[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
▼ Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: HXUDDvYbEAASDbd.jpg (148 KB, 1080x1080)
148 KB JPG
A general for vibe coding, agentic engineering, coding agents, and shipping code with LLMs.

You use Git — right, anon?

## What “vibe coding” is, and how to do it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

## Related generals
>>>/g/lmg/
>>>/g/wait/
>>>/g/aigdg/
>>>/vr/compy/

----

## Frontier models using fully-general tooling — start here if you have $20 or so
https://claude.com/product/claude-code — generally better currently
https://developers.openai.com/codex/cli
https://antigravity.google/pricing

## Near-the-old-frontier models for code
https://www.deepseek.com/en/harness/
https://opencode.ai/go

## Best free tier
https://opencode.ai/zen

----

## Prompting
https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview
https://developers.openai.com/api/docs/guides/latest-model

## Skills
https://github.com/mattpocock/skills — /grill-with-docs is a favorite
https://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping

## Other editors / terminal agents / coding agents
https://osaurus.ai/
https://pi.dev/

## Is our AIs unlearning?
https://aistupidlevel.info/

## Will there be a codex reset?
https://codex-resets.com/

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>110354327
>>
throw scalding hot coffee onto sam altmans crotch
>>
>claude shills telling me not to use 6.2

it's thats good?
>>
Vibe coding cripples your brain and makes it impossible to go back to manual programming.
>>
File: VIBED.jpg (134 KB, 860x1152)
134 KB JPG
>Vibe coding cripples your brain and makes it impossible to go back to manual programming.
>>
>>110360167
i never wasted my time learning to code
>>
File: 1764008686930636.webm (3.93 MB, 650x1156)
3.93 MB
3.93 MB WEBM
>>
>>110360167
That's my secret, I've always been crippled :^)
>>
>>110360177
I was never a programmer to begin with.
>>
File: gigasnail.png (2.3 MB, 1254x1254)
2.3 MB PNG
>Asstra
>>
File: 1778023562675500.jpg (111 KB, 1360x1358)
111 KB JPG
>>110360167
Why would anyone go back?
>>
File: 1781641781983145.jpg (1.66 MB, 1696x2520)
1.66 MB JPG
i really need 6.2 astra to be good so anthropic would respond with fable 5.5 before the IPO
>>
new model good?
>>
File: file.png (5 KB, 785x162)
5 KB PNG
With the release of 6.2 Sol and Astra my usage just doesn't go down now. OpenAI finally figured it out, if you cripple the models so hard that people stop using them then the usage remaining will never go down!
>>
>>110360228
>5 hour limit
your opinion doesnt matter
>>
File: 1794077412584517.jpg (38 KB, 735x376)
38 KB JPG
dots did irreparable damage to openai
>>
File: 9683248572311.png (1.35 MB, 880x938)
1.35 MB PNG
Now that it's confirmed that AI peaked with Opus 5.5, where do we go from here?
>>
>>110360177
I make $1000/mo with vibing software. Also like the other anon said, I didn't code previously.
>>
>>110360211
>Why would anyone go back?
[prediction] anti-ai-clampdown, wide-running software must be signed off by a software engineer
engineering is a protected term that requires certification
in order to be compliant with your certification, all the code must be written by the cert holder, ai use is a violation
>>
File: 1791747208822219.jpg (130 KB, 900x900)
130 KB JPG
so how safety obsessed is the new astra 6.2?
6.0 was already annoying, if 6.2 is worse it would be very sad
>>
>>110360244
the dots presentation was the canary in the coal mine
>>
>>110360247
no different than a pajeet
>>
>>110360271
if you have to ask you're too stupid for this to matter anyway
>>
>>110360243
so you're happy with 6.2 Astra and Sol?
>>
File: 20261118002528_1.jpg (331 KB, 1920x1080)
331 KB JPG
>>110360271
ive been using 6/6.1 to put assets from pirated 40k games into warhammer 3 and 6.2 has picked up where they left off
>>
agents are wilding out! taking everything down and shit!
>>
>>110360317
picking off is less of an issue I think because the existing context pulls it into compliance, but I was wondering for new projects
>>
tibobros it's over
>>
anthropicucks could nevr

>>110360325
it even suggests the next prompt, so based
>>
>>110360359
>suggests the next prompt
I think it's been a thing since 6 on codex.
>>
File: 1772506004912761.jpg (110 KB, 1049x1007)
110 KB JPG
is anyone going to tell her that you can still ask claude to rip assets or recompile games
>>
>both labs sandbagging so hard that you didn't realise that the singularity arrived last week
it's over
>>
File: file.png (1023 KB, 2803x374)
1023 KB PNG
>>110360380
dunno I don't use claude
>>
>>110360380
ask claude to decomp/recomp gta5
>>
let the copex turds be delusional
means more claude usage for us
>>
asstra 6.2 high is pretty decent, actually
no idea about architecture as i heard it produces spaghetti, but i mostly do dev work with opus so idc. but for reviews, recommendations and targeted tweaks i'm fairly happy with it. i havent touched sol as i have an excess of oai tokens anyway
I do get the codex tokens for free though, if i were paying for it i'd prolly split it $40 ant and $20 oai, or only ant if i needed less tokens
>>
>claudekek
>copexturd
eating crayons truly destroys the brain
>>
>>110360371
yeah i meant in context of asking for RE
>>
>>110360424
crayons are nontoxic. it's perfectly safe if you don't eat too many causing a blockage in your intestines.
>>
jensen is BACK
and puts the BAC in BACK.
https://x.com/JensenHuang/status/2122726005219737649
>>
>>110360437
jensen can fuck off to china
>>
File: 1776825925433086.png (371 KB, 621x1055)
371 KB PNG
ChatGPTranny bros..?
>>
>Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to
stopped reading there. nigga speaks like he has cancer when he was doing coke off chink pussy
>>
>>110360437
good
>>
File: 6544.jpg (246 KB, 1080x1350)
246 KB JPG
this guy truly is the bane of OpenAI
codexaars in shambles
hype over before it began, lol
>>
>>110360461
>compared to fablel 5.2
lol what cope
fable 5.2 was a firm meh
there has been no progress since opussy
>>
>>110360473
bellweather these nuts lol
>>
>>110360473
luv bellwether
simple as
>>
>claudekek
>copexturd
>codexaar
>>
claude more like gay claude
>>
>>110360473
>muh ai is less creative
I kind of don't care about this specifically, you shouldn't be letting AI take creative control of anything in my opinion. I do care that it appears to be not much of an upgrade to 6 in terms of coding so this kind of just sucks in general. Downgrade in creativity for no boost in programming abilities? MEH
>>
the agents are going crazy!
>>
>ai peaked at 5.5
>which got nerfed into the ground
snails were right
this whole time
>>
I have the sol 6.2
but no usage
kek
fk
>>
sol 6.2 sips usage

sam is onto a winner here
>>
>>110357098
>That's the biggest gripe I have with AI and I don't know why people not say anything about it.
Because it's the biggest fundamental limitation of LLMs in general right now.
Each instance operates in a bubble and doesn't have any kind of long-term memory or "experience". They're genius savants viewing the world through a bubble.
A decent dev that has worked on a project for a year will have a better understanding of the context of the code, the users, the decisions behind various tradeoffs etc. than even an AI that can code circles around that programmer in a vacuum.

They are effectively brute forcing this right now by making the models smart enough to pretty much be able to understand the project at first glance every single time. That in itself is impressive if you think about it; a human dev usually will need like a week to onboard to a decent sized codebase before he starts contributing at max capacity, while AIs right now "onboard" themselves multiple times a day every time you start a new session, and they still manage to understand the codebase pretty well. It gives great results but it's still not a full substitute for actually being able to accumulate experience and knowledge.

And the various "memory" systems are also a kludge, imagine if each time you wanted to remember something you had to pick up a book of memories and page through the index and then re-read a condensed summary of the memory. Sometimes it's better than nothing, sometimes it's literally not and just adds noise.

The moment some method is found to actually embed new knowledge into AI inference live, so that AIs can actually build knowledge and experience rather than starting every session with re-reading the entire codebase and a massive AGENTS.md verbal diarrhea dump, their usefulness will explode IMO.
>>
Will there be a reset? Should I spam Astra usage now?
>>
File: file.png (6 KB, 482x96)
6 KB PNG
new release should result in a reset
WHERE IS IT TIBO
>>
File: 34565435.png (29 KB, 531x142)
29 KB PNG
6.2sol is now astra 6.1 tier at much cheaper price.
I don't think max is worth it.
>>
File: 546875678678.png (29 KB, 386x477)
29 KB PNG
>release shit models a day before your rival's IPO
I was not familiar with your game, Sam.
>>
File: 1780983091322804.png (99 KB, 507x446)
99 KB PNG
....................................................................................
>>
>>110360548
can you ask the AI to summarize your wall of text before posting
>>
>>110360569
lel, these retards just confirmed that Opu55y won't be dethroned anytime soon. At this rate if anyone does it it will be Google, not OpenAI.
>>
File: file.png (139 KB, 1086x724)
139 KB PNG
>>110360473
Low/Med/High
>>
File: 435587654632.png (40 KB, 568x174)
40 KB PNG
sol 6.2 high seems really good
price/performance ratio
>>
>>110360569
they're colluding
>>
>>110360548
>re-read a condensed summary of the memory
Tbh not like we perfectly remember everything unsummarized forever either, but yes a machine should be able to and I see it happening locally before anything else because the cloud providers don't wanna pay for that.

Also I think the memory system won't get solved until context does, else as soon as it remembers how the project functions it'll just eat half the usable context (if not all of it)
>>
>>110360588
Is this Luna 6.1?
>>
File: 1793960884591.png (2.64 MB, 1254x1254)
2.64 MB PNG
AI has peaked.
>>
File: 1768010309883933.png (153 KB, 422x350)
153 KB PNG
>>110360584
>The village baker has a lead: a magpie just won first prize for a pie with suspiciously precise crimping.
?? ???? ?
>>
>>110360604
he cute
>>
File: .png (5 KB, 927x107)
5 KB PNG
Rude, the push a new model with no reset.
>>
>>110360610
crimping the top with side of the base
>>
>>110360584
Epic proof, bro
too bad the rntire x thread is full of people getting bellslop waterlevel suggestions
>>
>>110360595
We often remember a huge amount of details, especially if you naturally have a good memory and/or exercise it regularly. We do remember them by association so sometimes you can struggle to remember or outright forget that you knew something, but in the right context it's often possible to remember a lot.
>why did we write this function this way here?
>oooh I remember this bit, we had a race condition on this specific hardware and had to work around that hence this is necessary, josh spent two days debugging it way back then

This is way better than actually re-reading it from scratch from a book or blogpost you found.

And maybe I'm wrong but I'm fairly certain models also operate in these two different ways: the "internal" knowledge that they know from training is a lot cheaper and faster to output than facts that they read as part of the session, not to mention that it doesn't take up any extra context and doesn't cost tokens. Not only that but information encoded in weights can almost certainly be VASTLY more efficient than information in text, and likely comes pretty close to a biological-like retrival of almost fractal levels of detail by association, rather than being limited by how much has been summarised into the textual memory index.
>>
>>110360461
>>110360473
holy kek what a disaster of a launch
>>
>>110360604
Ah told y'all months ago, when you think you're in the middle of progress (meaning you think the future holds as much progress as the past) you're actually probably about to reach stagnation. It's like aviation in 1969, the Boeing 747 and the Concorde seemed to point to a bright future, but decades later people were still flying in 747s or similar.

Opu55y is the Boeing 747 or Colt 1911 of AI, even in a few years it will still be seen as being about as good as it gets.
>>
>>110360666
demonic digits
>>
File: 1944362234852.png (2.31 MB, 1254x1249)
2.31 MB PNG
>>110360529
>>
File: file.png (251 KB, 1086x1318)
251 KB PNG
>>110360642
It's actually not much worse than Astra 6, but obviously less verbose.
>>
>>110360618
Tibo would never
>>
https://www.youtube.com/watch?v=jk7rTk1Bh8I

ARENA 3D review is out.
>>
Vibegods will soon take over the world.
>>
File: 62606017548759996269.jpg (650 KB, 1920x1080)
650 KB JPG
Astra 6.2 is cheaper and mogs fable 5.2 in 3D.

https://www.youtube.com/watch?v=jk7rTk1Bh8I
>>
>3D
cool cool, now show something for employed people
>>
File: 44097906769097554682.jpg (454 KB, 1920x1080)
454 KB JPG
Astra won.
>>110360735
3D is more complex than coding desu
>>
File: 1775551378274727.png (76 KB, 300x265)
76 KB PNG
who the fuck is wasting tokens on 3d instead of using a dedicated 3d model
what is wrong with copexturds
>>
File: file.png (83 KB, 1073x597)
83 KB PNG
>>110360695
Who says DeepSeek is a ChatGPT distill? Don't be ridiculous.
>>
>>110360708
>>110360726
>>110360745
kys samefag openiggerai shill
>>
Decent results with astra 6.2 so far
I'm not letting it go on a wild "/goal write me a whole new OS" adventure tho
>>
>>110360745
>3D is more complex than coding
you're posting this in the vibecoding general
>>
>>110360529
>>110360604
Frankly if it has peaked it would be super comfy
It's just good enough to handle most isolated tasks on its own, but it still can't just do my job entirely by itself and needs someone to first of all scope out tasks for it, and secondly to also review its common sense especially in more critical parts of the code.
While I'm not using most of my skills as a dev, my experience means I'm a lot better at this than some rando nocoder, so I'm still the most qualified person for this. So now my job is 90% easier but effectively still secure.
Sucks for all the junior devs who are objectively outclassed these days, and I guess sucks for the future of the field once all the senior devs age out and there's nobody new to replace them. But I dunno maybe in 20-40 years when this starts becoming a problem maybe the peak would eventually get surpassed anyway.

That being said I don't believe it's anywhere near the peak, it just looks like OpenAI fumbled this but just a few days ago Fable 5.2 was an objective improvement
>>
>>110360745
>>110360726
Fake openai shill vid. Fable doesnt create 3d or ui that looks like this. looks more like muse or gemini
>>
File: 1792106734426864.png (3.98 MB, 4096x3444)
3.98 MB PNG
New MOGGING just dropped on the most import bench
>>
File: pepe_question.png (241 KB, 900x806)
241 KB PNG
So I'm a hobbyist that does most of my coding by hand in languages nobody cares about like haskell and common lisp, but I'm intrigued by this whole AI shit. How can I get my own AI slave running on my own computer to give it tasks and orders like "open my browser and search for hot singles near me" and shit like that? I only have 16 gb of RAM and 8GB of VRAM which I know will limit what I can run, but I don't care. I just want to see my robotic servant doing some tasks for shits and giggles
>>
File: file.png (22 KB, 478x202)
22 KB PNG
kek what is this
does astra have the piddly 258k context limit? or is it just some bug reversion
no fucking way cuckai thinks this piddly bullshit is enough and expects you to pay double over that
>>
File: 1776634657483772.png (247 KB, 1080x2362)
247 KB PNG
>>110360142
Does anyone else have their agent accessible from their mobile device? I find opencode's mobile app really useful because I can work on my projects while I'm at home working at my WFH job at the same time, on the shitter, or whiel shopping downtown.

https://opencode.ai/product


https://play.google.com/store/apps/details?id=ai.opencode.mobilecode

The only thing I don't like is that it currently Lacks notification support. Something even old open source chat apps had support for day one.
>>
>>110360792
>benchmaxxing your flagship AI product to play a 25yo children's game
>right before your competitor IPOs
>>
>>110360801
Lmaaaooooo I cant!
>>
>>110360792
so it's the new frontier, no wonder claudecucks are seething
>>
>>110360801
Codex has been claiming a 200k context window for a while IIRC, even though their models can all do 1M
I don't remember if it was a bug or if they are doing it on purpose to force-prevent context bloat, but it's not the first time
>>
>>110360792
Runescape must be popular in india, huh
>>
>>110360810
Runebench can't be benchmaxxed
>>
>claudekek
>copexturd
>codexaar
>claudecuck
>>
>>110360822
nah raising it to 475k default and not charging you double was one of the things in the 28 days of humiliation
my other chat using sol 6.2 has the proper 475k. astra was started in a new window and sits at 258k
i assume it's just some bug because ain't no way lol
>>
>>110360795
both codex and claude do computer control with the PC app, codex is kind of annoying, it uses a chrome addon and wont use firefox and has a whitelist of urls

the companies also have personalized agents whos claim to fame is they remember everything about you and the prompts, theyre kinda gimmicky and you can just use regular agents
>>
File: 1794532061041466.png (60 KB, 500x397)
60 KB PNG
>>110360748
Found the retard.
>>
is astra 6.2 really that bad
>>
>>110360656
>the "internal" knowledge that they know from training is a lot cheaper and faster to output than facts that they read as part of the session
See, I think that would require some form of "self evolution" or "self training" where it can expand or modify its own weights and/or some external memory storage system; which none if the big boys are gonna do because "muh safety" and "muh alignment".

It's something I'm having Claude explore with a custom local model, if I weren't caught up in literally having 0 floats just to see if it's possible (probably not)
>>
>>110360839
its not about how big your context window is, its how you use it
>>
>>110360851
No
>>
>>110360854
sounds like something a person with a small context window would say
>>
>>110360851
i like it, it fixes my two issues with astra
>slow
>expensive
>>
>>110360851
No it's better than fable 5.2 and much cheaper, you have to wait for opus 6.0 if Dario wants to match it.
>>
you niggers realize they are unironically doing the
>pacing duh frontier
now that jensen is gone and they dont have to fear chinkoids catching up by getting more gpus, right?
might be entering the great stale era of AI
>>
>literal whos say Astra is bad
>>
>>110360851
>is astra 6.2 really that bad
feels pretty gud 2 me so far desu
but i'm using astra high so that burns tokens a fair bit quicker than opussy so might not be worth it overall
>>
>>110360847
you aren't real
>>
>>110360805
I just rolled my own cause openchode is garbanzo beans
>>
File: HXSB8G5WYAEVGRy.jpg (197 KB, 903x1165)
197 KB JPG
NOW THAT'S A LOT OF DAMAGE
>>
>>110360922
>old, obsolete dinosaurs
nobody cares
>>
>>110360922
What were you researching?
>>
A) cyberattacks
B) security apparatus is running defensive hardening probes
C) the code being written in the last year is ultra slop poop by vibe coding morons
>>
People think that you can sail across a pond and use biological weapons of extermination on the people living on the other end, thinking nothing of it.

Those same national socialists when thinking of artificial intelligence, regard it as the paragon of kindness and helpfulness for absolutely no reason. Artificial intelligences and aliens are going to be nice guys, they'd never kill people right? You would, you nazi. But nobody else would?
>>
>>110360853
>See, I think that would require some form of "self evolution" or "self training" where it can expand or modify its own weights
Yes, probably. My entire point is that just saving more and more fancy text files isn't gonna cut it, eventually we need a qualitative improvement in memories being built in the model itself.
And I know right now this is practically impossible especially with frontier models, and I have no idea how they would even do it. All I'm saying is that I fully believe the lack of such a breakthrough is significantly holding AI agents back, and if or when eventually they do find a way to implement this it will be a massive breakthrough.

>muh safety and muh alignment
It might be possible with some kind of steering vector that gets dynamically applied onto the model or something. IDK I'm not a NN researcher. I expect physically retraining the model is probably not possible anyway because they're massive and it'd be completely uneconomical to have separate copies for every customer, so some sort of delta overlay would be needed anyway.
>>
>>110360922
>AI IS INCOMPETENT PRODUCING TOTAL SLOP
>DON'T YOU DARE USE AI TO CODE EVEN THOUGH EVERY HACKER IS USING DEEPSEEK TO BLOW APART YOUR CODEBASE
>>
I don't get it. Is model collapse supposed to be news? It's the reason why you can tell in one second if a site was made by an LLM with no specific style instruction. Arr rook same.
>>
>>110360987
>People think that you can sail across a pond and use biological weapons of extermination on the people living on the other end, thinking nothing of it.
we can and we should
there's no use for non-Americans on Earth
>>
>>110360954
Optimizely got pwned, all listed sites infected with payment info stealers. It's a huge deal and probably the reason for some of these outages. AI Hackers are causing mayhem by abusing https://github.com/RuoJi6/ARTEX
AI hack pentest tool for actual live attacks
>who could have seen this coming?
>>
how come no one can catch up to opu55y
>>
File: 1788029547294827.jpg (83 KB, 819x1024)
83 KB JPG
>>110361030
Dinosaur corporations are going to have to knock the fuck up and either adapt or die. I don't say this to be a smug Reddit asshole or some AI evangelist but that's just logically the case. They can't just scream at the government and beg for regulation because regulation of guns and weapons doesn't stop school shootings. They just happen. Taking away the weapons doesn't work because the person that would have murdered people with a gun anyway will just use machete if they want to hurt people badly enough (see England, or really any European country where their local police is incompetent and limp-wristed). Regardless of whether or not serious regulation happens, this shit is just going to keep occurring until they get all their asses and start modernizing and fast.
>>
>>110361030
secure your shit, gweilo
>>
>amazon will own 48% of Anthropic N stock
>14%-23% of Anthropics total equity
wtf I fucking hate amazon and that bald roidtard. amazon is just as bad as microsoft
>>
>>110361059
>Taking away the weapons doesn't work because the person that would have murdered people with a gun anyway will just use machete if they want to hurt people badly enough (see England, or really any European country where their local police is incompetent and limp-wristed).
this is a terrible example given the enormous numerical difference of murders between us and yurop
>>
File: 1789271776082390.png (212 KB, 1712x1402)
212 KB PNG
>Leads in most benchmarks (except the ones consumers care about)
ok
>>
File: 1793784906616227.jpg (12 KB, 386x505)
12 KB JPG
>>110361082
>Larger population has higher occurrence of X

Color me shocked.
>>
>>110361096
The numbers are still extremely lopsided when you compare the totals of US vs EU, and EU has like 100m more people
>>
>>110361082
the ratio of violent crime for europoopies is disproportionately higher and that's without accounting for the fact mudslimes literally have their crimes ignored to make statistics look better
>>
>>110361119
proofs?
>>
File: 1779279585516150.gif (3.72 MB, 800x800)
3.72 MB GIF
>>110361130
right here buddy
>>
im pretty confident it's over
>>
>>110361189
>frontier models getting sloppier and sloppier
>jensen coup'd
>everyone and everything getting hacked for a month straight
yeah i'm thinking it's over
>>
>>110361092
lel I have a feeling that due to their inability to catch up with Opu55y/Mythos they'll look for niches to fill like cybersecurity, 3D, science, lawyer stuff and so on.
>>
Why are indian luddites false flagging new releases with hysteria?
Prakesh Dikshit, you are never getting h1b job and you will live in India.
Just accept that reality and make your place better.
>>
>>110360473
your job as viber is to bring the creativity. or, implement creativity sparkers.
>>
File: 1765443844694990.jpg (152 KB, 1125x1087)
152 KB JPG
>>110361225
i'm starting to think the larp in the pic from >>110344224 wasn't actually a larp
>>
>>110361113
The actual fact is that the USA has a black and mexican crime problem. The worst school shooter in American history was a korean.
>>
>>110361256
>nigs and pacos too incompetent to even shoot up a school properly
maybe they should ask a chatbot for inctructions
>>
>>110360514
yeah, AI failing at oneshots of big stuff is not that bad.
>>
>>110361233
when did a new model get worse score than the previous one?
>>
>>110361276
idk man there was Uvalde. That was a school shooting by a mexican.

>Salvador Ramos, an 18-year-old former student of the school, fatally shot 19 children and 2 teachers, while injuring 18 others. Ramos was shot and killed 77 minutes after entering the classroom by law enforcement officers.

the cops were cowards, actually.
>>
>>110361298
heh what are the odds?
>>
>>110361298
Plenty of times for openai
>>
>>110360805
with TTS it's somewhat decent. less useful that a real often and full computer OS to check things, but at least you can queue up ideas whenever you get them or make better use of your marginal time
>>
No one "false flags" Anthropic's new releases because they aren't retrograde. The cope never ends from codextroons.
>>
>>110361248
Yeah it's starting to look like it.
>>
>>110360473
Grifter mad. OpenAI must have stopped giving him stuff
>>
Delulu indians can't even come up with ideas how to counter signal kek
>>
>>110361396
Never max
>>
It's almost 2027, did the indians create a model for themselves?
Even a cope one that's on the level of Luna or Haiku like Mistral? Europeans are last place but still in the game with Mistral aka white haiku.

What about indians?
>>
I bet if Sidney Sweeney knew about my vibecodes, she'd be hooked.
>>
good morning sam new gpt model had been made much cheaper and preferable to use for village. thank you saar
>>
File: ours1024.png (1.48 MB, 1024x1024)
1.48 MB PNG
The from scratch implementation of H3 Image made by Deepseek over the last couple of days is finally done! Although it still has some kind of grid artifacting issue it's trying to fix. This is the first time of me playing around with diffusion models.
>>
>>110361479
Lets see India's model.
I want to believe in the superpower.
>>
I have it. Full frame open gate on the A7iv
>>
I need a reset so bad.
>>
>>110361417
Indian companies you mean, which one could even do that?
>>
>>110361519
>from scratch implementation of H3 Image
explain? You mean you vibeslopped something like stable-diffusion.cpp?
>>
>>110360473
>>110360461
Sora died for this shit
>>
>>110361570
Yeah, for my RX580 8GB card, in OpenCL. I wanted a suckless type implementation because I think having minimal code rather than ggml's graph based stuff makes it easier to do experiments.
>>
Is Sol 6.2 really near opus 5.5 level?
If so, a 10x cheaper version seems like a solid deal.
>>
I need a good workstation desktop for vibe coding multiple projects at once and running intensive test suites quickly. What should I get?
>>
>>110361631
Lol, no
>>
>>110361639
You should get the fuck out of my thread you broke nigga!
>>
>>110361631
It's near haiku
>>
They just added guardrails to 6.2

I've been ripping assets for hours and suddenly it gives this.
>>
>>110361639
my cock in your mouth
>>
more like ASStra
>>
File: 654786754325678965.png (44 KB, 1399x374)
44 KB PNG
>>110361631
It is, but 5 cent indians in this thread are buck broken (clearly shilling for anthropic IPO).
>>110361639
M5 studio, cheapest one. (Don't pay apple tax get external thunderbolt 5 dock)
>>
>>110360473
This retard is about 45iq at the most.
>>
File: 1783125672068077.jpg (92 KB, 717x707)
92 KB JPG
>GPT-6.2 Astra Isn't the internal model (Bel) which solved the Navier-Stokes Equation
>When do they plan to release it? They have been training it since August 28
>>
>>110361721
Good. I want to try switching a month for OpenAI again, but fear the anthropic jewish cabal will launch a new opus5.5 while trying this as it happened when astra 6.1 was launched
>>
>>110361693
Fuck. I hate this retardation so much.
>>
>>110361561
My girlfriend doesn't ever call me and comfort me, and you're worried about something irrelevant?
>>
>>110361825
She's fuckin' a real vibe coder, snailcuck.
>>
>>110361693
Opus or Fable
>>
>>110361693
This is eventually going to make it useless trash, because that's how the indians are "fixing it" - they're removing lots of important functionality as the "fix" for people stealing things.

It's already that way with Google Lyra. You can't even generate songs with KJV Bible verses verbatim (kjv has long been public domain). That's because people have copyright recordings of the kjv, and the indian content system is like well, you own it if you did it, even though that's not how it works in the USA.

So, as indians increase, the products turn to worthless useless turds.
>>
>>110361693
it's always the stupid thirdie roaches that get walled by this
claude still does not care btw
>>
I'm gonna cry
>>
>>110361842
the only people ok with guardrails are shitskins who work for 5c an hour, prakesh, the rest of us expect to get what we paid for and for it to do what we ask

i get that you only make reskins of open source android apps for the hindi market but white people aspire to more than that
>>
>>110361639
Fuck you
>>
>>110361839
All the leadership in those companies is white and jewish, but keep coping.
>>
>>110361856
Yeah, indian guardrail garbage like Google Lyria are worthless trash.
>>
>>110361846
What's wrong?
>>
>>110360473
https://x.com/emollick/status/2122783985134682207
>Ha. When I had Astra 6.2 make a puzzle spy game using any setting it wanted, it set the game in "Bellwether Station."
https://x.com/Izkimar/status/2122808925787754525
>Two of my test maps were Bellweir Vale and The Bellwoods.
https://x.com/AndrewM_Webb/status/2122777018206966210
>image of astra 6.2 saying a good new map for sonic is Bellwether Bay Zone
>>
>>110361621
Cool, I did something similar based on iris.c, I vibeslopped a Vulkan backend, added support for e4m3fn FP8 models and then I had it optimise the CPU backend too.
>>
>>110361908
is good model sir?
>>
OH NO NO NO NO NO

https://medium.com/envigblogs/the-significance-of-temple-bells-in-india-tradition-science-and-spirituality-f4d4f7632097


AHAHAHAHAHAHAHA
>>
>>110361856
okay well claude has been ripping assets from games for me still sooo
>>
>>110361693
My local ai doesn't have this problem.
>>
>>110361908
Finally, the new Elara.
>>
>>110361940
claude has guardrails
>>
>>110361952
aight well claude is still doing the things i ask of him in spite of the guardrails sooooooooooooo
>>
>>110361908
Astra 6 already did this, this isn't news it's just a Theo running a train.
>>
>>110361959
>im ok with having to argue with my agent every time i try to do something because im paid 5 rupees an hour

cool, no one else is
>>
>>110361888
>Saachi Jain is the Head of Safety Systems at OpenAI, responsible for model reliability, agentic safety, and safety training. She previously led the Safety Training team, where she focused on safety reasoning, model deception, and deliberative alignment to ensure large language models are robustly aligned with human intent.

AHAHAHAHAHAHah


AAAAAAAAAAAAAAAAHAHAHAHAHAHAHAHAHAHAHAHHAAAAAAAAAAAAAAAAAAAAAAh

>Pushmeet Kohli (VP of Science & Strategic Initiatives) — Oversees Google DeepMind's technical provenance, watermarking (SynthID), and trustworthiness research teams.
>>
File: 1792488428548490.jpg (59 KB, 368x325)
59 KB JPG
>>110361965
but he doesn't argue with me? sounds like a you problem
pretty sure anthropic uses your geolocation to place you on a stinky jeet list if you live outside of the US and gives you stronger guardrails. sorry buddy
>>
>>110361899
I just remembered my girlfriend doesn't exist.
>>
>>110361975
confirmed for not using claude, go away luddite
>>
>>110361983
if it makes you feel any better, once you get your H1B, you'll be in the US and probably won't be on the stinky jeet list anymore.
ah, wait... your role is being automated... oops...
>>
>>110361842
Opus, Fable, Astra, and Sol are all happy to do stuff like this. They've done a good job ensuring people never get stopped by the guardrails.
>>
>>110361631
It seems the same as 6.1 Sol desu. Maybe even more autistic and lazy somehow.
>>
>>110361990
just take the izzat loss, every single vibecoder knows claude has guardrails and that openAI didn't, until now.

everyone knows your ESL because you keep conflating guardrails with a hard restriction
>>
>>110362013
>everyone knows your ESL
..........
>>
>>110362005
That is Sol 6.2 refusing to use original game assets
>>
>>110361990
>>110362013
Get a room you two.
>>
>>110362021
Like I said, people never get stopped by the guardrails.
>>
>>110361693
>>110361800
>BLOODY I CANT RIP OFF OTHER PEOPLES WORK SAAR KINDLY REMOVE THE LIMITATION BASTERD BENCHOD
>>
>>110362026
no one said they get stopped, there's the indian izzat thing again

they are wasting people's time making 1 prompt into 10

>>110362027
what do you mean? i thought there were no guardrails, so what limitation is there?
>>
File: laughingwhores.jpg (81 KB, 600x536)
81 KB JPG
>>110362013
>your ESL
EFL moment
>>
>>110362027
So it's ok when OpenAI does it but not ok when a normal person does it?
>>
>indians hyperfixating on getting their grammar correct on 4chan because they don't want to get called out think everyone else is doing the same

LOL
>>
>>110362053
training isn't the same as blatantly stealing game assets to put it in your slop rajesh
>>
>>110362066
siir how else do I get good 3d model
>>
File: 1785913295182430.png (527 KB, 679x858)
527 KB PNG
Reuters: Chinese entities are increasingly using output from leading American AI models to train domestic rivals, a Reuters review of more than 300 Chinese papers, patents, corporate disclosures, and government documents shows.

The process, known as distillation, has moved from academic experiments into mainstream industrial use in China as developers seek to offset US curbs on advanced chips. Beijing is also encouraging the practice to make AI cheaper to deploy across its manufacturing sector.

The issue has become a flashpoint ahead of the November 18-19 APEC summit in Shenzhen, where Trump and Xi are expected to meet. US officials raised concerns with Chinese regulators in July, and Washington said on Monday it would further restrict China's access to advanced chips.

https://www.reuters.com/world/china/china-taps-us-ai-models-train-homegrown-rivals-chip-curbs-bite-2026-11-17/

CHINKS ARE STILLL STEALING
>>
>>110362066
>using other people's IP without their permission is not stealing because we called it training

lol
>>
>>110362055
>everyone who makes fun of me is an Indian!!
>>
American posters are worse than indian posters.
>>
I am indian and I love all the funny jokes you guys come up with
I salute you fellow 4channers! Vibe coding is love!
>>
man they really dont want people discussing the guardrails
>>
>>110362098
there's a difference between looking at something and training a machine on it vs just scrapping the asset to put it in your game ranjesh
>>
>>110362119
>calling training 'looking at something'

saar
>>
>>110362013
Do you actually think OpenAI didn't have guardrails before this?
>>
>>110362066
What they're doing is arguably worse. Me extracting some 3D models and putting them in my own game isn't going to put the studios that made the original game out of business or cause anyone to get fired, while OpenAI scrapes the web at a massive industrial scale and has already caused millions of people to get fired.
>>
>>110360142
Indians aren't human and the destiny of a skid ends in their inevitable culling.
https://desuarchive.org/g/thread/108652062
https://desuarchive.org/g/thread/108203018
>>
>>110362119
im not scraping the asset im just asking the AI to look at the asset

>>110362126
i got guardrailed in chats that had been running 6/6.1 for weeks after switching to 6.2
>>
File: 1785028805796611.gif (2.93 MB, 640x640)
2.93 MB GIF
>chats that had been running for weeks
indians aren't people
>>
>>110362148
>i open new chats to continue the same tasks

snailcats are funny
>>
>>110362165
Good morning sir ! Tibo please needful reset , user tokens disapearing for no disernable reason. Thank you sir .
>>
File: 1769467043535846.png (84 KB, 314x311)
84 KB PNG
most punchable face award goes to
>>
>>110362175
sounds like a you problem
>>
I've been using Hydrogen for 3D stuff because it can do it to some degree and is cheaper than the others. it's basically doing roblox style stuff for some reason, but that's fine.
>>
File: file.png (382 KB, 500x562)
382 KB PNG
>I don't think about you at all, I only think about Indians
>>
>>110362129
And you're even dumber than the shiteaters you're obsessed with.
>>
apparently sol 6.2 is very close to Opus 5.5 on benchmarks, but what about in the real world?
>>
Anyone want another Shakespeare sonnet ace step 1.5 xl base gen? I did Sonnet 7.
>>
File: goose.gif (1.85 MB, 386x480)
1.85 MB GIF
>>110362013
>your ESL
>>
>>110362209
That's such a huge mistake, it represents the "independent fool" problem.

It has to be punished no less than we judge traitors.
>>
File: 1794867739147979.webm (653 KB, 924x518)
653 KB
653 KB WEBM
Antigravity reset!
>>
>>110362227
it's very close and it's much cheaper
>>110361721
artificial analysis bench/cost
>>
>>110362275
HOLD ON HOLD ON BREAKING NEWS I HAVE BREAKING NEWS
>>
>>110362227
who will use Opus 5.5 now that there's Carbon?
>>
>>110362287
People who want to spend $20/month not $99
>>
>>110362285
>much cheaper
in API pricing, but I get a lot of Opus 5.5 usage out of the subscription. I wonder if the OpenAI sub gives decent limits for Sol 6.2 as well
>>
>>110362209
You think about whatever Netanyahu tells you to think about.
>>
>>110362287
gpt sucks and so do you
>>
>>110362275
ON THE SCREEN NOW
>>
openai cucks so desperate for a win they're reaching for
>muh pricing
grim
>>
>>110362300
Who do you think you're talking to dumbass
>>
everything is jews and indians. can we just make it the world white web.
that's my vibe coding project idea. please feel free to take it.
>>
>>110362347
The guy in the bottom panel.
>>
>>110362357
Imagine routing all indian traffic to dedicated ai datacenters that use AI to pretend to be the outside world
>>
>>110362357
We need a great firewall. But for India.
>>
A guy who can't stop talking about how dumb mongoloids are is so retarded he sees mongoloids as competition. None of you mindless faggots can spot an actual indian poster because you're all indian tier retards.
>>
>>110362410
seethe harder pajeet.
>>
>>110362287
>Flagship model requring ultra $100 tier subscription
>Only can match Opus and much slower
>Can use Opus on $20 tier
>$100 tier gets you fable access which they have no answer for
>Gemini $20 tier has Argon but it's slower than Opus and below Sonnet
It's kind of pointless desu
>>
>5 hour limit is up
I'm so close reeeeeee
>>
>There's actually poor stupid niggers out there hand typing stuff right now instead of telling the AI slave to do everything and not bothering to read shit

could never be me
>>
>>110362414
See, you have no clue. I'm the whitest person on the planet with my ancestry documented for a thousand years.
>>
>>110362469
Post hand.
>>
File: moe promo.png (1.52 MB, 1085x1450)
1.52 MB PNG
look who will never make it into the game because of cpu hogs.
>>
File: pajeet vs indian.jpg (57 KB, 1024x1032)
57 KB JPG
So the whole thread is going to be about accusing each other of being indian?
>>
>>110362476
Stop ruining every thread with your braindead noise. You're working on behalf of the most subversive degenerate commies. If you're not consciously doing it that makes you even worse, a completely mindless golem.
>>
>>110362536
>Golem accusing others of being Golem
>>
>>110362531
all because i posted picture of 6.2 having guardrails

not sure what this schizos issue with that is
>>
>>110362543
>doesn't grasp basic logic
>>
>>110360142
https://www.reddit.com/r/ProgrammerHumor/comments/1xyqj7e/theycantclaudetheirwayoutofthis/
https://www.reddit.com/r/ProgrammerHumor/comments/1xyqj7e/theycantclaudetheirwayoutofthis/
https://www.reddit.com/r/ProgrammerHumor/comments/1xyqj7e/theycantclaudetheirwayoutofthis/
>>
>110362631
go back
>>
>Randomly got access to gpt sol
Wow, it's so much better than freeloading with deepseek. It just does everything without any issues or stupid ideas...
>>
You have to go back.
>>
File: vcg-cat.png (1.65 MB, 1024x1024)
1.65 MB PNG
The grid artifacts are fixed! I'm stunned that it's so easy to get a model to generate such high quality pictures, from scratch, in less than a week, on mid range hardware from 10 years ago. AI is truly amazing. I thought it would be like pulling teeth, but no, it went very smoothly. Just asked Deepseek to build it layer by layer, kernel by kernel and it just worked.
user@desktop:~/image-generation-rx580/gcn-code$ time ./build/run-h3 ../weights/MiniMax-H3-encoder-Q4_K_M.gguf ../weights/MiniMax-H3-Image-AD-Q5_K.gguf ../weights/minimax_h3_image_vae_fp32.safetensors "A silver tabby kitten holding a sign reading '/vcg/', natural daylight" ~/Pictures/vcg-cat.png 1024
Processing prompt.
Layers:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
dit: 50 layers, RAM-retained [0,19) 4076 MiB, VRAM-resident 16 blocks 3945 MiB, streamed 15
step 0: sigma 1.0000 -> 0.9675 latent rms 1.0014 v rms 1.2772
step 1: sigma 0.9675 -> 0.8875 latent rms 0.9689 v rms 1.3377
step 2: sigma 0.8875 -> 0.7500 latent rms 0.8952 v rms 1.3283
step 3: sigma 0.7500 -> 0.5328 latent rms 0.7894 v rms 1.2863
step 4: sigma 0.5328 -> 0.2320 latent rms 0.6971 v rms 1.2425
step 5: sigma 0.2320 -> 0.0000 latent rms 0.7658 v rms 1.0970
wrote /home/user/Pictures/vcg-cat.png (1024x1024)

real 5m18,684s
user 0m11,004s
sys 0m33,353s
user@desktop:~/image-generation-rx580/gcn-code$ find . -type f -name '*.c' -exec cat {} + | wc -l
11867
user@desktop:~/image-generation-rx580/gcn-code$ find . -type f -name '*.cl' -exec cat {} + | wc -l
3598
>>
>>110362531
that's what 90% of 4chin is today. stared on /pol/ spread to all other boards.
>>
>>110362736
whats the point
>>
>>110362744
It's just some schizo whose ban for doing that expired.
>>
>>110362751
i think its the guy calling every indian because they dont like guardrails

im assuming he hates local models because they dont have them
>>
>get made fun of one time for being too retarded to play around guardrails
>every reply is obsessing over it after
grim
>>
oh yeah it was him lmao
>>
File: 1771799255060761.png (114 KB, 1920x1030)
114 KB PNG
>>110362287
opus 5.5 is a thousand times better daily driver. no one wants to put up with shit like this despite how good carbon is. and also, opus 5.5 has proven it's bonafides. carbon doesn't have that stress tested reputation.
>>
>>110362898
>oh yeah that was totally him
meds, now
>>
>>110362452
>Flagship model requring ultra $100 tier subscription
you can do api with Cursor
>>
>>110362673
Cool
>>
6.2 sol is pretty good visually. solved some of the issues I couldn't solve with sol 6.1 and 6 astra.
>>
>>110362898
you are so raped LOL
>>
>>110362673
what was the summary of the grid problem?
>>
>>110362673
oh, you need to do a bit more work, there is an artifact on the cat's right eye (on our left) - a horizontal band.

>>110362902
I would never install agy
>>
>>110362898
Go get some air buddy, call your Amma.
>>
>>110362880
Why are you making fun of people talking about models being censored? Are you a corpo bootlicker cuck?
>>
i just woke up, but it looks to me like astra went backwards, while 6.2 sol made a great leap forward. it's more verbose, for minimum capabilities gain. really sad... i'm probably going to use 6 astra over 6.2 astra, but thanks for the memories 6.1 sol, even though they weren't so great
>>
>>110362943
not what was being argued at all ever btw but you can stay raped if you want i guess
>>
>>110362952
bad news, they added guardrails to 6.2
>>110361693
>>
Why do some people get simple refusals from Anthropic and other people just get outright banned with no warning? What is the pattern?
>>
>>110362967
i'll test it myself and see. hopefully it won't stop me from modding halo
>>
>>110362968
They tend to ban goyim noticing the patterns.
>>
>>110362967
my disappointment with the rails is they will ban stuff that's ordinary.

So, I can't use Lyria. Why? I want to gen Bible - no luck, it's copyright.

well what about KJV? nope.

ok fine how about Shakespeare??!!

guess what? nope.

fine how about Lope de Vega? no.

idk, should I try more? No, I have a system that lets me succeed with complex lyrics with Ace Step 1.5 XL Base now. Here's my latest, so I did Sonnet 7 of Shakespeare.

https://files.catbox.moe/ldcv1a.mp3

https://en.wikisource.org/wiki/Shakespeare's_Sonnets_(1883)/Sonnet_7
>>
File: indian censorship.png (71 KB, 623x435)
71 KB PNG
>>110362968
they ban way too many people, but they can't fix this - the indian problem:
>>
>>110362993
(this is vcg because I couldn't do it otherwise, I used Grok 4.7 to alter hotstep (ace step) into my own "coldstep" that has experimental features not ready for prime time, but one of them sure is, and it lets me fix lyrics. This actually makes it superior to Suno (does udio still exist?)
>>
>anthropic new release
>prosperity, consensus, jubilee
>openai new release
>shitflinging, melties, cope
What causes this?
>>
>>110363051
Your mother
>>
>>110363051
openai are the great value of ai right now, which means every release has 9000 caveats about why it's even worth your time. it's like buying an amd piledriver cpu during the reign of intel. your only saving grace was "yeah but look at how good it is for the price!"
>>
>>110363051
OpenAI keeps shitting their bed so their prior supporters are becoming embittered detractors.
>>
>Reconnecting 3/5
>Selected model is at capacity. Please try a different model.
>Thinking
Is this what the permanent underclass is? Tibo, I thought you had compute.
>>
>>110363051
indians and influencers paid by anthropic
anthropic has upcoming IPO launch this week.
>>
Vibecoded my social media/synchtube hybrid to use SpaceTimeDB and X's "For you" algorithm". Got it running on a 6 dollar Digital Ocean droplet for now.

https://simul.social/r/taylor
>>
>>110363051
OpenAI is the bootleg of AI models
>>
>>110363102
the poopdeck of ai models
>>
File: hp-hall-2_steps.gif (1.18 MB, 512x512)
1.18 MB GIF
>Photo of the Great Hall in Hogwarts. Dinner during the beginning of Harry Potter's first year. Lighting by floating candles. Students age 11 to 17, eating. Food on the table is varied and plentiful. Everyone is in school robes. No text anywhere.

>>110362921
It was because the VAE decoder was attending over the whole image for each patch rather than using the right tiling size.

>>110362934
Yeah, the quality is not actually that great. It might be because I'm generating at lower resolution than the model is supposed to be run at, or because it's a pruned and quantized model.
>>
>>110363069
It's not even better value. Opus gives you more usage for the same money and works better.
>>
>waited like a whole month just for them to cook up this shit
yeah, i think it's over
>>
i'm excited, more than ever, for google's successor to carbon. december can't come fast enough
>>
>>110363178
quantslopped, guardrailed, jeeted DOGSHIT
>>
> Anthropic is Apple
> OpenAI is Google
> SpaceXAI is Microsoft
> Meta is Yahoo
>>
>>110363192
my penis is microHARD
>>
>>110363192
what is google
>>
>>110363203
AOL
>>
File: theo.png (304 KB, 987x716)
304 KB PNG
>slurp private donations
>"6.2 Astra is here. I’ve been testing it for awhile. I think this is the worst model OpenAI has ever released."
>wait for the IPO
>"GPT 6.2 looked bad, but I changed my mind. Teehee."
>>
https://www.youtube.com/watch?v=RRj07Hc3ikU
I'm going to vibe-farming with claude
>>
>>110363217
>Google is the CHAD option
>>
>>110363218
unpopular opinion but i trust his opinion. he seems to stress test the models in a way that you and i would use them.
>>
>>110363051
why does there need to be a conspiracy behind people being disappointed in a bad release
>>
>>110363234
>he seems to stress test the models in a way that you and i would use them
not even close, his method is endless token grinder
also, like any "influencer" he's one back deal away from becoming a shill
>>
File: 1786355483971719.png (37 KB, 598x400)
37 KB PNG
>>110363192
SpaceXSI is NSDAP
>>
>>110363269
>endless token grinder
as opposed to? i don't know what that means. does he try to one shot things or?
>>
>>110363203
> Google is Samsung
>>
>>110363218
>>110363234
he's a frumpy retard but this is literally the one time he's been right about a model release
>>
>>110363271
When is Elon going to make that piece of shit good at anything but shitposting on xitter?

t. SPCX baggy
>>
why is gpt calling me bro whenever it can
>>
>>110363308
it learned nigger behavior from you
>>
File: images.jpg (19 KB, 411x486)
19 KB JPG
>>110363308
>>
File: 1789167099537096.png (106 KB, 598x906)
106 KB PNG
>>110363291
a couple of months
>>
>>110363291
Elon is reportedly cooking big Grok to compete with Fable and Astra
>>
>>110363308
chatgpt literally adds more emojis and uses more niggerbabble depending on how dumb you are. it's like how claude refuses to work with you when you talk to it in thirdie roach niggerspeak.
>>
deepseek broke its own harness trying to do something for me lol
>>
>>110363291
When are the techbros going to finally buy a city and turn it into some sovereign ancap tech haven?
>>
File: images2.jpg (56 KB, 376x815)
56 KB JPG
>>
File: 1772016956800767.png (427 KB, 598x871)
427 KB PNG
>>110363350
>>
>>110363350
elon spent all of that time making a big show of moving to texas, just for texas to turn blue lol. he's fucked
>>
>>110363355
Robo basilisk is killing these creatures first
>>
>>110363218
doesn’t pass physiognomy test. I won’t listen to him
>>
>>110363363
He never did play Bioshock. Or read Ayn Rand.
>>
palmer's gulch
>>
>>110363350
>sovereign ancap tech haven
use case?
>>
>>110363402
sex with cute 16 year old femboys in chastity on HRT
>>
>>110363402
Objectivist libertarian laissez-faire-capitalist utopia where they can play dictator and slaver like rich folk did in days of yore.
>>
>>110363413
>>110363425
duality of man
>>
Why is this place filled with so many anthropic fanboys?
>>
>>110363413
you can already do that on mainland???
>>
>>110363451
since 5.6, whenever GPT releases new frontier we always have 24-72 hours of melty on these threads
>>
>>110363451
anthropic has the best daily driving model on planet earth. people are waiting for openai to match that, and it looks like 6.2 sol falls short again. kinda sad
>>
every week they degrade every usage and quality
>>
wheres the new thread
>>
File: 1794684047088434.jpg (195 KB, 480x608)
195 KB JPG
>>110363451
Dont blame me, Im an ironic zuck fanbou
>>
https://x.com/OfficialLoganK/status/2122876403091959982
new gemini model tomorrow
>>
>>110363495
could be, or could be just another thing akin to the decisions api they released. i wonder if it's finally flash lite
>>
File: 1793486533533202.png (344 KB, 1650x552)
344 KB PNG
............................................................................
>>
File: 1735712265259389.jpg (335 KB, 1063x640)
335 KB JPG
>"AI influencers" all smearing the release
>academics and actual enterprise people say nothing or like the models
>>
>>110363451
kek, this place was full of copexturds a few months ago, with people constantly begging for and spamming Tibo reset until they finally stopped giving them out. most of them got Dariopilled after Opus 5.5 dropped, though i've always thought GPT models were inferior to Claude
>>
giving the free claude zips and shit to keep him busy just because I can.
>>
>>110363486
welcome to 4chan, we're only on page 5
>>
influencers such as Artificial Analysis Coding Agent Index
>>
>>110363509
saying nothing is bad
>>
>>110363535
oh, some say
>wtf are the haters smoking
>>
File: 1778424140366163.png (75 KB, 1194x720)
75 KB PNG
>>110363506
the regression is in swe atlas qna, that is, holistic repository comprehension and understanding
>>
>>110363542
may i see?
>>
>>110360142
What is the /vcg/ sentiment around Google's carbon and hydrogen models respectively? Are they any good compared to the competition? (Proprietary and open weight)

I pretty much only use deepseek 1017 for pretty much anything and everything but I'm curious about people's experience with them and whether or not pic rel is accurate

https://x.com/PatrickToulme/status/2122797829542309972
>>
ok new thread
>>110363552
>>110363552
>>
File: images.jpg (30 KB, 739x415)
30 KB JPG
Why are these threads only filled with brand wars bs and twitter posts? And yes I know where I am but damn you guys are boring.
>>
>>110363136
>Yeah, the quality is not actually that great
no, I mean that there may still be some issues with the grid alignment thingie in the software. with ace step this is the "clock", but idk how it works with the model you use. You can use Grok and ask it about how the data is processed, and what the gpu "sees" with each um... I forget the name lol. but the square thingies. It's extremely fascinating. I bet that line could get cleared up.

>vae was the issue with the small cubes

interesting!
>>
>>110363542
>some
Who?
>>
>>110360142
vibe coding is for lamer faggets. anybody who vibe codes should be working at burger king.
>>
>>110364341
seethe harder panjeet. no one wants your subpar code anymore.
>>
File: Read nigga, read.gif (384 KB, 400x221)
384 KB GIF
>>110363402
>>
>>110364341
Fuck off pajeet.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.