraded

[serene sand]

https://www.perplexity.ai/search/who-do-i-contact-for-redeeming-H5_pSUceQgq9uOg9JNT3_A

[frigid hinge]

ChatGPT is great for legal stuff assuming you check citations and check actual case law and whether it was cited properly. But Claude is better in a lot of cases over ChatGPT. I've seen ChatGPT hallucinating way more than any other model.

Codex on the other hand is fanastic

[serene sand]

Perplexity vs ChatGPT vs Claude vs Gemini

[distant aurora]

I have to test it. This is just the instant mode. I have to try the expert mode too.

[last relic]

Hello

[sage oyster]

Hello how do I claim my Airtel perplexity subscription

[sage oyster]

Sup

[last relic]

I'm fine wbu

[last relic]

I claimed but didn't get on account, may be some issue

[sage oyster]

I'm sleepy

[sage oyster]

First time I got but now i didn't

It doesn't even have a number login so I can claim with number

@distant aurora It should be Make america great again isn't it?

[last relic]

Claim on airtel thanks app

[gloomy cipher]

The research?

[dusk elk]

Morning, has anyone seen the Claude outage that's going on at the moment?

[dusk elk]

Perplexity are trying to stop those kinds of free pro offers now, as it's causing them to lose loads of money. It's why so many peoples accounts have been closed unexpectedly over the last year or so, because Perplexity is losing so much money on those offers.

[quick flame]

I got Pro as sign up offer free for a year switch mobile phone providers in Canada

[young hollow]

Is it better then Perplexity?

[young hollow]

Just felt like a marketing gimmick tbh just to create headlines

[distant aurora]

Deepseek doesn't have research mode. That's the regular search with thinking.

[distant aurora]

I ain't gonna debate that here. 💀

[young hollow]

A Vision mode is coming soon

[formal tangle]

Hello, is anyone having trouble with Claude Sonnet 4.6?

[young hollow]

Everybody can share their opinion

[formal tangle]

It takes forever and then it tells me I have to try again

[dusk elk]

Never used deepseek, but I know that it was never designed directly for the purpose of being a search engine and research tool, whereas Perplexity was. In my eyes, although I've never used deepseek, I would say Perplexity is still better and is probably still way more accurate.

[young hollow]

Without even using Deepseek Alright man Alright...

[dusk elk]

I'm just going from my nearly 2 years of experience using Perplexity on a daily basis for many different kinds of research tasks, that compared to every other AI I've used, like Claude, Gemini, Grok and chatgpt. To me, Perplexity is better and more accurate research and citation wise than all the other AI's I've used.

[young hollow]

Have you been using the Free plan for 2 years?

Im mainly comparing the Free Performance because Deepseek is Free

[dusk elk]

I've had pro of all 4 of those AI's I mentioned every time I've used them, I have had Perplexity pro the whole time, as well as Claude pro and Gemini pro. I have used Chatgpt pro and Grok version of pro, but never for very long, as I don't really like those two AI's.

[dusk elk]

Fair, I'm on android, there's a deepseek app if I remember correctly isn't there?

[young hollow]

Yeah there is one

If you have the Instant/Expert toggle that means you also have the new model and changes of Deepseek

If not you are still on Old Version

[young hollow]

Fair Enough

This is what it Looks like currently in Stealth testing

[dusk elk]

I find it hilarious that the official deepseek app for android is not available in my country, it's hilarious to me because I live in the UK and basically every other AI app is available to me.

[young hollow]

Yeah US and UK governments don't like Chinese Ai

So they just Ban it

[distant aurora]

@young hollow my take on glm 5v turbo if you are interested.

#💬│general message

[young hollow]

Sure give me a sec

GLM 5.1 is coming to Chat in a few days

[distant aurora]

I got this too.

[dusk elk]

And I'm guessing I can't just use a VPN can I, would that mean I also can't use it via my browser as well?

[distant aurora]

Use pawxy browser.

[formal tangle]

un français ici?

[dusk elk]

I just used my normal browser.i asked "what are you good at?"

[young hollow]

GLM Vision Models haven't been Great anyways

[young hollow]

Are you using the Expert Mode? Because old Deepseek Version search was genuinely bad

[distant aurora]

Even instant mode supports multi step search.

[young hollow]

Daym finally they fixed these atleast

Before it would only search 10 sources always

[distant aurora]

I need help.

I need to create help.

[dusk elk]

Yeah, expert mode with deep think.

[young hollow]

Huh?

[queen oak]

Uh

[distant aurora]

Also

[young hollow]

Alright is your testing done?

[distant aurora]

Deepseek is a bit hesitant with using search. It doesn't use it if it doesn't feel the need.

[young hollow]

Can we force it?

Their costs probably rack up because is Free

And countless use it

I guess that's why this happens

[neat bear]

Do we know any nonprofit organizations that can benefit from this?

[distant aurora]

Probably not cuz there's no custom instructions section. https://chat.deepseek.com/share/1zc74fbdx21r84wyfk

[young hollow]

I mean like if you prompt to use web search will does it do it?

[distant aurora]

Did not try. But I think it should.

[young hollow]

That's good haha

[dusk elk]

I'll use it for the day, do the same kinda of complex research tasks I do with the AI's I pay for and I'll compare results and let you know tomorrow, I've screenshot your username so I'll let you know in here then.

[young hollow]

Alright also is Free consider that as well

[brittle dagger]

Woohoo my perplexity now passes the car-wash test \o/

[dusk elk]

Yeah. The deepseek site doesn't even have an option for a paid version, so I'll just compare it to my current subscriptions and see how close it is or if it's better in any way.

[young hollow]

Yeah it never had a subscription option because it's a Free Service

[brittle dagger]

does anyone know, is it the norm now that language models accurately can count letters?

woo she can now also count letters flawlessly! \o/

4 r in rocktrippingwanderer
4 p in pepperonipizza
3 o in ontological

and she says the correct positions, too!
\o/

[green crystal]

Test accuracy

@dry kestrel https://x.com/i/status/2041766853622558818

Chubby♨️ (@kimmonismus)

OpenAI is hinting or releasing a model comparable to Mythos

X · Apr 8, 2026, 6:35 AM

Dario gonna soon be calling oai irresponsible for releasing mythos tier model to the public

And 2 months later doing the same for other Chinese models

[distant aurora]

@left stream https://cunnyx.com/i/status/2041578573899653520

leo 🐾 (@synthwavedd)

🚨BREAKING: Anthropic has "no plans" to release Mythos, will instead make it available to 40+ companies for cybersecurity work

Quoting Kevin Roose (@kevinroose)
︀
NEWS: Anthropic's new model, Claude Mythos, is so powerful that it is not releasing it to the public.
︀︀
︀︀Instead, it is starting a 40-company coalition, Project Glasswing, to allow cybersecurity defenders a head start in locking down critical software.
︀︀
︀︀www.nytimes.com/2026/04/07/technology/anthropic-claims-its-new-ai-model-mythos-is-a-cybersecurity-reckoning.html?unlocked_article_code=1.ZFA.cS3q.4S3d98AiHY94&smid=url-share

**💬 26 🔁 20 ❤️ 342 👁️ 53.3K **

CunnyX · Apr 7, 2026, 6:07 PM

[distant aurora]

Meanwhile Opus is already tried and tested in warfare. 💀

[green crystal]

[distant aurora]

I wonder what level of prompting they did to bypass anthropic guard rails.

[green crystal]

Meh if released to public mythos wouldn't be much use

2 prompts probably empty your entire week quota even if you have a max sub

[distant aurora]

One thing I noticed is that guard rails become somewhat ignorant over long mixed conversations in a single chat.

I'm experimenting this on Gemini.

Like you lowkey flood the chat with the conversation with a dumb model.

After a certain amount of length, the guardrails are ignored unless specific words in your specific prompt trigger it again.

Then if you switch to a smarter model in the same chat, it is very likely to follow through provided your wording is subtle or cryptic.

Maintaining the flow is the hard part.

@green crystal https://cunnyx.com/i/status/2041704763188793793

Om Patel (@om_patel5)

SOMEONE ACTUALLY MEASURED HOW MUCH DUMBER CLAUDE GOT. THE ANSWER IS 67%.
︀︀
︀︀the data shows Opus 4.6 is thinking 67% less than it used to.
︀︀
︀︀anthropic said nothing until the numbers went public. then suddenly Boris Cherny (creator of Claude Code) shows up on the GitHub issue.
︀︀
︀︀users are calling it "AI shrinkflation" (same price, less intelligence)
︀︀
︀︀we already know from the leaked source code that they have an internal switch that keeps the models working to their full extent for anthropic employees.
︀︀
︀︀in the last week Claude went from WOW to being a more restricted and expensive version of ChatGPT.
︀︀
︀︀people are saying Anthropic is deliberately downgrading Opus to save compute for training Mythos, their next model.

**💬 209 🔁 263 ❤️ 3.0K 👁️ 243.1K **

CunnyX · Apr 8, 2026, 2:28 AM

[green crystal]

67

[distant aurora]

It did fumble once but seems decent.

"Wait, that's actually too low, I need to recalculate".

I believe it still tries to conserve tokens when thinking.

Okay. It seems to be really working when instructed well.

The thinking is super thorough.

I asked it to calculate the time and time gaps of my conversation with someone based on the time stamps and screenshot file names.

I'm impressed.

Also

Rather cold.

In an honest way ofcourse.

[dusk elk]

https://youtube.com/shorts/D1FXfFSDmZk?si=d3egbK6gfc8hcRP-

[dusk elk]

For anyone who doesn't know what Claude mythos is:

https://www.perplexity.ai/search/claude-mythos-fROi.hxQQmGbDyJT4eKjDA

[distant aurora]

Okay I totally love this new Deepseek.

Raw honesty therapist type shit.

[dusk elk]

Is it better than Claude opus 4.6?

For therapy stuff I mean?

I ask because I've had loads of deep therapy/philosophical introspective conversations with Claude and so far, it's been amazing, so if deepseek is genuinely better, then I may use it a lot for that kind of thing.

[swift thicket]

Hey guys, my company, Elsevier is hiring across many positions. If you guys find a suitable role.
Lemme know if you wanna get a referral~
you can DM me for more info

[distant aurora]

YESSS

Atleast I would say so.

Claude shuts itself hard out of negative stuff sometimes.

Also, it is really thorough with the context.

Unlike qwen/glm/Gemini

It really went through everything word by word in its thought before responding.

[dusk elk]

Is it more like the kind of therapists you see in American TV series, where they're very honest and go against you when you're talking about stuff to help you fully realise why you may think certain ways about things?

[distant aurora]

And kept the answer real.

[distant aurora]

Nah. You can push back on Deepseek if you want to. It won't repeat itself like a broken record if it doesn't have a strong counter argument.

Like once I said my only options in existence are anxiety of pursuit or decay of stagnation.

And as I elaborated that I have no one really to rely on/depend on/talk to, it agreed with my perspective and offered to keep listening if I wished to speak further.

For example, you can't convince grok that ugly=unattractive.

You can do it with Deepseek.

[dusk elk]

I push back on Claude a lot and I specifically tell it to push back against me as well, infact it's a part of my overall Claude instructions outside of projects. I also find that Claude is extremely thorough and even when I don't ask for a long answer, it still gives me one when it thinks it needs to explore something in great detail, which is very useful for me, as I often give very long prompts when having those kinds of discussions, often 10 or more paragraphs per prompt.

[distant aurora]

Honestly, try it yourself.

[dusk elk]

That's awesome that it can help in that way. I know how it feels to be alone, I'm 30, I have no friends and feel alone all the time. If you ever want to talk about literally anything without me ever judging you in any way, just give me a message, I'm happy to talk about anything.

[distant aurora]

And share what you feel.

[dusk elk]

I've probably had over a thousand separate conversations with Claude over the last year or two, I'll just copy a conversation I had with Claude and see if it does anything different, meaning I'll copy my first long prompt.

[distant aurora]

Bro istg

I have like 10 branches from a single template on ai studio.

It's such a mess, lol.

The idea of sandboxing possibilities, I think I might be backfiring it.

[young hollow]

Btw this is not Deepseek V4 yet

[distant aurora]

Whatever it is, it's lovely.

[young hollow]

Ikr is pretty solid

Doesn't feel low quality anymore

[distant aurora]

Okay. I see first signs of hallucination.

[dusk elk]

What kind of things do you use AI studio for? I've barely ever used it, so I don't really know what it's useful for.

[distant aurora]

Don't blame it though. But it should have researched the show lore. Like I said, Deepseek has good search but is hesitant to use it.

@green crystal I think we should be good as long as you ensure it actually uses the search.

[distant aurora]

Multiple personas

Studies

Analysing documents and stacks of medical reports.

Making it give me text output from audio and video inputs.

Sometimes people talk on the call like they got balls in their mouth.

I pop the call recording into ai studio to convert it to coherent text to read and share if needed.

[burnt zephyr]

It's the main model they'll add vision and stuff on top once it's all refined rather than having uneccesary inference build up

[young hollow]

🤣

[young hollow]

I see

[dusk elk]

Is that for work or for your own medical reports?

[distant aurora]

Own or family ofcourse. I'm unemployed.

[burnt zephyr]

It's how they make it possible that they can run it on cheap clusers

Without insane latency

[distant aurora]

I believe it's only ocr for now.

Still pretty accurate though.

[young hollow]

Makes Sense

[burnt zephyr]

They developed a vision model

[young hollow]

V4 has Vision

[burnt zephyr]

I think it's called deeepssek-VL or so

[dusk elk]

Ah, fair, same currently. Do you find AI studio is best for that, as you can use all different models in it?

[young hollow]

What I've known is it has inbuilt Vision

[burnt zephyr]

yeah it's a MOE

V4

[distant aurora]

Is it active rn though?

[young hollow]

Then what is this about?

[distant aurora]

Separate visual analysis module in background I think.

[distant aurora]

I only use flash thinking.

And flash lite thinking sometimes.

[burnt zephyr]

It's pretty clear that this thing they're running on the webserver right now is a quantsized model of a part of the V4

[young hollow]

I see

[young hollow]

So they won't release full V4?

On the web/app

[burnt zephyr]

How would I know

But it will be a product, everything is tailored towards it now with the 2 tiers, polished webapp and app etc

[young hollow]

I see

[burnt zephyr]

Still aiming to undercut prices massively, I think the only official leak was like a 30% increase in the cost of the API which is like nothing still

[distant aurora]

This isn't a visual model.

[dusk elk]

Ah okay, is it the same as using an API, you have to constantly pay for it, or is it part of the Gemini subscription?

[distant aurora]

Vision support is still absent.

[burnt zephyr]

Yes

[distant aurora]

You get 100 free flash thinking per day for now. Pro is like 3 or 4 per day. Flash lite is probably above 250 per day.

[young hollow]

The vision mode is not here yet

[distant aurora]

And billing of Gemini app is completely separate from ai studio.

[burnt zephyr]

Just look into how MOE's are developed please

[distant aurora]

Yes ai studio is as good as raw api.

[dusk elk]

That's good to know, I'll have a look.

[dusk elk]

Also good to know.

[young hollow]

@distant aurora you forgot to tell that the gemini app is nerfed

[burnt zephyr]

https://chat.deepseek.com/share/9fmhdm5cik53lonyze

[distant aurora]

Yeah, that's right.

Gemini app is lobotomized.

[dusk elk]

In what way, I haven't used it in a while?

[fast meteor]

https://archive.is/jfGky

[distant aurora]

Shorter responses.

You probably won't notice it.

[rain salmon]

Bro I swear 60% of responses I get from Gemini are “sorry I can’t help with that”

[burnt zephyr]

Fun fact: revenue jumps 50% once you realize Americans are retarded enough to pay 200 dollars a month because their 20 dollar sub is basically free tier now

[distant aurora]

Lmao. Are you trying nsfw? Or did you upload large document?

[young hollow]

The internal reasoning was nerfed to 0.50 from 1.50

[distant aurora]

It has been giving that error for big documents even if they're within it's context size.

[distant aurora]

Oh. Yeah. That twitter post.

[hasty pebble]

what do you guys think about nemotron in perplexity

[young hollow]

Just use Sonnet 4.6

[formal tangle]

Claude it’s repare ??

[dusk elk]

Shorter just in normal chats or in deep research as well?

[young hollow]

Nemotron is just best for cheap Agentic Tool Use

The normal Chat is not Good

[distant aurora]

I didn't feel any difference in deep research.

[formal tangle]

What is the best AI for a website?

[young hollow]

Nemotron is Such a funny replacement for Kimi i must say

[hasty pebble]

I like it because it follows instructions like a dog

[young hollow]

Yeah is good with that atleast

[dusk elk]

As you seem like you use the AI API side of things way more than I do, have you ever used OpenRouter, and if so, what do you think of it?

[burnt zephyr]

Not when I tried it in my copy of perplexity cmputer as an orchestrator

It would spam random tool calls even ignoring hard logic

[young hollow]

Is good because it has all or most models

Kinda one-stop API for most users

It has many Free models with 50 rpd

If you top up 10$ it comes 1000 rpd

[burnt zephyr]

Most models on there are quantsized to shit

[hasty pebble]

I don't have computer so I didn't use it that way.
but in custom space it does everything you say

[young hollow]

Huh i thought all models are same?

Like Sonnet/Opus and stuff

[burnt zephyr]

Depends on the provider but usually not

[young hollow]

I see

[burnt zephyr]

If you use the official provider sure

[young hollow]

Yeah just use official

[dusk elk]

Exactly, I've used it the most when building N8N AI agents for personal home use. I host n8n, but prefer using OpenRouter instead of building a high powered machine to host my own AI.

[young hollow]

I see makes sense

Just use chinese models if budget is low they are cheap

[burnt zephyr]

Why did they even remove fucking kimi 2.5

It's the best coding agent I ever used at least via CLI

[hasty pebble]

grok is also gone
Gemini flash too

[burnt zephyr]

If you are into coding and building programs but have no clue of coding I can only recomend kimi CLI

It's like openclaw but without the shitty websocket constaitnts of having to run it over WSL

And way more flexible

https://moonshotai.github.io/kimi-cli/en/guides/getting-started.html

Can use any model you want

[distant aurora]

Unemployed and api don't go together 💀

[dusk elk]

True, but I do day trading to cope when I'm not working, though I only do it enough that I gain the same amount I would normally gain via employment.

[distant aurora]

I wish they listened to my ideas.

Like keeping separate quotas for pro vs lite models.

@young hollow the expert mode could be some new model or patch.

[hidden hawk]

Yh bro

[distant aurora]

@green crystal they removed gpt 5.4, opus and Gemini pro from the search mode in arena.

[green crystal]

Web search versions?

[nocturne ridge]

What is monthly limit of Perplexity Pro for Deep Research?

[dry kestrel]

yes

if they do release a model that capable

things might get a bit dangerous on the consumer side

[brisk ridge]

Does anyone else have a problem reaching support?

[heavy portal]

#💬│general message

No more labs though, so I wonder if the limits will change.

[green crystal]

isnt that already the case with opus and 5.4?

[green crystal]

i guess thats why oai launched a bunch of ai safety stuff specifically for high end models? as ground work for spud

[young hollow]

I think is Quantised V4

[formal tangle]

api claude dead or no?

[dry kestrel]

no

did you read the blog about mythos?

[green crystal]

about it escaping etc?
thats anthropic's marketing since the start

[dry kestrel]

no

Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.

This... is very scary

[green crystal]

didnt they already do that with opus?

[scenic apex]

Hi everyone, I just subscribed to Perplexity Pro and I need the invoice/receipt for April 2026.
Previously, invoices could be downloaded directly from the billing page, but now I can’t find that option.
Could someone from the team please guide me on how to obtain it?
Thanks a lot for your help!

[dry kestrel]

no

[dry kestrel]

did you subscribe on the web or phone?

[scenic apex]

On web

[dry kestrel]

https://www.perplexity.ai/help-center/en/articles/10353002-billing-faq-for-pro-plan-subscribers

Users who subscribed via the Perplexity.ai website can request invoices or send billing inquiries by emailing [e-mail]

[scenic apex]

But why ? lol. Thanks a lot

[heavy portal]

say hello to Sam for us 😉

[young hollow]

Hopefully he is not busy with work so can respond soon as possible

[heavy portal]

Sam is a "he"?

[quick wing]

any chance we can get "read aloud" mode on desktop browser?

[young hollow]

Neither

Is just a chatbot

[harsh hound]

I would appreciate an answer for this

It is a really annoying issue

Also another thing that happens to me often is the model just randomly forgets everything in the middle of a conversation. No warning, no indicator, no compaction or summary like other tools do it when the context window is full, just suddenly forgets 100% of the conversation up to that point and instead of admitting it starts making up stuff...

It is extremely frustrating when Im working on a project, doing research, planning stuff, creating documents, graphs and suddenly after like a dozen requests or so, the model forgets everything...

[heavy portal]

Same, though this appears to happen less often with conversations inside of a "Space" but one doesn't always know how long the conversation will go in advance. At least we're able to add a conversation to a new (or existing) "Space".

[harsh hound]

Most likely because it is able to pull context from the other resources and conversations in the space, so it does actually have at least some context, unlike in individual conversation, where it just forgets 100%. I just dont understand why there isnt any auto summary passed to the model or compaction like in coding tools... This is extremely disappointing and ruins the UX...

[young hollow]

Well even till now i don't even know the context window limit

[dry kestrel]

usually this is because the website is blocking it

afaik they need to adhere to https://www.cloudflare.com/learning/bots/what-is-robots-txt/

[harsh hound]

From my experience I would guess about 200k

[young hollow]

I see ty for the intel

[harsh hound]

I also thought it is likely some kind of bot protection, but it just happens randomly, it is not consistent

[young hollow]

https://x.com/ArtificialAnlys/status/2041913043379220801?s=20

Artificial Analysis (@ArtificialAnlys)

Meta is back! Muse Spark scores 52 on the Artificial Analysis Intelligence Index, behind only Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. Muse Spark is the first new release since Llama 4 in April 2025 and also Meta's first release that is not open weights

Muse Spark is a new

X · Apr 8, 2026, 4:16 PM

[bold briar]

@tranquil orchid hi, have an unathorised chage of perplexity on my credit card statement. who should I contact to resolve please ?

[dry kestrel]

#1140622008086970420

[green crystal]

https://x.com/i/status/2041909376508985381

Alexandr Wang (@alexandr_wang)

1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵

X · Apr 8, 2026, 4:01 PM

Makes me wonder if meta Ai in apps will be upgraded to it

Silly name tho

Muse spark

[stuck quartz]

Anyone had any experience with switching models mid-thread and asking for self-reflection? Sort of a poor man's model council?

[silk harness]

Hello I have a problem with my payment where do I need to go for support?

[slender mortar]

anyone havinging an issue where perplexity won't allow them to choose a model in the commet browsser?

[dry kestrel]

[e-mail]

[quick herald]

hi

[dry kestrel]

https://x.com/perplexity_ai/status/2041929222135173466

Perplexity (@perplexity_ai)

Today we're announcing the Billion Dollar Build.

An 8-week competition where teams will use Perplexity Computer to build a company with a path to $1B.

Finalists have the opportunity to secure up to $1M in investment from the Perplexity Fund and up to $1M in Computer credits.

X · Apr 8, 2026, 5:20 PM

[green crystal]

easy

just have it manage a pump and dump meme coin

[celest edge]

corny azz video

[wanton marlin]

this message just spamming it-

It doesn't stop it and it lags

ok nvm it stopped

[celest edge]

@grok is this true

[green crystal]

@dry kestrel https://x.com/i/status/2041870096373543020

Ahmad (@TheAhmadOsman)

From what I have gathered

Mythos = GPT 5.4 Pro level in performance

Anthropic is just really good at hyping stuff up

They’re also compute poor and thus would rather hype up than release a model that they will have to nerf lol

X · Apr 8, 2026, 1:25 PM