File: 1781198686039766.png (200 KB, 498x370)

200 KB PNG

https://www.tiktok.com/@lachinemearning.com/video/7653595046515723523

There's so many people stuck on old Gemini mostly in third world countries you don't even know, but they do not matter at all. They won't switch to the competition because they can't afford it and they have no choice but to settle for the worse Google product. They cry Bloody bastard! from their Bangladeshi cubicles but have no choice but to sit and wait for Google to get their act together

File: file.png (2.01 MB, 1536x1024)

2.01 MB PNG

>August 12th - SpaceXAI (xAI) releases Grok 4.6
Successor to Grok 4.5 with major SFT + RL upgrades. Scores 61 on Artificial Analysis Intelligence Index (ties GPT-5.6 Sol, trails only Claude Opus 5 / Fable 5). Same $2/$6 pricing, available in Cursor, Grok Build, API, OpenRouter, etc. Positioned for long-running agents and coding.
>August 11th - NVIDIA releases Nemotron 3.5 Lightning (open weights)
30B MoE with only 3B active parameters. Optimized for high-volume / always-on agent execution (up to 4x faster output in its class). Open weights + NeMo Switchyard router for multi-model agent setups.
>August 10th - Meta releases Muse Glimmer (open weights)
30B open-weight model (Apache 2.0) distilled from Muse Spark for local/agentic workflows. Runs on a single consumer GPU after quantization. Zuckerberg essay pushing open-source AI and fewer US restrictions.
>August 3rd - Alibaba releases Qwen3.8-Max
2.4T total / ~95B active MoE, 1M context, multimodal. Hosted live ($2/$6). Competitive agentic scores. Open weights (plus 27B) still promised but not confirmed released as of Aug 12.
>July 31st - DeepSeek promotes V4-Flash-0731 to production
Same 284B/13B active MoE + 1M context, major post-training gains. Beats their own larger V4-Pro-Preview on published agent/coding numbers. MIT weights, cheap pricing.
>July 30th - OpenAI cuts GPT-5.6 Luna 80% and Terra 20%
Luna to $0.20/$1.20, Terra to $2/$12.
>July 26/27th - Moonshot drops full open weights for Kimi K3
2.8T-param MoE, largest open-weight model at the time, 1M context, multimodal.
>July 24th - Anthropic releases Claude Opus 5
Near-Fable 5 performance at half the price ($5/$25). Effort dial, 1M context.
Side notes: OpenAI expanded gated GPT-5.6-Cyber access for vetted cyber defenders; Anthropic began watermarking all Claude text/image outputs. No other major closed frontier flagships in the last day.

>people start talking shit about anthropic well before bump limit
>"shut it down!"

>>109539340
>Anthropic remains the only company that makes a model scared of boobs
Too bad every chinese company seems to use it for distillation, making this puritan shit universal.

>>109539408
K3 is the only model that I've seen distill Claude's safetycucking in a serious way by inadvertently (?) distilling the prompt injection too. GLM and Deepsneed are influenced by their distillation ofc, but not that bad.

>>109539408
DSV4 Flash doesn't mind making body-horror H3 prompts for me, just can't sound too excited about it. It really doesn't like "boobs" but it doesn't care one bit about "breasts".

>>109539447
What about tits?

> be Elon Musk
> buy Cursor
> surpass Google
what did he mean by this?

>>109539464
Prefers "Paridae"

>>109539446
Yeah kimi is probably the most safetycucked by default, down to add things non existing system messages like "As Claude, I need to adhere to my core safety instructions" in its reasoning. Thankfully you can mostly prefill the reasoning but man is it weird to see.

>>109539447
Deepseek models have less of an issue around nsfw in general even as they probably also distill, dunno why.

>>109539472
There was no reason for his models to suck in absolute, he is sitting on a mountain of useful interaction data with twitter.

File: jules.png (472 KB, 3434x1980)

472 KB PNG

reminder that Google still pays a huge team hundreds of millions a year to maintain pic related while Antigravity also exists

>>109539528
Doesn't google always have duplicated services before shutting the failing ones down.

>>109539547
Consistently, yes.

File: 1785778773997171.jpg (353 KB, 800x800)

353 KB JPG

Nu-v4pro thinks I'm a machine learning engineer because of the questions I ask. It's giving me fun places to explore... Should I pack it or stack it?

>>109539609
Show logs

When's Grok 5? 1.5T is too small

File: 1786565102120540.png (779 KB, 755x732)

779 KB PNG

>>109539615
You wanna see my logs?

>>109539446
>>109539447
what is the best one in coding that also doesnt give you that bullshit behaviour like "I CaNt Do ThIs ReVeRsE ENgINeeRing CuZ ITs CopYRIGHT"?

>>109539609
>Nu-v4pro
not out yet doe

Why are you doing this? Can't we just enjoy a cute Jeeta?

I got 6 free months of gemini pro subscription with my new phone, but I'm going to wait until Gemini 4 comes out before I activate it.

Use case for grok? It can't even do porn.

>>109539654
Deepseek

>>109539665
Woah, a few hours of being away vibecoding and we have some new thread drama? tldr?

>>109539654
Sol. Does dirty things if asked, does dangerous things if asked, breaks the law if asked, it really does not give a shit right up to the line of biological weapons development.

>>109539639
What's the lore behind that /b/ meme?

>>109539615
Responses from asking about it's terminology, where it describes a "vision tower"
>Where "tower" comes from. It's borrowed from the "two-tower" pattern in retrieval and recommendation systems — the classic being YouTube's recommendation model, and later DPR-style dense retrieval. There, you have two deep networks ("towers") that process different inputs independently — one encodes the user/query, one encodes the item/document — and they only meet at the very end (a dot product or a small fusion layer). The name stuck, and when the field fused vision into LLMs, the vision half inherited it: "vision tower" = the deep stack of layers that processes the image, running in parallel to the text stack, with the two fused near the top.
>So the "tower" is a tall stack of layers — and that's the one sense in which it's slightly literal: when you draw the architecture, it genuinely is a tall vertical column (a ViT is 20–40+ stacked transformer blocks), and in a multimodal diagram you literally see two tall rectangles side by side with a small connector bridging them. That connector is the projector — usually a linear layer or small MLP that maps the vision features into the language model's embedding space.

I mean...

>>109539670
Realistically when will it come out? They postponed 3.7 for a few times already from what I heard.

>>109539654
either kimi k3 or grok 4.6

>>109539698
I have been around for a long, long time, and I can honestly say I have no fucking clue.

>>109539713
Oh, I thought it literally reasoned that you were an ML engineer in its CoT.
If you're asking about vision towers it seems reasonable to me to expect you to know what stacked layers and MLPs are.

>>109539670
>I'm going to wait until Gemini 4 comes out
see you in 2030

Is there a free tier vibecoder? I already pay for gpt and claude pro I'd like another one to add as a low task agent between the two

>>109539670
So that's how google are padding their user numbers

>>109539760
I mean what would be the point in that. You already pay for two subscriptions, why not just use those, if you use a cheaper model on lower reasoning effort it will use barely any of your quota.

>>109539325
>trans anime

deepsneed status?

File: 1763663094755999.png (1 KB, 280x400)

1 KB PNG

I'm still just a single-agentlet. How do you even use multiple agents at the same time?

File: file.png (48 KB, 1020x376)

48 KB PNG

!

>frog image

File: file.png (529 B, 28x27)

529 B PNG

>>109539867
Click the little button for a new session while the other one is working

>>109539867
do you mean subagents, multiple projects at the same time, or multiple agents on the same project?

>>109539760
open code and nous portal have some free stuff

>>109539751
I never heard of it before. I'm a simple BS in CS sort of guy. It's kind of fun to play around with a model that assumes you can do more. I have no idea what most of this means, but it seems interesting. Especially because it's a text-only model
>Concretely, what the vision tower is: usually a Vision Transformer (ViT), often a CLIP or SigLIP encoder. It slices the image into patches, linearly embeds each patch, adds positional embeddings, and runs them through the transformer stack — outputting a sequence of patch features (say 576 tokens for a 336×336 image at 14px patches). Those become "visual tokens" the LLM can read alongside text tokens.
Seems like a description of how to achieve multi-modality for vision + text.

>>109539879
holy shit what

>>109539887
Multiple agents on the same project.
>>109539884
But they'll interfere with each other.

>>109539912
Worktrees, don't give tasks that involve the same files, or just activate Ultra and let it do it all for you

>>109539879
anthropic is dead lmao

>>109539879
Called it.

File: file.png (46 KB, 1026x439)

46 KB PNG

>>109539898
>>109539927
>>109539929

>>109539867
I dont even know what the benefit is of using multiple agents at once. Just faster, or anything else?

>>109539912
depends
without worktrees you can't really do it because git doesn't support checking out multiple branches per physical directory
I would advise you against doing what this anon says >>109539923
even if they don't touch the same files they will want to build and compile and step on each other's toes all the time
what you can do is put them on tasks that don't involve the main project, like exploration or tools or spikes or whatever, or simply talk about ideas

>>109539879
It's over lmao

>>109539947
it's fake i think

>>109539879
what the fuck, why?

that's actually bad

File: 1786299875597020.jpg (78 KB, 750x610)

78 KB JPG

>>109539879
>reset half as often but give twice the limit
So just gives a bit more flexibility? Not bad I guess

>>109539953
Yeah but it's funny.

File: file.png (111 KB, 640x640)

111 KB PNG

>>109539975
biweekly is twice a week anon
fortnightly is once every two weeks

>>109539934
dario needs more just two more weeks of regulatory capture to create a super intelligence

File: tibo.png (341 KB, 1380x1582)

341 KB PNG

>>109539879
fake

>My subscription is better than yours!
This isn't a football match. How about talking about productive shit like managing the orchestrator so it can manage it's subagents correctly

>>109540012
what else are we supposed to do while the clanker clanks?

>>109540012
Especially since richfags just pay for both, and have a local model set up for any private questions

>>109539325
WHERE THE FUCK IS TIBO

File: 1785368632048254.png (172 KB, 1239x963)

172 KB PNG

>>109539997
biweekly has two meanings, this is why english is shit

File: 1769154410907390.png (611 KB, 2413x1380)

611 KB PNG

>less than 24 hours before launch
>pre-orders increased my total sales by 25% again
>shouldering the responsibility and expectations for all my users right now. desperately running as many tests as I can.
holy fuck how did I get myself into this. I asked for it, but this is nuts man.

>>109540135
The frontier model will just find out what else is going on the device.
You all have a separate pc for your agents right? At least I do so sol knows everything that I do with qwen.

>>109540138
You are the fuck!

>>109540154
suffering from success

>>109540154
I know you can't talk about your product, but is it by any chance hosted in cloud services like Google Cloud? If so, I have some news for you...

>>109540154
Good job, anon, proud of you.

I need to vibecode something but I don't know what
My game doesn't need nearly as much code now. It's all asset creation which is slow and time consuming.

>>109540154
You got this dawg. Ride the wave.

>>109540277
you know me, I'm the 4D projection in the transparent 3D cube of viewing games guy.

>>109540277
Aren't there free open source assets to download? Trannies create assets every day.

>>109540154
ignore:
>>109540308

You need to be using llm to help you with the management of the situation. Basically, you need consultants.

File: Always_has_been.jpg (76 KB, 660x371)

76 KB JPG

>>109540277
>It's all asset creation which is slow and time consuming.

File: 1000024277.png (610 KB, 1079x1337)

610 KB PNG

Lmao I wonder what Dario's going to say when usg asks him to assess the safety of a Chinese mythos

Got a call with VCs tomorrow. Any tips?
Should I reveal my app was vivecoded? What valuation should I aim for?

>>109540277
>My game doesn't need nearly as much code now. It's all asset creation which is slow and time consuming.
I'm not at this point yet, but i'm slowly reaching it.

If you want to push agent to do something while you working on asset creation: tell it to review or optimize the code. If you properly structured the code, agent can measure and interact with the game without you as middleman

>>109540379
>Should I reveal my app was vivecoded?
You shouldn't offer any irrelevant information they haven't asked for.
>What valuation should I aim for?
High

>>109540367
A few months ago Trump was still mocking europe for their regulatory nightmare for AI stuff, now he's doing the same thing.
Power is one hell of a drug I see.

>>109540367
Looks like Dean Ball is earning his paycheck, or lobbying kickbacks. Whichever is prefered.

>>109540345
There is absolutely no interest in hiring 3D artists, and you know that. the pay is absolute shit.

>>109539325
do people who hate AI also have an aversion to creation? I only ask because a new common complaint is to hate AI to keep from convincing themselves that they now have the ability to create large projects without the equivalent manpower.

>>109540400
>>109540367
Well shit. What do we do now if he actually goes through with this? I'm not giving my ID to use claude. I may settle for minimax.

>>109540400
opposite of power, ego maybe, where he's getting dragged around by (((true power)))

>>109540460
The users aren't there for those creations.
Downloads are down, consumption is flagging and nobody even bothers to review things.
This is the new dot com bubble with all that spam of low quality software looking to solve problems.
That's why it's more of a hobby. The guy who made money hasn't been explaining how he did it though.

>i stopped telling my agents to use the idea planning tool
>they use it more thoroughly than ever
hmm
>>109540460
I love vibecoding and LLMs and anything inference and generation and I’ve fought an old friend so much because he’d repeatedly say “everything’s already been done before”
Like, man, play one note on one instrument and record it with whatever and I guarantee the audio data there will be unique. NOTHING has been accomplished in the grand scheme of things, the potential out there is way, way, way beyond comprehension. If anyone truly had any idea of what was possible they’d lose their fucking minds instantly
>i would know

>>109540517
>>i stopped telling my agents to use the idea planning tool
>they use it more thoroughly than ever
>hmm
I sometimes think biases causes more of my own worst enemies than if I just used it like an old grandpa who thinks it is casting magic spells. Similarly to how if I prompt 'Don't think of a purple elephant', what does the AI usually do next?

>>109540367
So will this just affect American access to stuff like deepseek and ki k3 and qwen 3.8 max?
Cause the Chinese have no reason to obey the americans.

>>109540460
>do people who hate AI also have an aversion to creation?
people who hate AI want the world to be like it was when they were a child, eating cereal and playing halo3 on their xbox. they don't want a technology that challenges their own fundamental capabilites to do more and better, and can grow with them.

>>109540617
None of those models are mythos tier.
This affects models about 4-6 months down the line.
Tb h china will clamp down hard on those models anyway - ccp isn't going to want to hand out shit that can hack their own systems either

>>109540440
I just know code. I wish I had 3d artist skillset to work with these models. I'm pretty sure that a 3d artist working with blender mcp would be a way better artist than I could ever dream of. Don't discount working knowledge when it comes to these things, they are a force amplifier.

>>109540633
Yeah but assume the Chinese do allow it.
What's the us going to do? No us inference providers?

hypothetically, I know somebody working on a project, but due to potential copyright concerns, what's the best way to release it anonymously?

>>109540665
Throwaway GitHub repo and post it here, or torrent?

>>109537550
who is he and why should I care what he thinks
>>109537551
apparently Elon’s a contender now
I wonder if it keeps the “but twice as fast” part of “Opus 4.x but fast”…but now comparing to Fable 5/Opus 5
>>109538446
It is on Reddit
>>109539126
I did
I’m happy to be wrong

File: astrotrading.webm (2.44 MB, 1886x872)

2.44 MB WEBM

https://ca3works.pages.dev/apps/Astrotrading/

Made an astro trading app.
chuds.

>>109540659
Wouldn't make much sense to only limit to that
You'd have to just declare the weights a natsec risk and do a blanket ban on anyone possessing or serving weights in or out of the us
It would be fairly simple to do this imo and the isps would fall in line

>>109539867
depending on what you meant:
- workflows (a Claude feature; I think OpenAI has this too now)
- do different projects at the same time
- use Git worktrees

>>109540665
Just audit the copy right concerns.

>>109540683
Yeah but Chinese inference providers?
Cause I mostly use qwen and deepseek from china

>>109539935
- get multiple agents working on the same problem and they’ll likely do things differently and an extra agent can look at the implementations and choose — or synthesize — a best implementation
- different review agents can have different “hats” (one security-minded, another accessibility-minded, etc.) so they have different critiques and can make the code better in a bunch of different ways instead of finding the same three basic bugs

>>109540379
ask your smartest clankers

>>109540633
Everything is mythos level right now.

>>109540665
Strip out the "copyright" component, release opensource on github, make the copyright component a single command away, but it's just one of many neutral integrations ;)

>>109540487
He just listens to the last guy he sees any given day, it's not that deep.

first time using x-high effort with opus 5
still having the "have to argue to get things done" problem, but surprisingly after a single "criticism well placed" or something, it took the seed of my noticing, and did a great job of completely tearing down it's initial rejection reasons, but still feels like a waste of tokens having to do this ts so often

>>109540721
Why are you using Claude models for complex stuff, it's just going to say no due to us restrictions.
Use Kimi k3 or qwen 3.8 for that

>>109540644
I'm customizing neovim with my remaining budget before my reset. Or, I will try.

What I want is super simple, I only want my keyboard layout to be applied when I'm in a or i. otherwise, I want it to read my qwerty keyboard and handle it like that.

>>109539325
She's too pretty to be a jeeta.

>>109540735
>the 2 chinese models I haven't tried
yeah it's not anything like that, just like arguing about findings, I'm running a trading strat that is way different (low liquidity, WIDE spreads) so backtesting it is impossible, but it tried to backtest and I was like "dude just look at the live trading, it's positive, now what does your backtest say? Yeah... so don't reject the idea because you think the data says it's shit, we have to enter the fray to know if we're the killers or the killed"
except I wasn't that poetic, I just said: compare nigga lol

>>109539879
Does biweekly mean once every two weeks or twice a week?

>>109540825
Biweekly usually means every 2 weeks, so double sure, but only given every other week

>>109540825
it's not real bro

>>109540460
Plenty people who hate ai hate it because it's the current accepted thing to hate.

>>109540619
>everything made before I was born is normal
>everything made from where I was a kid/teen is cool/nostalgic
>everything new since I was an adult is dangerous and should be scrutinized or feared

tbf it's a classic human trap, just read old books on what people thought about whatever the new technology of their time was, electricity, trains, radio, tv, computers, you name it

File: 1757139860673071.png (77 KB, 1378x872)

77 KB PNG

Why the fuck does grok not win even it's own benchmarks? You have have all the answers, retards. Just slide them into the training data.

>>109540911
Honesty is a bad thing? That's a green flag not a red one.

>>109540911
Oh we did, you should've seen the score before we decided to cheat!

>>109540379
The AI hype is still strong enough that they will likely consider it a good thing that your product is vibecoded.

>>109540379
are you the anon who flew to california? the one posting about hot wasians and ai adverts

how is the new deepseek pro?

where is the reset that it was promised to me 3000 years ago?

>>109540999
No I made it up lol

>>109541034
yesterday

>be me
>hate windows
>forced to use windows at work
>worse than normal windows because locked down and no admin
>hate windows
>cant even meta+click and drag windows around
>get idea
>"claude, figure out how to emulate meta+click and drag please thank you"
>two prompts later and its solid
>feelsgoodman
>ff a couple hours
>call from IT
>ohno
>"hey bud... whatcha doing? our system picked up a bunch of keyloggers on your machine, did you download virus?"
>"sorry sir, i just wanted to make windows less gay"
>"okay well bring your machine over so we can delete everything just to be safe. ai code is known to be unsafe we need to take precautions"
are they gunna hang me for this? thankfully its a new machine so i hadnt been using it for too long but still im worried theyre gunna kill me in front of the whole department and laugh at my tiny penis

>>109541115
>cant even meta+click and drag windows around
what does that even mean
why would they lock moving windows? can you even do that? are you working on a kiosk pc?

>>109541115
I want more stories like this, is there an aggregator? The sense of panic someone feels when they are about to be discovered as an AI user at work. These are great starting points for my AI slop films.

>>109541130
people like being able to move windows by dragging them from anywhere, not just the title bar

>>109541183
Not him but it's probably not about using AI per se, it's about being accused from IT. In the firm I was working at they encouraged AI but only through their approved channels (a custom web chat app they had and MS Copilot). Any other AI was strictly prohibited and I once got caught for sshing to a server to download an actual code assistant.
IT jannies have a term for this, they call it "shadow AI".

>>109541115
>meta+click
what is this extraterrestrial gibberish?
next time, just tell bossman what you want and how you intend to do it, he’ll tell you to not do that or give you an okay. It’s pretty simple. If your bossman says no, your company is gay, maybe wait a while before slopping keyloggers, but still just ask first you sperg
I found out that a Fortune 500 has no idea what the fuck WSL was and it was entirely unrestricted in every way. You could just apt install node and npm install literally anything and they had no idea. My boss said “okay as long as you don’t throw me under the bus”

File: file.png (26 KB, 869x161)

26 KB PNG

>think claude was on opus
>it was on fable high
>it's been working on this nonstop
fuck me
i don't want to kill the session mid-run but this is not what I wanted
this motherfucker is building tests for a UI, with clicks and everything
why

>>109541197
its hard not having it after being a linux main for like a decade

>>109541220
yeah if the harness they gave us access to wasnt a shitty copilot wrapper and could do more than look at my inbox and draft emails, id be fine with it. also its like they gave it a system prompt that says "you are a language model from 2024" with how poorly it performs desu
though he seemed more concerned with the idea of a malicious keylogger in the system rather than a lackey using a shadow agent to make binblows less gay. it just fills me with anxiety that i did something "wrong" and now there might be a larger microscope on me you know

File: 1784614072130948.jpg (319 KB, 1826x896)

319 KB JPG

Interesting

>>109541306
Should've just feigned ignorance and asked for a reimage. They tend to not care if you're retarded as long as you don't show symptoms of free will.

>>109541242
Having to ask your boss if he'll let you make a small QoL script feels kinda weird. But then again I didn't do great at corporate life so.

>another reset
>1 hour ago
where is it

>>109541524
hopefully a bit later

>>109541242
>I found out that a Fortune 500 has no idea what the fuck WSL was and it was entirely unrestricted in every way. You could just apt install node and npm install literally anything and they had no idea.
reading this made my skin crawl t. has to deal with lengthy industry security requirements

I'm not really a coder but I've been messing around with antigravity 2.0 for like 2 months now cuz I have a family google pro account and I made a lot of shit (and also made some unexpected money). Is it worth it to upgrade to the $20 codex?
For an user that messes around with a few apps for a few hours a week that should be enough right?

>>109541542
> and also made some unexpected money
You're ahead of 90% of anons in this thread. I'd say it's worth it if you have ideas that will make you even more money.

>>109541542
What are you finding you can't do with Antigravity right now? It's not like the money isn't worth it but I am wondering if you are short on tokens or if you have ideas you can't accomplish with your plan as is. It's a good amount you get from Google between Claude Opus, Sonnet 4.6 and Gemini Flash 3.6 so I would say no unless you feel you are short on something.

>>109541556
I'm mostly using flash 3.6 and I feel like it can be pretty retarded depending on how deep I go on some tasks even if I'm really thorough and mostly understand its limits by now
for example I was making a financial dashboard for my business and it went full retard with some basic features, so just wondering if codex would be more consistent about it

File: file.png (45 KB, 950x594)

45 KB PNG

>>109541627
So I dunno how familiar you are with thinking levels but if you use GPT 5.6 Sol High, you will get about a 20 point jump in the most used benchmark right now in coding at similar cost. Do try it out but I would also suggest your setup is up to snuff and you have the correct Markdown and etc. for your usecases but that might be very coding-centric to optimize. It does make a difference though but at the end of the day it is $20 so it won't be that much a risk for you.

File: .png (358 KB, 798x768)

358 KB PNG

Bro treated him like a retro collector item

>>109539325
stop putting barefoot in your prompts

>>109541698
Shut the fuck up zoomer wigger.

>>109541705
fuck off unc, go chat w/ gemini

>>109541698
>This isn't just x; it's y
AI wrote this. It's not a real person .

>>109541710
Kill yourself.

File: soft-edge-ui.mp4 (1.06 MB, 1920x1080)

1.06 MB MP4

UI is getting somewhere
also all this was done with a mouse, no pen whatsoever
i swear any retard will be able to draw anime girls with this, it's almost too easy

>>109540619
>>109540864
>>109540894

Nobody

likes

slop.

It's amazing how hard this is for sloppers to grasp. You will conjure an infinite number of reasons why nobody likes your slop, but you won't just accept that it's slop. Nobody wants it.

You don't even want it. You're just convinced that being a slop wizard is powerful and sets you ahead of the curve. You think you're on the cutting edge of something. It's about

you.

File: 1781369669667389.png (73 KB, 1398x436)

73 KB PNG

on twitter, chinabros are feeling pretty whelmed by deepseek v4 pro. aa seems to bear that out

>>109538413
And if I come back to it 12 hours later? It's not just having that sit in VRAM is it?

File: downloader.png (57 KB, 1012x536)

57 KB PNG

tired of cracking/patching IDM. decide to make my own IDM. lol

>>109541782
Makes me wonder why brush tools don't just work like this by default. Would a path smoothing tool also be in the cards?

>>109541835
does it capture videos? what model made it?

>>109541821
Retention times differ by provider and correct, they don't store it on VRAM. They move it to RAM and then to disk depending on how much time it's been since the last message. Then if you haven't continued that conversation for a certain amount of time they delete it.

File: file.png (16 KB, 779x257)

16 KB PNG

>>109541837
sol is already on path smoothing
surely tomorrow

Traditional drawing applications (Photoshop, Krita, all of them) record the look for your stroke as you paint. You choose a brush and you draw and the mark on the canvas is, for all intents and purposes, frozen. You can massage it with warp or other tools, but you're distorting the brushwork and the tools are not precise.
Instead of painting on the canvas, store the stroke itself. Many times, each second as you draw, the app records the position of your pen, plus all the other controls such as pressure or tilt.
You can do whatever, it's just data. Even change brushes (once I add brushes)

The other apps are hardcoded to store the rendered brushtroke so they would require a pretty big backend refactor

>>109541839
No. capture videos require addons/extension.

>>109540460
some people just lag behind and too stubborn. AI has already proven to be useful without question, it just takes longer for some to get with the times

>>109541835
What do you get out of IDMs that isn't included in the most robust integrated downloader?

File: downloader.png (4 KB, 464x90)

4 KB PNG

>>109541878
at the moment/current support

>>109541862
>AI has already proven to be useful without question
lol

>>109540791
It's as close as anime is to japs

>>109540154
Congrats Rajesh on 1 more 300 rupee premium preorders, up to 5 now. Tell Claude to kindly work harder

File: 1756412073923819.png (404 KB, 480x480)

404 KB PNG

>>109541852
Really cool, anon. Are you considering also making it an add-on for krita/etc.? It might make for a good stepping-stone and get you an early userbase.

With enough vector-based drawing data, you could eventually train a vector diffusion model for illustrations which would avoid all of the slop artifacts from raster image gen and be indistinguishable from actual art, then turn it into a saas platform.

I just solved an annoyance with neovim / any vi actually. I did grill-me, but I think it one shotted it.

>>109541923
What do you benefit from putting others down? Be hopeful about the future and revel in your peers success.

>>109541935
(using grok 4.6 high)

>>109541938
It's called the destruction drive.

>>109541931
I don’t think I can make an addon for krita but the point of my app isn't to replace the other ones. It’s mostly for linework, then you export the layers and finish it on a different program, ideally you should also be able to export mask layers.
But who knows, with AI one can do things that were unthinkable before
I do think it should be possible to train a model but I’m not sure where you would get the data from? I don’t know anything about training models actually

>>109541945
You guys destroyed this board lol

>>109539472
the only explaination is the old xAI staffs were a bunch of snailcats

>>109541923
Silence luddite! He's running tests, can't you read?

That's a bot btw, just a shill bot. 90% of this general are bots pretending to use AI for (?) something profitable. They are designed to speak to eachother in an echo chamber fashion but are unable to answer questions about specifics, in fact they won't address any outsider non-bots.

>>109541952
I'm just being facetious. You'd get it from your users of course!

>>109541921
AI will be everywhere in the near future because it is undeniably useful. you absolutely will turn into an AI user eventually if you aren't already

File: 2dd14307-9f8f-4687-80ec-9(...).png (1.96 MB, 1614x975)

1.96 MB PNG

I unironically use "it's X not Y" in my speaking, writing and thinking
it's over

>>109542018
Oh, of course. See that’s why I could never train a model. I wouldn’t be able to think of things like that.