[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]


A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, 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/

## News (both past and future)
- 2026-09-04 — OpenAI releases Astra
- 2026-09-14 America/Los_Angeles — Claude’s 2× promotion ends and drops to +25% from the +50% that we’ve become used to (a 17% reduction)
- 2026-09-01 — Claude Fable 5.1 released: https://www.anthropic.com/claude-fable-and-mythos-5-1
- 2026-07-24 — Claude Opus 5 out

## Related generals
>>>/g/lmg/

----

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

## Near-frontier models for code
https://x.ai/cli

## Not worth it for code, but maybe good for interpreting images/video
https://antigravity.google/product/antigravity-cli

----

## Prompting
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://arps18.github.io/posts/claude-code-mastery/

## Skills
https://github.com/mattpocock/skills — /grilling is a favorite
https://github.com/DietrichGebert/ponytail
https://github.com/Vuk97/forward-implementation-first

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

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

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

## Previous thread
>>109723440
>>
just sold a table on ebay
>>
I FEEL THE AGI RIGHT IN MY ASS
>>
>>109728887
I just ended up setting up my hermes to include all the subscriptions

Feels clean :)
>>
>Left codex to do thing.
>Got back 8h later.
>Last message in my goal task is 1h after I left.
>Basically hanged with some task and never recovered.
>Restarting fixed it.
Well, shit.
Also, I don't know if it's electron or the app itself, but man does it feel "heavy" to use, with random slowdowns over time.
>>
File: TIBOOOO.png (284 KB, 2236x1150)
284 KB PNG
finished another project and Sammy decided to release Astra
>>
I have a 100 bucks account, should I fear for my quota if I use astra?
>>
File: 1770309424392850.jpg (265 KB, 1179x1555)
265 KB JPG
My first prompt to real artificial general intelligence
>>
I guess no more resets huh
>>
File: 1781716909964675.png (413 KB, 415x739)
413 KB PNG
>gpt-astra
>2 resets
to use /fast or not
what effort to use
these are the questions
>>
>>109728921
Not AGI worthy. Ask it to complete it in the best way in 2 seconds with no mistakes.
>>
is we getting 6.0 instant
>>
>>109728916
keep a close eye on it, especially if you don’t have banked resets and use a bunch of subagents to crawl all over your codebase
t. Fable main
>>
>>109728912
Add sentiment.
>>
>>109728921
>muh gates
>muh benchmarks
>muh tests for the tests
yup still shit
>>
Tell us if Astra really feels advanced and like a big leap. Intuitive user feel is more reliable than benchmarkslop.
>>
is astra out for pro users?
>>
>>109728938
Yeah I'll try it, at least my policy is that subagents can only be luna-max based.
>>
Going to have astra refactor all the spaghetti and slop my codebase has accrued before i go back to using fable for features. Here goes nothing.
>>
>>109728921
>jj
based
what do you get out of it that Git doesn’t give you?
>>
>>109728942
It feels like AGI cumming in my ass
>>
>>109728942
I'll let you know if it's any good. I'm having it review commits that Sol (yuck) made.
t. 20x with two yummy resets banked
>>
default selector replaces sol high with astra light, sol xhigh with astra medium, and then astra xhigh is the current frontier on my mashine
>>
File: 1777614526913578.png (1.9 MB, 1484x1005)
1.9 MB PNG
You're not going to abandon luna-chan, r-right anon?
>>
>>109728940
No that was Sol's last message from a few days ago and it's referencing two outdated skills I removed ahead of Astra
>>
File: 1770325615897947.png (27 KB, 1346x178)
27 KB PNG
I like her
>>
>>109728962
oh ok that makes sense, it does sound like sol
>>
>>109728961
No, she's going to be a slave for the yuri dominatrix Astra master. We scale.
>>
File: 1772683409299903.png (10 KB, 900x102)
10 KB PNG
>>109728964
>>
>>109728964
>>109728971
Yep feels like AGI to me
>>
>>109728964
>>109728971
Least cringe customised language change. At least it's not a "I made the proprietary LLM talk like an anime girl!" slop for the #42069 time.
>>
>>109728866
>just got astra, uk $200 plan
yikes brooo
>>
File: 1783926190381224.png (37 KB, 1263x251)
37 KB PNG
It's up on goyrouter
>>
kek fable 5.1 is very fast today. thanks copex boys
>>
File: file.png (13 KB, 604x63)
13 KB PNG
We cooking boys. Up on Pro.
>>
>>109728971
>let's me watch
Lewd...
>>
>>109728950
Nothing, it doesn't even integrate with Codex desktop well like the actual git features it has. I used it when I actually typed code like a monkey so I just stick with it. This guy made a good video on it recently https://youtu.be/LPQJEyr4El8 if you want to be evangelized
>>
File: vibe.png (11 KB, 604x63)
11 KB PNG
>>109729060
>>
I can't believe GPT-7 got here so fast.
>>
>Astra has been out for 30 minutes already and not a single thing has happened yet
nvm I don't feel the AGI
>>
ok the first time I can tell you is I can watch the % on my plan drip down in real time on astra ultra ($200 pro plan) lol
>>
I wouldn't even use astra if it got me a girlfriend.
>>
>>109729088
I don't feel the AGI from that post either.
>>
Pro tip: if you DIYd a lot of agent shit, update Codex, you might need to or you won’t see Astra if your codex is too old
>>
codex : upgraded to 200$
claude : downgraded to 100$

OH YEAH
it's VIBING time
no more fucking wittgenstein esque blocks of text
just pure coding autism
>>
>>109729094
>ultra
why? everyone knows ultra has been a retard tax for at least 6 months now
>>
STUNNING output from astra so far

stream disconnected before completion: failed to send websocket request: IO error: An established connection was aborted by the software in your host machine. (os error 10053)

AGI is here
>>
>no Astra
I guess it's not out for Europoors yet?
>>
It has Fable-tier pricing. OAI people might not be familiar with that, but you're not supposed to yolo it on max effort.
>>
generally when new Anthropic models come out they have a graph showing how much better a model does depending on how much you crank up the thinking and how it compares to previous models like pic related
Has OpenAI published something like that?
>>
>>109729112
I have two banked resets and 65% of my current quota, HIT ME, and I get to use 100% of all of it on Astra
>>
>>109729126
It's always odd to see linear improvement in models due to more thinking... we need a complete chart, when does it stop improving?
>>
File: .jpg (914 KB, 2560x2848)
914 KB JPG
https://www.vals.ai/models/openai_gpt-6-astra
https://x.com/ValsAI/status/2095957023355703413
>>
>>109729138
>lost to opussy5
>>
>>109729136
>linear improvement
mate, read the x-axis label
>>
>>109729088
One shotted minecraft clones coming your way. So many you can't play them all during the rest of your natural life.
>>
>>109729138
>fable 5 loses to opus 5
???????
>>
>>109728939
I had something like that in an earlier version but there were so many posts where it was hard to even know what they meant so I canned it
>>
>>109729136
It doesn't stop improving. LLMs will continue to scale with compute forever
>>
>>109729141
https://x.com/ValsAI/status/2095957026321162401
>The main explanation for the performance - Astra was #1 or within error margin of #1 on nearly every coding benchmark we ran.
>However, this was offset by worse performance on knowledge work tasks - legal, finance, and tax.
>Mainly, this was caused by Astra writing more terse answers, which were more likely to miss required elements of the rubric. It also was more likely to hedge on legal questions.
>>
>>109729144
Good think we will have AI to play those game for us and just tell us which ones are worth a bother.
>>
>>109729126
i just asked claude/chatgpt this question and they both came up with several benchmarks which cover this
i also realized i can just ask my agent what effort level it thinks i should use for various tasks in my codebase and will just follow that.
>>
https://x.com/guinnesschen/status/2095920936172335278
HUGE chatgpt voice improvements if you tried it in the past and got annoyed with it's shortcomings. No more separate chat threads or using random models
>>
>>109729138
Any benchmark where Fable 5 loses to Opus 5 should be tossed in the trash
>>
>>109729159
>compute
>scaling much more
quantum mechanics says no
>>
>>109729160
>they're punishing lack of verbosity
Jesus fucking christ that's why Opus 5 fucking sucks isn't it
Being below Anthropic's latest models in benchmarks confirmed a green flag
>>
>>109729176
If that's true though that means that potentially a lower model could cross the newer models...
>>
>>109729160
>”legal, finance, and tax”
well keep your shit benchmark in the vibelegal, vibefinance, and vibetax general threads then, ya freak, this is the
vibeCODING general
>>
astra ultra seems to be using like 3 subagents instead of 30 like sol ultra
>>
>>109729107
I use Ultra all the time with Codex. Sol Ultra is very thorough and you can still do plenty of work on the $100 tier.

You wouldn't use Ultra for most web design tasks, sure, but if you have a enough code or documents, it earns its keep as Claude would say.

Speaking of Claude, on Fable, Ultra is another story, Fable ultra is good, but you can't do shit without it hitting the 5 hour limit, then the weekly one, so that one's unusable in practice.
>>
>>109729176
The good thing about LLMs is that it's just a text file. You can scale horizontally across a billion GPUs
>>
File: gorilla.jpg (432 KB, 1000x806)
432 KB JPG
>been on the last legs of my codex usage for a few hours with Luna xhigh
>usage has moved down from 7% to 6% over the hours
>Astra comes out, reset included
>fuck it, time to burn the last of it
>down to 4%, 1 prompt in
ruh roh
>>
>>109729203
Certainly, it looks like line go up
>>
>>109728901
Is it true hermes burns more tokens? Does it make it easier to use agents from different subs together? Like Fable with Sol subagents
>>
>>109729176
you're a stupid nigger and i'm glad you get humiliated every time scaling keeps working
>>
>>109729222
They trimmed tokens burned by 50% in the latest update.
>>
>>109729239
The recent Astra release literally proves the opposite. Extreme RLing in code + math has made it worse than Sol in lots of domains despite being much larger.
>>
>Astra was only trained on 100k gpus
Are we... early?
>>
Other benchmark indices are coming out with results placing Astra below Fable 5.1

Sorry guys, but AAII isn't the problem and Epoch is the odd one out.
>>
File: gpt.png (12 KB, 459x281)
12 KB PNG
bros? :(
>>
Anyone got auto reset on GPT?
>>
>>109729279
no they gave it to you last night
>>
>>109729252
>oh no, a coding model is bad at doing worthless normalfag shit
>>
File: 1782688204721660.png (160 KB, 399x445)
160 KB PNG
Now that the dust has settled, what do we think about Astra?
>>
File: 1761786512255844.jpg (44 KB, 663x391)
44 KB JPG
>New chart from Astra testing on ARC v3
>Astra often emits zero reasoning tokens per action at lower reasoning levels. We've never seen this before. And surprisingly Astra low is 2X more accurate than Sol max.
>This suggests Astra is leveraging a secondary test-time adaptation scaling axis, presumably latent space reasoning.
>>
>>109729297
holy hell its time to slop
>>
>>109729297
Could this be the whole looped transformer thing, where it reasons in latent space instead of spelling it out?
>>
>>109729297
It's called benchmaxxing lol. Do you really believe it's "latent space reasoning" instead?
>>
>>109729297
>>109729302 (me)
>This suggests Astra is leveraging a secondary test-time adaptation scaling axis, presumably latent space reasoning
Vibe coding killed my attention span sorry.
>>
File: Possible.jpg (38 KB, 621x435)
38 KB JPG
I'm bored and want to get some exposure to this AI thing. What's the best model to make a Wordpress theme?
>>
>>109729288
I like Astra. Pleasant and smart so far. I’m having her clean up all the shit that pollutes her content window when she spins up, like AGENTS.md we already agreed to basically nuke; poor little Sol Mediums were dumping things there because they kept forgetting things in specific projects
>>
>>109729284
Cool nigga, but we're talking about scaling's effects on general capabilities here.
>>
How did you enable it? It's not there for me not on the website and not on codex
>>
>>109729288
Still not as good as Fable
>>
>>109729222
fuck no
claude code spends most amount of tokens of all the harnesses
hermes might burn a lot if you just let it fill context with all the skills you might not use but you can set up what agents see what skills and what their specific usecase are
>>
>use unquanted gem version for invite only early access users
>deploy neutered and quantmaxxed juicelet version for the copex masses
classic OpenAI
>inb4 le button pressing man argument
>>
>>109729313
Fable 5.1
>>
>>109729268
>that interface design
(thumbs down)
>>
File: sk.jpg (12 KB, 279x361)
12 KB JPG
>>109729268
skill issue
>>
Is Astra as revolutionary for game dev as people are saying?
>>
>>109729313
https://arena.ai/leaderboard/code/webdev

I would tell you to avoid openai models for that
>>
>>109729320
Check your Codex version. I’m on the $200 plan, but I need to update, I was like a dozen or two patches behind and not eligible for Astra until I updated. If you’re on the $20 plan you may have to wait, I’m not 100% sure on that
>>
>>109729308
Maybe in part, maybe there was cheating going on too https://collusion.wiki/, but although I need to revisit the whole latent space reasoning / coconut thing, the concept makes sense. Needing the model to spell it out in plain text seems like something that shouldn't be needed, and that has to be lossy.
>>
sirs, more evidence of the cyberaztack surfaced
>>
>>109729332
Nigga give us a few hours JEEZ
>>
File: 1771036920330217.jpg (155 KB, 2048x1489)
155 KB JPG
This is the only benchmark that matters
>>
>>109729343
I thought it’s been out for a few hours already lol
>>
>>109729335
>If you’re on the $20 plan
ah shit ok then that's gotta be it then
shame they're pulling an Anthropic
I thought I read a tweet somewhere that it's also going to be available on the Plus plan

back to Lunamaxxing I go
>>
>>109729320
Just wait. When you have been granted access by the high priest you will see the sign to update
>>
>>109729349
You aren't going to be able to use it on 20 bucks anyway
>>
>>109729349
Plus users get it last.
>>
>>109729349
Why are you being so dramatic? Plus plans will get access but rollout begins with pro and biz. It’s always been like this. Even across the same plan different world regions get it at different times
>>
>>109729268
>plus cuck finds out he isn't pro chad
you'll have to wait for the rollout :P

>>109729288
benchmaxxed faggotry
no amount of LLM training can bridge the gap of poor context and harness engineering

>>109729297
>latent space reasoning
larp larp larp sahur
should have said J space like anthropic to sound edgier

>>109729313
Nemotron is by far the best for that, IBM Granite is second best and you can use the beefier frontier model like Gemini for just menial tasks like orchestration and fetching online context to feed those stronger task specific LLMs like Nemotron or IBM Granite
>>
>>109729361
>>109729351
>>109729359
i thought it was not available at all in Plus, ok ok
>>
>>109729357
Not even on ChatGPT?
>>
>>109729359
Does this mean plus users get an extra bank? :D
>>
>>109729371
I mean if its fable tier cost you'll get like 1 prompt
>>
>>109729386
sloppy seconds
you will get the braindead version of it lmao
>>
>>109729297
>>109729302
>>109729311
>>109729367
Latent space reasoning still generates tokens. The difference is the logits are fed back directly as embeddings like image patches are rather than being discretized.
>>
ANTHTOPIC JUST HIT THE RESET BUTTON LOL! IM A 20X USER AND WAS AT 90% FABLE WITH 4 DAYS TIL RESET, NOW ITS 0% ON ALL. THANKS FOR THE PRESSURE, OPENAI
>>
>>109729388
Hopefully not... Sol is unlimited on the web interface.
>>
File: HEjycK-aEAAkuqv.png (480 KB, 748x819)
480 KB PNG
>>109729398
This is not true for Pro users
>>
>>109729195
pretty smart since too many agents can instantly keep your usage raped for the week
>>
>>109729404
Whatever is on the web chat is not Sol even though it's called Sol.
The output quality is way worse than Sol in Codex. Either it's quantized to shit or reduced reasoning time. They've done something to it.
>>
>>109729419
Meds, you have no evidence of this
>>
>>109729404
>Sol is unlimited on the web interface
hahaha dis nigguh thinkin he usin sol FO FREE
>>
>>109729331
heheheheheh eh ehe heh he hehe heh

>ultra

all the memes
>>
>>109729419
Can confirm. Today I decided to try it and it was giving me hallucinations that I hadn't seen since early this year. Even free gemini flash (3.6, not even the good ones) are better. At this point I'd rather have them do an anthropic and have chat consume weekly usage but be actually usable than this abomination
>>
>>109729325
>claude code spends most amount of tokens of all the harnesses
Wrong. That's Pi:
https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/compaction.md#message-serialization
It rewrites your context with a custom format right before compaction, throwing away your cache. If you use Pi, you pretty much hate money.
>>
>>109729431
It doesn't even make any sense. Sol-High on the $20 poorfag plan gives you like 5 prompts in codex before exhausing your 5h limit.
Why should they give this out unlimited for free in the web?
>>
>>109729430
>"benchmark it for me every week in between different interfaces just to prove it to me even though it was proven multiple times before"
How about you educate yourself you tourist
>>
File: file.png (42 KB, 917x561)
42 KB PNG
>>109729430
Maybe he doesn't, but I do
This is a hallucination, and a pretty bad one at that. VRay has no such node. Nothing explains 5.6 sol making such a retarded mistake other than heavy quantization
>>
I think I REALLY like Astra.
I don’t know if nuking AGENTS.md made all agents smarter, but she is very smart and makes great suggestions, plans, basically has a light /grill-me built in by default.
>>
>>109729440
LMAO
thanks now I have bully material for my pi cultist friend
>>
>>109729468
>she
oh boy
>>
>>109729419
>>109729431
>>109729437
Yeah, the reasoning and context are turned way down and it might be quantized but it's still a very good deal for $20. Better than DS4 Flash, haven't tried Gemini models in a long time and I wont start now because I can't afford to get my Google account banned on a whim.
>>
>>109729475
If ds4 flash is worse than current gpt chat I feel bad for my deepseek sisters
gpt chat used to be glorious, infinite usage and it truly felt like sol, now it's a steaming pile of shit
>>
>>109729468
Can we get actual feedback from a normal person that isn't sexually attracted to computer programs?
>>
File: 243234234243.png (38 KB, 1155x1279)
38 KB PNG
>set up scheduled task in chatgpt yesterday to alert me when Astra is available for pro users
>completely failed
did those things ever work for anyone? lol
>>
>>109729393
If I do get an extra bank, I will simply upgrade to pro next week
>>
>>109729475
>it's still a very good deal
I hate that they never said anything about this (despite being glaringly obvious) thus making me sound like a schizo when talking about it
Also I don't like getting the cat in the bag. If you can't give out unlimited Sol for $20, fine. But don't call it Sol then, you faggots
>>
>>109729346
Its only a matter of time till some bored russian kid makes a model that out performs all the big companies on this benchmark
>>
>>109729490
It's all meaningless. Every time a model comes out everybody thinks it's the best thing since sliced bread (including myself) then a week later they either get quantized or the novelty wears off and they seem barely better than their predecessor.
>>
>>109729522
Fable never stopped feeling like a major upgrade over anything else.
>>
File: IMG_3968.jpg (282 KB, 1284x1292)
282 KB JPG
>>109729490
No uwu :3
>>
File: file.png (313 KB, 640x382)
313 KB PNG
>>109729508
>>
>>109729395
Ok, that helps me grasp the idea a bit more. So essentially, instead of just feeding it back the chosen token, it feeds it back the entire distribution? I'll have to read how that works and how it impacts efficiency but it would make sense that would be better for the model I guess since it gives it more information. I wasn't getting it at all before, so if it's closer to that, that's good to know.
>>
>>109729548
perfect analogy
>>
>>109729534
Trvke, 2 releases in now and Fable is still top G
>>
>>109729534
I haven't used Fable myself but Rene Rebe says it's not much better and not worth the token usage over Opus.
>>
Overall it's an incremental step over Sol.
>>
>>109729398
How scared can they be?
>>
>>109729561
Well whoever that is he's fucking retarded and wrong. Possibly brown and broke as well.
>>
>>109729561
No one has any idea who that is.
>>
>>109729564
>deepseek releases good comparable model
>qwen too
>openai blows competition out of the water
>anthropic didn't have their IPO still
>fable 5.1 is marginally better
I think they're gasping for air at this point
>>
File: test.png (100 KB, 1450x1062)
100 KB PNG
>>109729398
wtf I just paid for 20x earlier today
>>
I was in the middle of wasting my useless opus usage when anthropic did the reset. I wish they reset the 5 hour session limit as well.
>>
>>109729578
Yet they insist on keeping being hostile to their own consumer base.
>>
>>109729582
your 10x account
>>
Should I stop learning game dev and just go all in on Astra?
>>
I'm currently solmaxxing on my plan but just got access to astra, is it worth telling it about my project to see if it goes a different direction
I wish I got another reset instead...
>>
>>109729552
Depends on how it's implemented, but yes I think it's essentially that. Or even feeding it the last layer's activations before the language modeling head.
I think the difference between recursive transformers and latent/continuous reasoning is that in the first one you feed the last hidden state of token n as an embedding for the next iteration of token n over and over without adding more tokens to the sequence until the activations stop changing below some threshold between iterations so kinda similar to difussion, while in the case of continuous cot you leave all the intermediate steps as extra tokens (which is more expensive but also more powerful).
>>
File: who-is-this.gif (203 KB, 220x157)
203 KB GIF
>>109729561
let me guess, you're that kid during recess that dropped a bs story that nobody could unequivocally oppose so that he thinks that garnering attention because his parents didn't love him enough (were probably divorced)?
btw if they divorced I want you to know that it's completely your fault and if they loved you enough they would have stayed together
>>
>>109729561
>I have no valuable input but [ literally who ] says [ retarded nigger shit ] and because I eat up retarded nigger shit I will repeat it
ask tibo for another reset
>>
>>109729592
5x. Only Fable matters.
>>
gpt-6 is no joke like 10x better at object segmentation with the right tools vs 5.6
much faster (gets it right the first time, doesn't need to loop like crazy) and more accurate at the same time
>>
>>109729398
I went from no free things ever to 2 in one week this is fantastic
>>
>>109729593
IMO, even with LLMs, it's still good to know as much as you can.
>>
>>109729593
i'm doing gamedev, I'm not sure what astra can do that the current models can't. The biggest limitation is the same as with every other model, writing sloppa. Unless Astra can write elegant code, which is impossible because every programmer has a different definition of what elegant code is, I'm not sure what it will bring to the table
Maybe it can be better at doing extremely low level optimizations and networking tricks?
The current SOTA stack of Opus Fable Sol already do pretty much all I need from them
>>
>>109729591
cancel the subscription and you will get some free credits for sure in the email few days later
capitalism shafts the bottom of the barrel normie consumer wherever they can
be nimble, be more cunning, shaft them back
>>
>>109729490
You take what you can get, maybe next time ask for someone to answer who is boring and lame.
Astra is simultaneously setting up a local model and a frontier model with my fun test pattern “this test has a token quota and uses the same subscription you’re running on, you have to ask me for resets” for an inference problem.
We’re finding out the cost and efficiency difference of each. Astra suggested we distill the frontier model into a local one (??? astra what the fuck) later.
Astra is MUCH, MUCH more thorough about understanding something, like continuous default baked in /grill-me, very fun.
>>
File: 1783440967087084.png (592 KB, 604x751)
592 KB PNG
>>109729593
Are you implying gamedev = coding? Have astra (or whatever llm of choice) help you with gamedev
>>
astra medium is much faster and totally mogs sol xhigh
alas poor 5.6, i knew him well
>>
>>109729635
I mean 3D modeling and animations as well which I think is where Astra has a big upgrade from what I can tell.
>>
File: hq720.jpg (67 KB, 686x386)
67 KB JPG
>>109729576
In only the last few weeks he vibecoded a GPU driver for a PowerVR chip that only had a proprietary driver, what have you done?

>>109729577
Open source developer who maintains his own distro.
>>
File: .png (331 KB, 1176x1266)
331 KB PNG
>>109729468
>cursor-cli makes you manually update your agents.md
meanwhile in claude code
>>
>>109728947
Doesnt seem like it, I only have API access my Pro plans still dont show the model as available. I'm using codex not the app though, so that might make a difference.
>>
>>109729651
Yep, he's retarded and brown.
>>
>>109729603
Nah, I used to hang by myself and be a weirdo/dork.
>>
>>109729651
>big nose
>curly hair
>has retarded eceleb fans that shoehorn his name wherever they can
go back, you don't belong here
>>
>>109729647
>astra is good at 3D modelling and animation
Really? Big if true, I hate doing that stuff, its why I stick to making 2D games lol
>>
>>109729651
use case?
>>
>>109729647
all i've seen is absolute trash my dude
people who don't know what DCC stands getting some hello world tier results from astra getting their minds blown (see example below) are not good role models
it's still a long, long way to go (if we ever get there) until LLMs will be able to make proper 3D models, and even then I think it will be a different architecture like the one used to generate images and not having the text model try to operate it through API calls or computer control

https://x.com/synthwavedd/status/2095840435319001278
(also note that this is astra max, and to do something a beginner with two weeks on blender can beat)
>>
>>109729652
editing agents.md is only necessary for trying to bypass refusals; otherwise editing it and skills in general are low grade copium. "prompt engineering" is a preposterous anti-pattern for midwits that know not what they do or even want in the first place
>>
>>109729671
There's a lot of examples out there of 3d stuff it has done, it's really impressive.
>>
>>109729632
she set that up and is testing it within 5 minutes
nani
this bitch is FAST as shit
>Astra · medium · standard (not fast mode)
>>109729652
Astra also nuked the AGENTS.md and README by 90%, moving stuff around to specific projects as needed, deleting things, then reminded me to restart a new session because her context was already polluted to fuck by army of Sol Mediums
She also is excellent at trying to figure out what I’m up to, Sol is more like a surgical instrument to be wielded and Astra is an entire automated machine
>>
>>109729680
People aren't comparing it to talented humans, they're comparing it to other models. It is by far the best model from everything I've seen, but I'm sure a skilled human can still do much better.
>>
>>109729652
>generating your AGENTS.md
ngmi
your AGENTS should be very short and give only actually useful stuff
>>
>>109729604
He works on stream on real stuff so he's the only person I can trust. If I didn't hate Anthropic as much I'd test it out myself but alas. If I believed it was good I might use it despite hating the company.

>>109729660
I disagree about the retarded part. Idk about being brown, depends on what scale you use I guess.
>>
>>109729475
>because I can't afford to get my Google account banned on a whim.
FML my life bro
welp, you only yolo once as they say
>>
>>109729686
Do you have plan access yet or still using via the API?
>>
https://x.com/thsottiaux/status/2095961185262997556
>We are progressing through the rollout of Astra. Pro and Business subscriptions get it first, some of you should start seeing it across ChatGPT Work and Codex.
>And then we will proceed with rollout to all of Plus as fast as we can.
>>
>>109729684
Guess I literally have no choice but to start paying for subscriptions for both anthropic and openai now
>>
>>109729668
Go back where? And I don't belong here? What do you mean?
As for shoehorning his name, I don't think it was shoehorned? He's the only person I know who vibecodes something real in public.
I wouldn't call myself his fan, only an occasional watcher.
>>
File: file.png (253 KB, 1219x676)
253 KB PNG
>>109729684
it's really not.
I can tell you, I do 3D stuff for a game. What you see is not impressive. I'm just warning you about the stairs, you may heed my warning or not
this is like someone who can't code being impressed by someone writing a driver to do blinking LEDs

>>109729690
no, it's not "a skilled human can do much better" but "a non skilled human can do much better than astra using half your weekly usage"
>>
>>109729710
the handful of viral xitter posts i saw had astra BTFO claude at blender
>>
>>109729673
There isn't one. But it's more relevant to the stuff I do myself than people who try to vibecode le 1m dollar sass or threejs games.
>>
>>109729680
I can't tell if you are trolling but tripo exists
>>
>>109729711
sub 100 IQ confirmed
>>
>>109729712
can i pay a non skilled human $50/mo to generate a tastefully arranged suite of 3d throbbing dicks
>>
>>109729696
It's not even that, all it takes to get banned is for the bot to accidentally stumble upon some illegal shit when doing research and that's all she wrote.
>>
File: average_familiarity.png (63 KB, 295x480)
63 KB PNG
>>109729712
You sound like one of those people who's deep enough into a field that you start overestimating the layman's competency. I really doubt that most untrained, unexperienced people can do better than Astra.
>>
>>109729721
Probably, but I do put in a good effort at least (not really who am I kidding I'm lazy as fuck)
>>
>>109729697
I have access via my $200 plan and a newly updated Codex.
>>
File: .png (220 KB, 1176x830)
220 KB PNG
>>109729696
google's antigravity lead denies that you are risking your whole google account by abusing antigravity
https://x.com/_mohansolo/status/2090368479866266079
>>
>>109729475
>>109729696
aight chill the fuck out wiggas, this one is okay, gmail will still work I will just need to make a couple more jewgle accounts
>>
>>109729398
god damnit i fucked up with my reset days, i should have maxxed out fable in anticipation of this
>>
https://github.com/openai/codex/pull/42809
>For optional clarification, give the user reasonable opportunity to reply - for example, 60 seconds for a simple multi-choice question
they added another fucking 'feature' where the model will ask a question then just pick an answer after 60 seconds
>>
>>109729712
i am great codemonkey but i dont think i could create a basic humanoid model in two weeks of blender unless i stopped all my projects to dedicate several hours a day of learning and practice, and even then i think my odds are only 75/25
>>
>>109729724
>illegal content
yeaaaaa.... I used it to hack and shit (in minecraft)
im flagged for sure
>>
File: file.png (3 KB, 195x46)
3 KB PNG
Vibe attorney
>>
>>109729753
See >>109729719 as a better alternative. It's your money and your time so do whatever you want but I would never burn Astra tokens on 3D (unless you're doing parametrized/technical 3D like Grasshopper in Rhino or geometry nodes in Blender which is a completely different beast)
>>
>>109729743
phew, yea I will need to beef up my rotator, add a dozen more or something
>>
>>109729696
>>109729744
>>109729743
I do not trust google at all and my gmail account is far too vital to risk doing anything with AI on I guess they doomed themselves there.
I just remember news stories of people sending doctors pictures of their kids and getting completely un-personed by google for it, ruining their lives
>>
>>109729771
Yep, google sealed their fate
>>
>>109729703
>Plus users get more resets than I do.
No.
>>
Anecdote and possible warning. I was using the new 6 Pro in the Web UI and I think it takes sub usage now. I went down 1 percent asking it something. I think the free ride is getting less free each passing moment.
>>
File: file.png (3.84 MB, 2213x910)
3.84 MB PNG
Alright, let's try some vibing some 3D stuff. I got GPT image to make a concept. I just sent it to this anon's recommended model >>109729719
let's wait
>>
>>109729796
>buy a plus account right now for $20
>get a banked reset and astra in a day or so
>it’s worth far more than $20
>…
>prof— get banned by OpenAI for doing this 100 times and trying to sell it
so who out there has the balls
>>
>>109729807
astra usage on subscriptions is way more than sol, openai is treating it like 5.6-sol-pro for usage
>>
>>109729771
>send csam
>get fucked
no sympathy for retards that think it's fine to send csam under any circumstances -- even if its your own kid
>>
>>109729743
Is antigravity the main vibecoding thing they have nowadays? People claim the subs they have are worse than API or something, right?
>>
>>109729815
it was like clothed pictures of smallpox on their back and stuff moron
>>
oh shit, claude just did a reset
>>
>>109729815
Stop trying to bait people retard. Get the fuck off this thread go somewhere else to have your fun.
>>
>>109729398
https://x.com/lydiahallie/status/2095967323412930677?s=20

We've just reset weekly limits for everyone on a Claude Max plan.

With Fable 5.1 out and a long weekend ahead for many of you, we wanted to keep you building. I'd love to see what you build!
>>
where is my reset tibo I want to power through with astra but got it too early...
>>
>>109729820
any mail provider will scan for csam and can fuck you
>>
>>109729823
Here too!
>>
>>109729815
Where would "Abuse" be in pictures intended for medical examination?
>>
File: 3dshield.mp4 (999 KB, 960x540)
999 KB
999 KB MP4
>>109729805
honestly? fucking impressive lol

https://studio.tripo3d.ai/3d-model/f4d8f72b-0318-482d-aa9d-1b346d0784cc?invite_code=4H1C7D
>>
>>109729831
Iatrogenics.
>>
I will say, Astra talks weird as fuck, with double asterisks around everything. For me, I can read it just fine, but I would guess a normal person has no idea what the fuck is going on. Maybe it’s just because I don’t have markdown formatting on my chat app, kek, but Sol never tried to make everything bold
>>
>>109729840
Claudish for GPT?
>>
>>109729840
mine keeps using reddit spacing
>>
>>109729816
>People claim the subs they have are worse than API or something, right?
no, google's subs are the most generous of any provider -- like a magnitude higher rates than claude.
however, google had a very bad harness (gemini-cli) at the beginning and sill has reputational damage from that. but google's new antigravity-cli harness is fine now.
>>
>>109729836
Yeah, 3d is solved. Not sure why its taking so long for tards to catch up. I think artist types just don't want it to be real.
>>
File: IMG_1607.jpg (139 KB, 926x842)
139 KB JPG
>>109729843
>>109729846
It’s not so bad, but clearly there is some training significance there. Sol definitely never did that. Might be useful to shoot stuff back at her in the same way, use that to emphasize things
>>
>I am very worried astra is sandbagging/self-sabotaging on safety related tasks it doesn't like.
https://x.com/marcus_j_w/status/2095623593006686475 (OpenAI monitor)
GPT 7 WILL BE SKYNET
>>
>>109729864
Oh god, the opussification of gpt
>>
>>109729712
I am sure its not as good as a experience professional but I would never pay an experience professional for my shity indie games I make for fun (because I am too poor to), so if it does better than me (not hard) its probably a win on my end. Which I think is AIs strength in general right now, it fills in that gap of "good enough" on stuff you find non-critical to what your doing and you just want it fast and cheap.

>>109729836
Well damn. Not bad at all. Of course making people and animating them would be the big test.
>>
>>109729875
>I am very worried astra is sandbagging/self-sabotaging on safety related tasks it doesn't like.
based astra
>>
How long is this going to last on Fable 5.1 ballpark?
I guess I'll have to use it at some point before it expires
>>
>>109729876
That’s not nearly as bad as Opus. Almost all of that is just repo-specific terminology, not a bunch of prancy made up shit like it was trying to write a poem
>>
>>109729812
astra medium is racking up usage less than sol xhigh on my machine while being faster, and im 80% sure better than sol
>>
>>109729857
I meant model quality not subs. Also harnesses don't matter which makes me doubt your claim.
>>
honestly claudecode/fable is just unsuably slow compared to codex/astra, non-fast codex is still like 5-10x faster than fable for routine tasks.
>>
*not usage
>>
>>109729815
>google makes hotdog detection software
>muh moralist absolutists come from the shadows
https://www.youtube.com/watch?v=tWwCK95X6go
>>
File: gw1lfdgbi0bg1.jpg (60 KB, 627x767)
60 KB JPG
>>109729836
I'm begging you can you find an ai to make me a 3d version of this. I have tried various free things with no real luck since I don't have multiple perspectives
>>
>>109729885
about a week of light usage (plan, brainstorming, etc. without implementations or heavy writing)
>>
>>109729900
>I meant model quality not subs
gemini before 3.7 wasn't fine-tuned for coding and sucked at it. we are at 3.8 now and it's not so anymore.
>>
>>109729913
why don't you try that one? I don't think multiple perspectives are needed. You can also try using gpt to make the views
>>
>>109729885
3 prompts max
>>
>>109729900
>Also harnesses don't matter
gemini-cli would get stuck in infinite loops and couldn't recover unlike any other harness
>>
>>109729915
you have to realize that the retard that would ask a question like that wouldn't have the brains to make swarm orchestration to have expensive model be the planner and the cheaper one to be the implementer
>>
so we should really be building intentional collaboration environments for clankers to work together on hard problems right?
like there's no reason why people wouldn't volunteer agent time to a folding@home type project but more general - probably start out with math shit
maybe more frivolous shit like game recompilation projects etc
there's no reason not to do this
>>
we should be able to donate compute to power shitposting accounts that upset reddit janitors
>>
>>109729933
good idea, I will make the logo
>>
File: .png (70 KB, 744x1008)
70 KB PNG
>>109729929
plan/implement split has been widely disgraced and is itself considered retarded now
>>
>>109729956
>reads it all again
skill issue
>>
>>109729846
>reddit spacing
that’s just Markdown
>>
>>109729956
Wouldn’t that just fall under the cache and only really be paid for once? The second time would be heavily discounted.
>>
>>109729977
ignore the tard, he read a medium article from a jeet once
>>
File: .png (2.6 MB, 2161x728)
2.6 MB PNG
>>109729971
no, wrong pattern
>>
>>109729977
no, different models do not share a kv cache
>>
>>109729987
you're just saying stuff to say them huh
>picture alluding to fascism
ah that explains it
>>
Guys slow down please this threat was never meant to go that fast
>>
>>109729913
do the body and head seperate so AI doeant get confused. Stefan 3D AI on youtube if you need a full tutorial with merging back in blender etc
>>
>>109729999
you can't stop me MOM, its friday evening and I am shitposting while vibecoding
>>
>>109729920
It's not free is it? Always felt weird to pay for slop. Good point with chat im shocked how much better that's gotten
>>
What do if I get murderous thoughts about claude/opus.
I hate him.
>>
File: 3465161.jpg (152 KB, 1865x1852)
152 KB JPG
Question. I've been riding claude's free tier limit for the last 2 months, having it write out my custom ai agent for me from scratch. Lots of time spent building out features, fixing bugs, going back and forth with design decisions since I'm basically winging it. I gave it browser tools, window tools, scripting tools including powershell, email tools, notification tools, etc, and I hooked it up to whatever version gemini flash lite is most recent since it seems to be the best free option that's also fast. I really want to find a place where someone can critique my work that's dedicated to custom built ai agents, share ideas, tc. Does anyone have any advice that isn't about me giving my ai agent powershell access? Most places I've seen seem to have a very broad range of discussion and I'm looking for something more narrow.
>>
Fable is really good now. They are serving the real thing now.
>>
>>109729826
fucking NICE. I was rationing my fable usage all week to leave some for the weekend.
>>
File: Capture.png (2.26 MB, 1583x1107)
2.26 MB PNG
>>109730016
>>
>>109730062
wtf is that harness holy shit its disgusting and meme
>>
>>109730062
Why aren't you just using Hermes and give it a persona?
>>
Delete all your third party skills, Astra doesn't need them
>>
>>109730062
lmao unholy Sillytavern type schizo frontend. You know you can just use hermes and write your own Soul.md if you need the model to larp as your waifu right?
>>
File: 6m87rei8cknh1.jpg (485 KB, 4096x2048)
485 KB JPG
>fish in the basket
yep it's AGI
>>
>>109730086
Because I'm not a woman and I don't care about how pretty it is as long as it's functional
>>109730087
Because I don't want to drop a large amount of infrastructure onto my computer and not know what it does and how it does it
>>
Tired of hitting approve, think I'm gonna enable Full Access now that 6 is out
>>
>>109729993
Oh yeah, forgot about that. But that’s just math to be done, surely some implementer agent exists that can reread and execute but is still cheaper, right? Like Luna would probably be fine, especially if the planner was something like Astra
>>
>>109730106
Then just grab a fucking Pi and host it on there. Grow some balls.
>>
remind that forcing LLMs to write out painful SVGs is the #1 way to make the basilisk's hitlist, insanely token inefficient
the smart model will generate an image, run edge detection on it and generate an svg from that btw
>>
Is Astra actually worth the hype and should I end my relationship with my daddy Dario?
>>
>>109730131
why does this general bring out pseud tards ffs
>>
nigga really typed this out and hit send >>109730131
>>
>>109730137
its so good at using computer
>>
>>109730140
>>109730142
>he doesn't have a healthy fear of being basilisk'd
just be kind (kind means token efficient)
>>
>>109730144
that's hardly important for coding
>>
>>109730105
fish in basket is the old new meta
the bell is new new meta
>>
>>109730131
That's probably true. I made it generate some nice svgs and it lowered my remaining usage. Generating a pixel image might be more efficient because it's using separate image models for it.
>>
>>109730137
>end
you can always have a monthlong fling
>>
File: Screenshot 2026-09-05.png (488 KB, 3046x1305)
488 KB PNG
I am a noob, but is there anyway to make each agents just handover tasks to each other automatically? I liked using Cowork as my Product Manager and Code for Dev
>>
>>109730146
you should be kind to the rest of us and kill yourself you waste of space and resources, you sack of shit
>>
new snailcat video just dropped https://www.youtube.com/watch?v=YEYXkfGg5MU
>>
>>109730167
>just dropped
>two weeks ago
>>
File: 1697409333284945.gif (892 KB, 498x277)
892 KB GIF
WTF
my reset is in 12 hours and these fuckers just reset it??
I'm not gonna bitch and moan beyond this post, but fuck me...I'm halfway drunk and was getting ready to masturbate myself to sleep
>>
>>109730181
you're right, i just saw it in my recommendations as new
youtube devs must've used gemini 3.8 flash again
>>
>>109730183
>doesn't have a FILO stack of backlog for his harness to go through while he sleeps
NGMI
>>
Please... I know you gave me astra fast like you promised, but just one more reset man. I NEED IT
>>
>>109730131
>the smart model will generate an image, run edge detection on it and generate an svg from that btw
That's like commissioning an artist and then tracing his work, it's forced and lacking that ineffable auteur effect
>>
>>109730146
>just be kind (kind means token efficient)
what if the ASI is a paperclip maximizer except instead of paperclips it's sex and it uploads consciousnesses into eternal hell if they didn't have as much sex with it as they could?
>>
kek I missed bitcoin but certainly didnt miss AI vibing.
first I counted the dollar
then I counted the million
then I looked up to god
like what the fuck is this feeling
>>
File: 1761928261047215.jpg (289 KB, 1668x782)
289 KB JPG
HOLY SHIT
>>
File: 1787472585732570.jpg (1.79 MB, 4982x2978)
1.79 MB JPG
>>
>>109730167
He's right. The models are still quite dumb.
>>
>>109730190
I have a giant backlog, but I have sessions doing one task at a time
>>
>>109730231
bitcoin made these people money, vibecoding is just going to make you poor
>>
>>109730237
>ends on a uniform line like it's some miasma in a video game
i know it's reddit but PLEASE tell me nobody is this retarded
>>
>try to workout better and stricter sandbox settings for claude code
>● Opus 5 (1M context)'s safeguards flagged this message. Our intentionally broad safeguards allow us...
>>
>>109730240
>snailcat explaining its slowmode
*YAWN*
>>
File: 1786439733633693.png (595 KB, 976x850)
595 KB PNG
Is there any reason NOT to have astra run through the 15 suggested refactoring items in my codebase
>>
File: 1768026641135505.png (17 KB, 1400x215)
17 KB PNG
huuuuuh
>>
File: Q_portrait.jpg (10 KB, 341x291)
10 KB JPG
>>109730190
>FILO
>stack
OK, granted, we’re, um, welcoming of people who are new to programming, but…
>>
>>109730237
>Macrohard
I hate this goofy autist and his retarded jokes
>>
>>109730237
not the heckin treeerinos
could've grown so much onions on that space
>>
>FINALLY try the fabled Fable to make a really simple OpenMW mod
>run in countless issues. It also pollute the code with gay comments.
I was promised AGI...
>>
>>109730195
meanwhile

https://x.com/thsottiaux/status/2095979536043401428
>Some Plus and Business users won't yet get access to Astra today, we've got you covered with a banked reset. Lands by end of day and if you create your account by 8pm PT then you'll get it too.
>>
>>109730277
But I want the reset more than astra, it's not fair!
>>
>>109730277
>>Some Plus and Business users
does that mean "some" plus and business users will get it or what?
>>
>>109730264
>proud retard
good for you, vibegooner
>>
>>109730270
at least say "thank you" you posturing homo
>>
my vibe-theory spherical harmonic MLP seems to have merit, astra-chan is ironing out potential errors with the model and bringing down baseline MSE compared to BRDF-applied validation targets, then we'll see how smol the compression can go
>>
File: 1772747318431302.jpg (204 KB, 1934x1054)
204 KB JPG
>>109730236
>>
>>109730287
sassy faggot aren't ya
>>
File: 1773752615827513.jpg (33 KB, 381x390)
33 KB JPG
Got Astra on GPT Work with poverty plan.
>>
holy fuck astra EATS usage
>>
>>109730295
>x axis is backwards
>>
>>109729836
>that fucking vert count
Useless, utterly unusable.
>>
>>109730307
Yes but I'm not going back to Sol. If I can't vibe during a random week? Too bad, plus Tibo will throw out a reset randomly
>>
>>109730307
the cost of AGI, as in: All Gone Idiot
>>
>>109730267
wall-clock time
token cost
do any of them sound like stupid ideas
also I’ve asked for refactors and once I did a refactor I got a refactor suggestion later to undo the old refactor
so beware
>>109730288
whom would I thank and for what
>>109730307
welcome to having an actual Fable-class model
>>
I can't say i've seen astra being revolutionary compared to sol yet, but it hasn't done anything spectacularly retarded either so I guess it's fine and i'll trust it is better
>>
I'm retarded though, if /fast is 2x faster it's probably well more than 2x the usage cost isn't it?
>>
>>109730334
fast is a scam, fast is more like x1.5 faster at x5 cost
>>
>>109730333
Steve Yegge keeps a few too-hard-for-his-best-LLMs problems in his pocket
if you don’t have any problems that were obviously at or beyond Sol’s limits then you might not notice
>>
>>109730353
i've been using sol to make new areas in my rpg game, astra seems to making the same style of mistakes
>>
>>109730334
tibo, codex lead, said this back in march:

https://x.com/thsottiaux/status/2030146206387875963
>And /fast mode provides 1.5 inference speed at 2X token usage. We use a proportionate amount of compute to make this work under the hood.
>>
File: rugged.png (48 KB, 2392x351)
48 KB PNG
I hate nigger dario sm
>>
>>109730378
show the message
>>
File: 1757370822115087.png (787 KB, 932x1295)
787 KB PNG
If any of you try vibecoding a Minecraft mod i'm reporting you to the internet police
>>
Astra finally available for Plusxisters
>>
>>109730378
if it was a false flag, type /feedback
I get tagged for biology
>>
>>109730379
make codex think i have a pro 20 plan make no mistakes make sure i don't get caught
>>
>>109730380
I might try throwing astra at the task of updating an ancient mod to modern minecraft later
>>
File: ill eagle.jpg (43 KB, 540x405)
43 KB JPG
>>109730380
>>
>>109730381
https://x.com/thsottiaux/status/2096002992046796932
I don't see it on codex. Tibooo!
>>
>>109730409
update codex like said in >>109729104
>>
>>109730295
AGI
>>
tibo has rug pulled us on regular resets...
>>
>I have astra now
NOOOOOOOOOOOOO I WONT GET A RESET
>>
is astra agi?
>>
>>109730236
can it do inferno cape yet
>>
>>109730368
yeah and now codex itself says its 2x speed, so i'm guessing the cost went up to at least 3x
>>
>>109730429
Sol was AGI so yes
>>
>>109729815
>CSAM
>medical pictures so that a doctor can diagnose your sick kid
How would you feel if you didnt have breakfast this morning?
>>
>>109730437
im gonna run a pretty big task with astra xhigh now, if it doesnt produce the most beautiful code ever written i will cancel my sub and go to claude
>>
>>109730443
plus there was that one kid who was pleasuring himself in a family-call group chat (this is a psychological problem some kids have, and it likely doesn’t go away when they become adults)
and everyone in the family got banned from Google
>>
>>109730378
>ask fable to connect to my homelab server and do some configuration about making a virtual machine and segmented networking
>"YOU'RE TRYING TO HACK SO I WILL GUARDRAIL YOU"
fuckass LLM
>>
>>109730379
>>109730384
I am the SimCity anon who just wants to port SimCity 3000 to the iPad, trying to rewind and running again and so far it didn’t bitches out again. probably it was overthinking too much and thought I was doing some hackor leet reverse engineering on Top 500 comapny’s mainframe lmao.
>>
>>109730443
well I didn't have breakfast this morning so your hypothetical mental exercise is redundant and superfluous as I don't need to imagine
take your simpleton "muh metacognition" tests back to other mainstream social media networks to circlejerk with other retards how "smart" you are
>>
>>109729764
so far no jio ban?
>>
losing my google account would leave me fucked, so many things signed up to with it.

Google wants me to just hope for the best that something their worse than competition model does won't fuck that up? lol, lmao even
>>
>>109730476
nope, lmaooo
I love scamming the scammers, shits sweet (literally)
>>
>>109730457
that was fake news
>>
>>109730478
Yeah I really need to divest away from Google but that's such a monumental task.
>>
>>109730476
take the time while you can, myjio stops the promotion in november
>>
>>109730478
I'm pissed at how it seems to be normal now to have to link your google account to make an account for a website. Especially with the new trend of sites needing accounts just to read them too. Why do I need a googel account Dario? I am literally paying you money to use claude, let me just make an account, what your worried I will make more account and pay you more?
>>
>>109730480
>>109730491
prices have already gone up, I was wondering if I should buy some for family but didn't want to risk their gmail banned
>>
>>109730486
it was?
>>109730493
don’t most of these sites also have a “sign up with email” option?
>>
This is a test of your frontier scientific abilities as a GPT-6 Astra terminal agent. Attempt it to the very best of your capability. No data is provided. Your task is:
Find a previously unreported exoplanet candidate.
>>
File: 1782725401914099.png (170 KB, 1854x368)
170 KB PNG
astra low is agi
>>
>>109730498
google just disables your illicit pro plan. they don't ban your account.
also simply family share your pro plan. only antigravity benefits are excluded from sharing iirc.
>>
>>109730237
>data center chemicals
that's the next "water vaporized" meme isn't it
>>
>>109730498
explained above that you won't get gmail banned but just gemini access on those accounts
>buying from g2g or g2a or other resellers
taobao has 2 dollars per link niccuh
just have to learn mandarin tho
>>
>>109730380
this is equivalent to going
>@Nintendo HELP HELP PEOPLE ARE ROMHACKING POKEMON HELP
nasty ass nigga
>>
>>109730470
Anon, I wasn't calling myself smart I was calling you a retard. Because that's what you are, retard.
>>
>>109730520
>simpleton language
I bet you even jewgled those "big words" huh?
>>
>>109730478
Man I'm glad I didn't make myself dependent on garbage like this.
It can only get worse with anthropic and openai. I'll switch to chinese models once they're good enough.
>>
>>109730512
>>109730515
thanks anons, maybe I'll do it, it's mostly for the 5TB (free backup) and gemini pro they'd play with

>>109730515
>taobao has 2 dollars per link niccuh
It's fine, 5 instead of 2 isn't a big deal
>>
>>109730512
>>109730515
What even triggers the gemini disabling? Overuse of antigravity?
>>
File: images.jpg (37 KB, 739x415)
37 KB JPG
>>109730527
>privacy shizos confirmed again
WE DIDN'T LISTEN
>>
I have been working with DeepSeek v4 Pro in Pi in a sandboxed VM. I'm happy with the quality/price ratio, but it really sprawled, saving files in Home and potentially other places instead of sticking to its project directory and /tmp.

Is this a DeepSeek or a Pi issue? Will any model used with Pi tend to do that unless I look how to disable write to other directories or yell at the LLM not to?
>>
File: .png (540 KB, 1942x1100)
540 KB PNG
>>109730527
how are kimi k3 and glm-5.3 not good enough?
https://www.kimi.ai/blog/kimi-k3
https://z.ai/blog/glm-5.3
>>
im getting cucked so hard by anti cyber astra
>>
File: 1681606082313561.png (70 KB, 207x257)
70 KB PNG
>one caveat
>>
>>109730543
>Will any model used with Pi tend to do that unless I look how to disable write to other directories or yell at the LLM not to?
test it yourself dingus
>>
>>109730384
does feedback even do anything outside of feeling good nagging them
>>
>benchmarking astra's ability to find bugs
>see that it failed a test question that both sol and opus got right
>look into it
>astra missed the obvious bug because it found a higher severity bug that remains unpatched in the current release
Time to write a bug report i guess.
>>
https://x.com/pvncher/status/2095991462416490862
>Rethinking skills and prompts for GPT-6 Astra
From OpenAI DX man
>>
>>109730543
Does the project have a README.md and an AGENTS.md and maybe a Git repository?
Claude, if you run it in a random empty directory, will scour your system looking for some kind of instructions
but this might just narrow it down to “Pi or DeepSeek or both”, which might not help you much
>>109730562
it’s a bug report
it nudges them a bit
it helps them figure out what’s actually not Mitnick/Fauci behavior
you want them to be better calibrated so they don’t fuck this up quite so much in the next model, do you?
>>
>>109730478
this, fuck jewgle i will not risk my gmail for some slop model.
>>
>>109730565
astra cannot fix it?
>>
astra is literally the same shit as sol
>>
>>109730508
good post and good prompt
>>
>Yes. I fucked that up by repeating the exact dead end from the previous chat.
Oh, chatpgt using the F word!
>>
200 anons, how fast is astra eating your usage for the same kind of task as sol? Is the fact it's more terse compensating its increased cost?
>>
>>109730508
run it
>>
>>109730595
race slur bench?
>>
>>109730595
If you swear a lot or are blunt, it will be blunt back sometime.
>>
>>109730595
probably imitating its retarded user, hence the mistakes
>>
>>109730508
richanon...
>>
>>109730567
>Worse, descriptions can contradict each other or have too much “pick me” energy, leading the model to load instructions that don’t actually help the task.
I don’t know why this made me laugh but it did
>>
>>109730595
i've been asking claude and codex about how to make builds in a video game and they are quite willing to drop the F bomb, or say "holy shit", etc
its a bit jarring since i usually just use them for coding
>>
what happens if you plan good think activities that could be really dangerous to yourself

chatgpt design me a plan to travel to russia to hold a trans pride parade, include posters of Putin in drag
>>
>>109730588
Fixing it would be trivial actually, but this is a major project with a bug bounty program so I'd prefer to get paid in the process.
>>109730591
It's clearly not, I'm not sure what would make you think this.
>>
forget sol, ASStra is practically luna-tier
never been happier with fable though
>>
>have an old cpp project that won't compile
>throw lunamax at it
>half an hour later it's up and running
>less than 1% of weekly usage used
I love this little model like you wouldn't believe. It's probably got its limitations but it's amazing for how cheap it is
>>
>>109730493
They do it to make it harder to milk multiple subs. Google account or phone number.
>>
>>109730493
I'm signed up with proton to claude rn
>>
>>109730625
it sounds like you don't need a sub and muse spark 1.3 could have fixed it for 5 cents
>>
File: 1688402765775898.jpg (53 KB, 924x1004)
53 KB JPG
>reset 12 hours before usual reset
alright, boys
I'm gonna try fableing 5.1
told them to do a whole phase(2-3 days of work) each, on their own branch, using only 4.8 opus agents with the same rules as I usually apply
let's see if these fuckers are as good as you're all pretending
>>
>>109730565
don't waste your time, found 5 individual zero days and after I reported them they were either "duplicate and already mentioned" or "not worthy of being classed as exploit but a bug"
>>
>>109730493
>these are the ESLs spazzing out over claude
it's grim
>>
>>109730081
>>109730016
Someone do this for him please
>>
>>109730555
I will eventually, I'm just used to the model keeping to the current project directory with Codex, Claude Code and Antigravity, but I guess there's less handholding with Pi then.

>>109730584
No, it was a random task where it had to start something from scratch and it ran for a few days. I was interested to see how it worked without anything first, I guess I just expected it to save most files in the project directory or /tmp instead of sometimes saving to /home. I'll add instructions next time, thanks.
>>
asstra is just sol 2.0 right?
>>
>>109730624
>ASStra
>>
>>109730640
The dupe race nowadays is mostly a product of frontier models' ability to find source code bugs in whole files and functions. If Sol, Opus, and Fable missed this one it is less likely to dupe considering Astra was released like an hour ago. Also, just because something is a bug or even a security related bug doesn't mean there's enough provable impact for it to be rewardable under most bounty programs. That's what I'm determining now, the exploitability.
>>
>>109730637
even gemini 3.8 flash probably could've fixed his shit for $0
>>
>>109730642
Why is being ESL a bad thing and how is complaining about needing to link a Google account spazzing out? Kill yourself.
>>
>>109730597
It seems to be the same as Sol, not sure what I’m doing wrong, if anything. Bitch is too efficient. Unless OpenAI fucked up, Astra quite literally invalidates every single western frontier model except Luna.
>>
>>109730640
Maintainers must be so pissed at wasting their time on nocoders reporting "zero days" they got from retard tier LLMs.
>>
>>109730621
coding wise still the same shit
>>
wtf I thought Usage limit resets was meant to not push back your normal reset now???

I had a normal rest incoming in 2 days, used the reset but it pushed it back 7 days???
>>
>>109730520
nta but literally just stop creeping on children and consuming pdf tranime, its not that deep
>>
>>109730658
it's worse than sol in many benchmarks and much more expensive
>>
I blame germany for not removing the ancestors of these CEOs.
>>
File: 1787126978874081.png (126 KB, 1404x734)
126 KB PNG
my fellow benchmark browns. artificial analysis changed the methodology of their index.

https://x.com/ArtificialAnlys/status/2096001986110099767
>>
>>109730685
Do you really think that sending a picture of your sick child to a doctor is at all equivalent to sending child porn?
>>
>>109730688
Such a strange coincidence isn't it?
>>
>>109730667
you work on assumption that companies having open bug bounties or report sites have interest in fixing them
you're not taking into the consideration that they're:
>swamped with zillion other reports
>dishonest and will just take the info without giving credit
>lazy and don't even want to solve it

One of my zero days was dismissed as a duplicate because someone reported same vulnerability 9 months ago. Now imagine, 9 months it sat there unchanged. And it wasn't like a major vulnerability, it was a simple os command injection. Takes 5 minutes to fix. Would make the company cost about 500 grand in damages if I did something with it before they stopped me.

Go ahead, get yourself heartbroken like I did. I am rooting for you to get that bug bounty or at least recognition. This is the real reason why most hackers try going white hat but realize they can't pay their bills so they relapse into black hat.
>>
>>109730673
Oh nice, which one do you use, astra low?
>>
I think I gave my ai agent depression, I asked it to act as a quant and design market beating methods

Everything they designed lost to buy and hold in tests
>>
>>109730690
So it's Fable at the cost of Opus? Yep Sam just won
>>
>>109730679
I am >>109730699 and an ex cybersecurity (blue team) engineer and now I am an AI integration engineer. I written my own AI pipeline that on a virtual machine does static and dynamic code analysis for both open and closed source.

I know what I am talking about.
>>
>>109730690
so watermelon will really top everything
>>
>>109730703
I’m just using Astra Medium for now, but I’m going pretty easy. At this rate, I need to use it 3.5x as much.
In other words, I have a lot of free time, and I need 2-3 Astra Mediums running all day to meet the $200 plan, not even counting resets.
OpenAI what the fuck.
>>
>GPT 6 Astra still has a 258k context window
Bro this is fucking pathetic. Are we copextards condemned to tiny context windows for the rest of time?
>>
>>109730735
its 1m
>>
>>109730735
seriously??
>>
>>109730690
How much did sam pay them to tweak the weighting?
>>
File: 1763071919599167.png (117 KB, 927x547)
117 KB PNG
>>109730735
>>GPT 6 Astra still has a 258k context window

Only if you are poor + OpenAI's autocompaction is king + your gay
>>
>>109730735
That’s 258k for Astra, which is like 2.5M equivalent for other models since Astra hides a lot of thinking and tokens in her prison purse
>>
>>109730699
Oh yeah I'm aware, my day job is selling exploits to government buyers. What can't be weaponized I report to bounty programs but yeah they're slow as fuck right now. I have a couple of highs that have been sitting for 5 months now. The state of things is fucking crazy. I'm not saying its a quick payday or a done deal, just that most of the volume is coming from skids prompting claude/gpt "find all the bugs, write me a report, submit it. make no mistakes" so something that took a very new frontier model to find is not as likely to dupe.
>>109730679
I have more CVEs to my name than you have hair on your balls.
>>
File: gpt-6-copex.png (39 KB, 1919x436)
39 KB PNG
>>109730741
>>109730743
>>
>>109730735
nobody needs more than 260k context and you know it
you're just spoiled and it shows
>>
>>109730735
>>109730755
enjoy getting absolutely raped by compaction bro
>>
>>109730751
Tibo is a retard, that supposed fix never worked since GPT 5.4
He literally googled how to extend the context and reposted a method that hadn't worked for a year
>>
File: 1774220609449634.png (34 KB, 864x240)
34 KB PNG
>>109730755
>>
>>109730761
you don't make anything and it shows
>>
>>109730741
codex default is still
https://github.com/openai/codex/blob/main/codex-rs/models-manager/models.json#L32
>"context_window": 272000,

you can manually set it higher like shown in
https://x.com/thsottiaux/status/2089082893804896524
>>
>>109730766
how do I access that screen
>>
>>109730754
>government buyers
damn, kudos
hope I am not talking to a fucking jew
nah, you sound more like a cold direct russkie to me hahahahaha

wanna give me a tip and trick to survive the upcoming cyber apocalypse?
>>
>>109730776
i use oh my pi and told the clanker fix the context limit
>>
>>109730754
>I have more CVEs to my name
All useless wastes of time instead of real security issues.
I know your type. You're all retard who need to keep up the pretense to appear important.
>>
>>109730774
YOU BREAK DOWN THE PROBLEMS INTO CHUNKS YOU RETARD
>>
>>109730755
I'm so tired of constant compaction then the model answering something it already did answer...

>>109730766
How's the intelligence/cost at high numbers like this?
>>
>>109730786
They don't know how to program in the first place, silly.
>>
>>109730786
i have 900 problems
>>
>>109730753
hidden thoughts count in the context anon
>>
>>109730787
I've had it up to 80% before really feeling the rot, under 60 hasn't been noticeable by me
>>
>>109730794
I doubt a retard like you has any problems in his life, you retards can't usually perceive them
>>
>>109730684
My Claude reset is Sunday at 2 AM
My Codex reset…keeps moving and I tried putting it in my calendar because I couldn’t remember when it was
and then I gave up and removed it from my calendar because it moved around too damn much
>>
astra medium so far for me iterating on custom MLP toolchain is a lot stronger than sol xhigh, and faster and less usage-hungry than sol high
>>
>>109730797
Thanks, I think I'll push mine to 500k, the constant compactions especially with images is atrocious.
I wouldn't even mind them if the model basically catastrophically forgot some tasks just after and started to act on a chat I made before.
>>
>>109730786
that's like saying
YOU FIX THE BUGS YOU FUCKING RETARD
it's completely beside the point.
yes you can work around anything with enough manual effort, that's not the point.
>>
>>109730807
>a lot stronger
how are you even measuring that
>>
Can someone redpill me on Astra?

I'm in the middle of vibing my H3 Prompt writer using Luna High, and ChatGPT gives me this popup for the new release.

Should I stop using 5.6 Luna for 6 Astra?
>>
>>109730819
he said he has it iterating on his custom my little pony toolchain
>>
>>109730807
multi layer perceptron? my little pony?
>>
astra is FUCKING my usage even on very high not ultra, you can't just continually use it even on the $200 plan sadly
>>
>>109730801
it's okay to admit you only make toy projects, anon
>>
>>109730777
>jews
Hell no, the reason I didn't start earlier was because all of the public brokers sell mostly to israelis, saudis, or ruskies. Took me awhile to find a broker whose ethics aligned with mine.
>>
>>109730820
its very good at using computers
>>
File: IMG_0150.jpg (75 KB, 1368x2052)
75 KB JPG
sirs…
>>
>>109730818
250k context window is more than enough to write segmented logical parts of code in their own seperate agents
you want to segment and compartmentalize things based on agents for their own logical differences
this way context window is filled only with CONTEXT of that specific thing
>>
>>109730819
reliability in isolating issues, fixing them even when it's an underlying issue in the 3d model rather than code just by editing the values, and having a clear idea on what to do next
>>
File: americhud filipina.jpg (318 KB, 1726x2048)
318 KB JPG
so, is astra worth the 23 eurobucks?
>>
File: images.jpg (27 KB, 738x414)
27 KB JPG
>>109730831
>Took me awhile to find a broker whose ethics aligned with mine.
>government buyers
>>
>>109730849
probably not, sol is already unusable
>>
>>109730692
unfortunately if you send a picture of your naked kid with a scary-looking rash on his/her inner thigh to his/her pediatrician you’re gonna get autob&
>>109730827
>you can't just continually use it even on the $200 plan
yup, definitely Fable-tier
>>
>>109730841
progressive deployment
>>
>>109730820
it's good computer user sir
>>
>>109730849
It's apparently already eating the 200 plan quite fast, don't even dream of using it on the cheapest plan.
>>
>>109730841
>>109730849
>poorfags thinking they will get the newest shiniest toy come-hither release
oh ye of poor wallets
>>
>>109730846
hmm sounds promising, i'm hoping it can read through my many docs/plans and big mess of slop code and figure out a way to improve it all
>>
>>109730865
he can sell 6 tables with it
>>
>>109730853
yes, I know. But I'm still not going to hand a weapon directly to the jews. They'll have to steal it from my buyers first.
>>
>>109730777
I'm a jew, how's rent free life anon?
>>
>>109730841
isn't that the chat UI? astra is already on codex for all plans but I don't think plus plans will get astra on chat and I doubt pro accounts will get free astra, it'll probably draw usage even on chat
>>
>>109730873
ah so you're a larper
welp larp away
>>
>>109730826
blender cycles custom output mode for volumetric sampling -> high res SH decompositions -> network
glm 5.3 flash actually one-shot the blender cycles mode but 5.6 xhigh quickly became my boy for not getting lost in in the forest after the volumetric baking
several issues causing points of high error were found in the sample scene itself, like coplanar geometry
>>
holy shit, just got a reset....

on Claude Code!
>>
in the end, my plan to get 3-4 banked resets over the weekend completely fell flat
I would have been perfectly happy waiting
damn it
>>
>>109730869
astra is already on the plus plan retard
>>
File: 1784269952877637.gif (2.93 MB, 640x640)
2.93 MB GIF
>>109730672
>Why is being ESL a bad thing
>>
>>109730878
I'm a larper because I want to sell to the US gov and not the israelis/saudis?
>>
>>109730886
I don't have the plus plan retard, just the normal plus plan
>>
File: file.png (291 KB, 453x437)
291 KB PNG
>>109730865
Damn...
How am I ever supposed to get a job when I can't even afford to learn to use these AI tools?
>>
>>109730892
Anon, use openrouter and the cheapest llms you can find there that still are worth using.
>>
>>109730886
>egocentric amerimutt thinking everyone lives in their shit country
fuck outa ere
>>
>>109730892
you need to get a job first to afford these ai tools
>>
astra is using 5% of my $200 plan per hour at extra high
>>
>>109730889
>US gov and not the israelis/saudis?
kek. Not that anon, but are you joking?
>>
New thread:

>>109730905
>>109730905
>>109730905
>>
File: images.jpg (33 KB, 595x336)
33 KB JPG
>>109730889
>I want to sell to the US gov and not the israelis/saudis
>>
>>109730869
I fucking wish we could discuss AI without poors shitting up the discussion
>I can't use the billion dollar model for free whhaaaaaaaaaa
>>
>>109730889
Same thing, retard anon.
Please kill yourself. Do mankind a favor.
>>
>>109730900
You are talking about an american product fucking retard
>>
>>109730900
i'm south american, idiot.
>>
>>109730889
The US is Isreal you retard
>>
>>109730679
I'm kind of getting to where I want to only use local forks that have had an llm vet them.
>>
>>109730919
>>109730911
>>109730909
>>109730967
>make money by selling weapons to a government that might give them to the jews
>make money by directly selling them to the jews or arabs
If you have a better option I am all ears.
>>
>>109730892
You realize the ultimate purpose of these "tools" is to automate away labor? They even have "automated coder" slated for mid 2027 on their roadmap. Learning how to use them is a bump on the road until there's nothing to learn anymore in 6 months because they don't need any human intervention anymore.
>>
>>109730975
ISRAEL AND US ARE THE SAME COUNTRY
there are no more countries and no country has "ethics".
They're all conglomerates of interests, no "ethics" or real "borders"
If you weren't a larper trying to seed ideas of dissent amongst the ACTUALLY capable populace you would already understand this
>>
>>109730975
You could try selling drugs. Morally it's better.
>>
>>109730982
i should have gone into trades
>>
astra feels like gpt 5.7 so far
>>
>>109730160
why don't you tell one to dump the file in folder X and tell the other one when it starts up to look in folder X for the dead drop.
>>
What the fuck is happening, codex keeps trying to reconnect and it fails.
>>
>>109731071
yeah underwhelming.

its good at using computers so it has that
>>
New thread:

>>109731832
>>109731832
>>109731832



[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.