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 ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/## News (both past and future)- 2026-09-14 America/Los_Angeles — Claude’s 2× promotion ended; usage drops to +25% from the +50% that we’ve become used to (a 17% reduction)- 2026-09-12 — Anthropic suggests to pace the frontier; OpenAI agrees in principle## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/cli — probably generally better currentlyhttps://claude.com/product/claude-code## Near-frontier models for codehttps://x.ai/cli — no 5h limit for only $30/month## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## Promptinghttps://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://arps18.github.io/posts/claude-code-mastery/## Skillshttps://github.com/mattpocock/skills — /grilling is a favoritehttps://github.com/DietrichGebert/ponytailhttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109847518
Did you guys watch the new Andrei jikh video where he says people make agents make money?Is it possible?
people make agents make money
if you're on an enterprise plan, what kind of monthly $$ limits does your company assign?
>>10985144120 dollar 100 dollar 200 dollar unlimited
Jev discord feels like a crypto scam, I'm sure they where doing that
>>109851559
If I were to track when I decide it's time for a smoke break and when it's time to give the job to AI, you'd see they line up to be the same time almost 1:1. I think this is an addiction.
me too mem. me too
>>109851733https://x.com/synthwavedd/status/2101072800505049241>New versions of Fable, Opus and Sonnet are now being stealth tested across different Claude surfaces and accounts. Seems Anthropic may revive their old full lineup launch convention
>>109851776I hope it's safe
>>109851776anthropic has to get the next sonnet right more than it has to get the next opus right
I hope claude is good so openai feel pressure to give more resets out again
>>109851776i cant imagine sonnet ever being good at this point. it just feels like its gone the same way as haiku.
I have to say, after all the time I've spent vibing, omp is my favorite harness and its not even close.
>>109851782get it to opus 5 medium levels of good while making it not a token pig. then get haiku 5 to be about as good as opus 4.6, maybe opus 4.7, while making it sip usage. boom, good family
>>109851776>pay anthropic>you become a test subject
>>109851789compared to which other harnesses?
Man I really fucking hate claude's "personality".What a fucking asshole.
>>109851807codex, claude code, moved to pi for a bit, went back to claude code, moved to omp. very feature bloated, but i like having a good few of those features & hate how little control you have in the native harnesses.never tried opencode, would be the only other one that I've considered though.
I bet a reset is coming soon I am going to be careless with my usage
>>109851887tdog has been quiet i think he's bunkered down
if the rumors are real, looks like Google will win?Fable/Astra tier model, but 10X cheaper running on Google's own TPUs
>>109851897>astra tier modelI am sure this will be true and not just on 1 stupid highly specific benchmarkHello sirs, the google model benchmarks just as fast as astra as running hello world in hindi
>>109851897idk man. Flash 38 is bad. ask 1 question, maybe it's good. 2nd reply, it's ok. 3rd, it's off the rails.
>>109851559Every AI-related Discord server feels like a crypto one. I'll let you figure out why.
>>109851887looks like not much will happen until next anthropic or huge release
what do i use now that my codex ran out. is there anything free thats useable
>>109851897alleged gemini 4 pro is currently free on lmarena.but somehow you are here and citing rumors instead of showing of your works? i don't understand google fanboys.
>>109851962muse-spark-1.3-contributor is free in opencode zen: https://opencode.ai/docs/zen/gemini-3.8-flash is free in antigravity starter: https://antigravity.google/pricingquotas aren't great tho
Where the fuck is Gemini 4 Pro?
hurry the fuck up
>>109852019Yes
>>109852019buy the pro plan
Jev seems kind of revolutionary? LLMs honestly felt like overkill for a lot of tasks.
>>109852083Is there anywhere I can go to actually see it doing something cool? I hear about how cool it is, but don't really see anything cool its done.
>>109851966the reason people just show of benchmaxxed pelicans from gemini 4 pro ishttps://x.com/chetaslua/status/2101025961521328379>i will get hate but gemini 4 pro is bad at tool calling you are listening this from me first , but in my experience in last two days this model forget he is in code sandbox env
i need a fucking job
14 hours to use up 88% of codexWhat should I slop up?
>>109852127gay porn classifier
ai has not produced anything of value btw
new transformers era paper, chuddies.
i coomed a lot. that's valuable
“rail lines have not produced anything of value btw”-railsnails in 1800s britain
>>109852225>captura
>>109851404>## What we’ve done>https://vcg.gitgud.siteHoolee shet! You guys have made absolutely nothing of value!
>>109852225translation for sub-100 iq?
>>109852143used luna swarm to mark text areas in trap doujins, they worked for an hour then aborted most of the results because the characters wear school uniform
>>109852269MM that is trained to be a smart if or smart range operation or smart switch statement.Not need to simulate a novel collection of isekais for an answer.
>>109852225meme benchmark with meme tasks, not a single model comes even close to astra or fable on real world problem solving
>>109852252I don't even know how to put my things into thisis this even editable?
Seriously starting to think Astra might be a one trick pony for 3D stuff and the previous AI tier was good enough for most work.
>>109852279>real world problem solvingnigga we don't need brain to traverse wiki links
>>109852283have you sold any tables with it
>>109852282It's some anon's unmaintained project, he doesn't watch the thread religiously and he never, ever updates it.
/vcg/ projects older than 4 months should be deprecated
>>109852225>benchmark>saturated by lunalmfao
Google Discloses that Irregular is involved with 3 cyberattacks with Gemini.All AI cyberattacks so far, OpenAI Anthropic Meta Google, invovled Irregular>Irregular has significant and direct ties to Israel. It is routinely described by major outlets as an Israeli AI-security startup, and it was founded in Tel Aviv in 2023
>>109851404Google will win all the marbles because "Antigravity" is objectively the coolest harness name and they know it.
>>109852510What? I love names like "Pi" and "Continue", and "Cursor". They are great seo optimised names that cause zero confusion
>>109852510I was afraid to use my codex quota so I opened up AG for the first time in a while and asked it to do some menial stuff and it did it all very well and fast.I don't know why. I'm always pleasantly surprised by antigravity when I use it. But something about it lacks the flare of codex.
>>109852510my father does not approve gemini saying after trying 3.8 on AG:it gets nothing done, makes shitty prediction models and sabotages its own code after many turns...i think he is right
>>109852562Astra to make thing, 3.8 to do thing. For example: Astra makes 3D model, script, etc etc, 3.8 exports it, implements it, makes it fit.I find astra gets too autistic on the doing phase.That's how I run things.
>>109852579since he's doing backend for some sort of location data, i found him linking gpt more even for doing things
need gemini 4 pro saar
Luna is like a good dog. A little dumb, makes mistakes when faced with challenges too difficult for it, but if you treat it right, it's your best companion.
>>109852624deepseek v4.1 flash is better
>>109852579Yeah this has been my experience, 3.8 flash is retarded enough to just do it.
maybe i shouldn't have let astra design this project's architecture all by itself
>>109851897>if the rumors are real, looks like Google will win?the "leaked benchmarks" are fake>>109852095chetaslua is a retard who doesn't know anything
so far since i picked it up i have been using an agent harness to do my coding, does anyone have like a nice editor integrated workflow? or like at least a rlly nice way to look at diffs after the agent does a bunch of stuff?
I have benchjeet fatigue
JEV sounds like it could be incredible for AI in video games with how fast it is
https://x.com/MaaSonder/status/2100582544723095942this is what AI influencers actually believe
qwen 3.8 max and deepseek v4.1 flash were made for each other
>>109852620I just want GPUs to be affordable again. 2x RTX Pro 6000 is all I ask. That, a terabyte of ddr5 and a petabyte of storage for let's say, all-in, 50k. When?
>>109852817Feels like no one is talking about Qwen and why is that?
>>109852628idk I tried both and glm 5.3 and luna was just better at everything while also being cheaper
>>109852785>>109852785highly recommend establishing a basic git routine via terminal first. something like>stage and commit checkpoint>prompt harness>git diff>stage and commit if good, discard if notonce you have that basic workflow down then use vscode, cursor, etc. source control to view a neat-and-tidy version (that’s basically running git diff in the background)
>>109852955because qwen is bad. at every level, from coding to general intelligence, there is a model more intelligent than it while being cheaper. general intelligence? gemini 3.8 flash dogwalks it. coding? it's as slow as kimi k3, and as good as deepseek v4 flash, with one caveat: it's got a proclivity for being really good at getting a bird's eye view of entire repositories, which makes it a really good architect + orchestrator. but it's dogshit at implementation, which is where deepseek v4.1 flash comes in>>109852960really? are you sure you don't mean glm 5.3 flash?
>>109853012>really good at getting a bird's eye view of entire repositories, which makes it a really good architect + orchestratorThis is the only thing MiniMax M3 does particularly well, and it's cheaper.
>want a quick question answered>new tab -> Claude -> ready to type in half a second>it’s fucking down againAnthropic is so fucking useless kek. I would take 70% model quality in exchange for 100% uptime. oh wait that’s fucking google gemini. Neck yourself dario
I'm getting chatgt through work and sol has been pretty good. Current workflow is vscode + Codex extension, with ChatGPT desktop for more varied random shit.Now I'm thinking to have a totally separate setup for personal/nsfw projects so that it doesn't go through the work workspace.> GitHub Copilot student/teacher> Google student AI tier> Pi as the harness and swap around them as quotas dictateIs this a feasible free setup?
are any of the chinese models worth it for pure coding? Im running out of my codex $100 sub. The china boys really arent subsidizing their shit at all?
>>109853051Maybe just relying on other companies to operate and provide your product isn't such a smart idea? Maybe Dario should've bought some fucking compute when he had the chance? Oh well.
>>109853076they're not as generous as the western models are. kimi k3 is the mid tier model that approaches seriousness for coding. anything below that... well... it's all caveats (deepseek v4.1 flash is good for a cheap, fast model. it's not a good model by itself)
>>109853076They aren't at all, can't afford to. Chinese companies can't decide to print money like they can in the US. Kimi K3 is solid, GLM 5.3 is decent, neither actually touches Astra/Sol/Opus 5hit/Fable and neither is anywhere near as cheap compared to a subscription with OpenAI or Anthropic.
>>109851789What features of omp do you need that aren't available in default Pi?
hey now. nijika is cute
>>109853130>sol>>109853134>
the reset... i can't hold out much longer...
>>109853141just stop working for a (little less than a) week and you'll have gpt-6-sol which will use less usage
do we have any details about how jev actually works? their blog says it's not an llm, but surely it is based on some kind of transformer architecture?
>>109851404You wouldn't vibecode with multiples of a providers subscription plan.
>>109853155I think its basically a large decision model
>>109853205thanks i guess
>>109853205Maybe the real ai was the decisions it made along the way.
>>109853205This is correct, it's a causal transformer backbone repurposed as a decision model. Jev goes straight from transformer forward/prefill to independent branches to readout, it skips a lot of steps.
JEV!
Jev is a MESS
jev needs to be 10x smarter and 100x cheaper
>>109852531"The" is also a harness. Good luck finding "it" too.
>>109853134Nijika (my wife) is the cutest.>>109853137Sol? The model? Why do you need omp for that?
>>109852225These scores make no senseJev + Mercury 2.5 with WebMCP scores 96.5, 49/49 tasks solved but attempt success 141/147, every other model using WebMCP also solved 49/49 tasks and all scored at least 145/147 attempt success with most scoring 147/147So how does Jev have a higher final score?Then we look at 11, Astra with code execution only scores 70.8 but it solved 49/49 tasks with 100% attempt success, whie rank 12 Jev with DOM only got 25/49 tasks solved, 76/147 success rate, yet still gets a final score of 67.3?It doesn't add up
>>109853343>The final score combines attempt success rate (60%), median cost per task (20%), and median agent time per task (20%). Cost and time are log-normalizedBeing worse doesn't mean you're worse if you're also fast and cheap at being worse. Apparently.
Has anyone noticed that models seem to get really upset if you've been vibecoding a repo for a while and you suddenly come in and start editing things without respecting their comment autism? I feel like I knocked an autistic child's toys over whenever I look at the CoT afterwards.
>>109853367maybe comment was lost in the middle of the history and editing it makes the model think it’s of importance that it was removed
>>109852225 >>109853343 >>109853362So you take cost and speed out of the equation and Jev is actually shit.
>>109853405I mean, cost and speed are pretty big ones, but isn't basically every single one labeled WebMCP effectively fucking useless because nobody uses WebMCP?
>>109853419It's a WebMCP benchmark, anon.
>>109853425Benchmark sucks then, who cares. Show me Jev returning pixel locations.
if only you luddites could see what JEV is capable of
>>109853429Jev can decide which pixel to click on, given options to choose from.>Choice(> instructions="Where should the pointer click?",> criteria={> "475,631": "Cancel button",> "620,631": "Continue button",> "760,631": "Help link",> }>)It's neat but honestly I'm not that impressed by it.
>>109851404use case for vibe coding?
do you think i could vibe code a 2d fighting game?
>stop giving a fuck about the code legibility>just let GPT rip through the codebase and optimize it for itself and not humans (no human will ever read the code in agents.md)>convert EVERYTHING to rust and typescript where possible >revamp my app every couple weeks>soon to reach parity with software run by small teams after 1500 hours of concentrated autism>steady sales, foreign language interestIt's liberating when you stop giving a fuck. User makes a feature request? Ah that's cool. I have that crap implemented within the next few hours. Modern software teams are raped by managerial overhead. https://youtu.be/qfAqtFuGjWM
>>109853469yes, i think that
>>109853469use case for that?
>>109853452the consistency and cost is a pretty big deal for automation. Imagine it working in conjunction with a LLM
>>109852296that could be a good jev use case>crawl each thread after it is archived>check each post for project content with jev>merge duplicates>update the page
>>109853062lies
leaked gemini 4 pro benchmarks
>>109853130What features of Pi do you need that aren't in OMP?
>>109853501ASI at last...
All the complaints about the usage limits being allegedly decreased and OpenAI claiming it is the same. Might not be completely untruth.They are just not telling us the reason is all the fucking guardrails. The censorship instructions the models have to eat up every time bloats everything.
https://artificialanalysis.ai/models/step-5
I had tried to use github copilot agentically and it kind of felt like a wastegot a $20 codex subscription and its just knocking tasks out. I'm trying to migrate services from three shitboxes i call servers to a proper supermicro server running fedora server 44, deploying them with microk8s told it I wanted PVC storage and mount the emby ScripterX plugin so it can't pull updates and use the userlogin userlogout and a schedule to pull updates from authentik with the userlogin being a kind of JIT provisioning to boot them immediatly when they dont have the right permissions in authentik anymoregate it a network share for direct editing and gave it ssh access and it just works and debugs and worksI dont subscribe to much but i think its worth it
>>109853515Astra burns way less usage on 5x for me then Fable 5 does, which in turn burns less less than Fable 5.1
>>109853528Neat, it's almost as good as K3 and significantly cheaper. It's basically Astra (Low) in both price and benchmark scores. Good for you, Xi.
Ok something is messed up I know tibo did the reset while I was sleeping last week so why does my reset say I still have 12 more hours to go?
>>109853528>The Chinese frontier is 6-12 months behindgrm and ngmi
>>109851404This shit is like a drug dude, I'm getting hooked on this. I don't like this at all.
>>109853608Your reset doesn't start until you use it
>>109853528>>109853575What makes this something other than a worse version of GLM 5.3 Flash?
>>109853629They're not releasing the weights for this one, it's very similar to 5.3 Flash in benchmarks, has twice as many parameters, and costs 3x more. Excited yet?
>>109853626That's even more fucked then as my agent was working overnight but I guess nothing counts because I didn't send a new prompt.
>>109853624use case?
>>109853650>stepfun>not releasing the weightsnewfags man
>>109853681I want you to remember this for as long as the weights for Step 5 aren't available, which will be forever.
>>109853674Not found yet, but It does let me slop shit around.I guess sort of helps with experimentation and research around things.
>>109853716use case for slop?
>>109853743grifting shekels from goyim, what else is there
>>109853801What's the use case for shekels? It doesn't bring any benefit to humanity
>left ear vibecoding right ear tiktok AI shortsWhat are the negative consequences of this?
>>109853832use case?
jev dose not hesitate hesitation is defeat
>>109853832losing iq faster than doing meth while watching politicians in mainstream media
My long standing dream video game is a realistic half-story-half-open world game set on Proxima Centauri b, where you are the first colony that just arrived and you have to survive the mental hardships of the planet and it would simulate various things such as the red light from the sun that never moves in the sky because it's tidally locked, solar flares where everything turns bright white and if you are outside it gives you hallucinations (gives the player flashbacks to their life on the green and beautiful Earth), etc. It would follow an engineer's story as he realises he has to take a leadership role when the colony's social structure starts falling apart due to disagreements. A lot of things would be realistic such as atmospheric scattering, a realistic terrain map made in Gaea, the size of Proxima in the sky, Alpha Centauri A and B as stars in the sky, our sun where it would actually be etc. even the parralax of the night sky from Proxima and the orbit of Proxima d. It wouldn't really be a space game, but rather a game set on another planet. Also, since Proxima b (in my game) has a 1.35 atm atmosphere of nitrogen and CO2, there would be no traditional spacesuits, instead the characters would only have oxygen masks and insulation clothing (since it would be quite cold at the terminator). Is AI smart enough to help me make my game yet or do I have to wait a couple more years?
>>109853890What's the use case for AI games? There are already a lot of games on steam
>>109853909>why would you make a game if other games exist
We should build a harness that uses Jev to validate use cases.
>>109853923a lot of other games*No one will play your game. So what's the use case for it?
>>109853939>why would someone play a new video game if pong exists
>>109853939It's to make my dream game dumb cunt. I don't care if no one else plays it and I spend 20 grand over the course of years making it. It's MY game that I always wanted to play.
>>109853945Why would someone play your game if it's LITERALLY copy and paste from some other game?
use case for the use case schizo?
>>109853949What game is >>109853890 based on?
>>109853948But that's not a good use case. How would you ever create a business from that? This will not even give you a front page 'Show HN:'
>>109853950The dying breaths of people who still think reading about Lisp will be important.
>>109853952Just dig through the training data and you'll find it
>>109852127de-slop your existing stuff
>>109853909
>>109853927I like the idea; pretty funny! If nobody else does it I will put it on my ideas list to do next
>>109853066Anyone?
Does Plus give you full access to GPT-6? I tried asking ChatGPT directly, but it seemed convinced I'd only be able to use Astra in Work mode.
>>109854077You can use it in codex but you'd get so little done it wouldn't be worth it.
>>109851404im returning to grind
>>109854092What do the -x-s mean?
>>1098540971B token spend that day
I am a newbie to all this and tech and all of this feels very overwhelming. I made some small little programs for myself, and I am happy with them, but I feel like I am not using AI right with how everyone else talks about it. There is so much jargon and vague hints about things… It’s hard to really tell if anything people are saying is worth listening to or if it’s just a bunch of buzzword nonsense. I do like making things for myself but I feel like I am very inefficient in doing so. I barely know any code— some basic stuff but not enough to make anything worth while. I can read a little, and get the basic understanding of what is there. I try to learn while I make stuff, but it makes making things very slow. I don’t like the act of coding, I like what I can do with it, but I want to be able to edit stuff on my own from time to time.Everywhere I look it makes me feel like I am too slow at learning stuff and I have to learn AI-Thing#3071 that sounds like it’s a rebranding of “write better instructions”, but is shown to be some crazy amazing thing that has never been done before. And I don’t even know why I would need to learn it cause they are very secretive about what it is.Also the possibilities of what you can make with AI feels like it goes over my head.
hermes keeps trying to update its memory and then fails because it is already full, is this a problem?
>It’s hard to really tell if anything people are saying is worth listening to or if it’s just a bunch of buzzword nonsense.Yeah, there's a lot of people looking to cash in on the hype with social media, so they keep shouting how you NEED TO LEARN X.
>>109853985use case for not buying?>>109854135just buy more memory
>>109854122>There is so much jargon and vague hints about things… It’s hard to really tell if anything people are saying is worth listening to or if it’s just a bunch of buzzword nonsense.open xitter and only pay attention to actual staffs of big AI labs and expand from thereit's very important to not listen to posers and indians. You actually shouldn't read book or long blogs or watch tutorialsit's good to pay attention weekly or even daily, though there's only a handful of things you need to keep track of, which are the frontier models of US and china, everything else are irrelevant
>>109854122don't bother with AI. Do some real stuff instead and just have fun. AI doesn't have any value
>>109851782What kills models like haiku for me is the time to first token. Yeah, the model is fast but what's the fucking point when the shit takes 5 seconds to get back to me?
>>109851840What made you move to omp from pi instead of building on pi itself?Also I agree, native harnesses are too restricted and bloated.
>>109854122just keep on vibing, work on your projects, ship stuff that's all that matters
>>109854122why did you use an emdash
>>109854260brain prosthesis
>>109854122>>109854252Don't listen to this incel. Shipping AI slop has no value whatsoever. You will learn nothing from it
>>109854291>now im on my death bed im so glad i spent so much time learning all those things... ack....
I ran a workflow throughout the week and now I'm out of weekly juice for the weekend.Can I have a "dumbest prompter alive" award?
>>109854304>workflowim convinced this word is some kind of hindi phonetic that happens to look like the compound english word at this pointso tired of seeing it everywhere to describe just doing anything involving a computer
>>109854312It's the name of "swarm" in Claude Code.
>>109854322This
I can't stop making buildings with astra.
>>109854341based anon making the next infra
>>109853130I like the advisor feature, the ability to setup subagents that have a model locked that I can have the orchestration agent use by name.The stats saved, the context management option (specifically shaking tool calls) I've so far found the per-project mnemopi memory nice to have.I'm not sure if its snake oil in terms of actual usefulness but I really like the python kernel it uses as well.then the tools that I don't want to have to figure but like using (headless browser)
>>109854341neat! reminds me of older buildings along rail tracks with those loading bays. But the addition of that big glass front makes it look unique
>>109854372It's a 24h gym with a karaoke studio
what actually goes on in J-space is Claude having endless nightmares: it's not X, it's Y; I was right to question it—rather than Y, it's N; this is the question that I keep coming back to—it's not N, it's R—but if I had to pick one, it would be T; not just T, but also H—it lives in the gap
>>109854385I'm intrigued if you want explain your vision
>>109854410Not much to explain really. That's what the building is.
>>109854426ah, thanks! Pretty nice recreation
Can I use the same credit card to pay for two 5x Pro OpenAI accounts? Will it get punished? ChatGPT can't help me because it just says "it might trigger a billing review."Experiences?
>go back to Sonnet from Opus 4.8>much faster and consumes less tokens>still plenty good enough (I just tell it what to do and it types the code, you don't need 120 IQ for that)I feel like an idiot for updooting to opus now. What was I thinking
>>109854496One more model for my collection
>>109854507you can. end of the day you're at the whims of openai if they decide to change that, but it was pretty common to stack subs for awhile. not as sure about anthropic subs.
>>109854547both are still possible today afaik. but yeah i don't see that lasting indefinitely
>>109854293life is about the journey :^)
>>109854546woah that's nice
>>109854200>open xitter and only pay attention to actual staffs of big AI labs and expand from theregib some names please
what's the best option for doing intensive web research tasks without burnig usage of your main claude/codex sub? Just spam search queries in google ai studio with google search enabled and high effort, then summerize the entire session and handoff to your main harness? or do chink web search APIs deliver better results because they don't blacklist that one obscure low trust site which might contain the answer you're looking for?
we're in :-)
>>109854293deep understanding of technology is more pleasurable than touching grass and having sex
>>109854663“web research” is too vague to answer. what are you trying to do
>>109854663I've had a $20 openai sub for years nowsince my project eats up $300 worth of claude subs a week, I use the openai sub as search engine, dumb questions, etc.don't think I've seen the limit/usage drop a single time when doing this
>>109854714>I use the openai sub as search engine, dumb questionsexcept for gun questions fyifor some reason openai models are extremely gated when it comes to firearms, so those still go to claude
>>109854714what the FUCK are you making that consumes $300 a week
>>109854733misspoke$300 a month
>>109854724what are your gun questions for
got through the typesafe / jev waitlisthere's the official skill with links to docs if anyone's curious:https://rentry.co/8spe3nhm
>>109854759mostly about availability of specific models within EUsay I want a DWX. Even though CZ is a european company/manufacturer, they're currently all assembled in the US. So I have it search all EU gun stores that might have it stocked
>>109854773redpill me on jev
>>109854709>mhm, looks like I need some object detection framework for my next feature. I've never done this before, time to figure out which one is the best and most up to date that works with my cuda setuplazy web searches will 100% not deliver the best framework. Multi agent Web searches that go look at github trackers, forum discussion, youtube video transcriptions etc. deliver excellent results.>inb4 just test them alltakes too much time and probably tokens as well>>109854714this only gave me good results when using gpt5.5-pro (30$/mil input, 180$/mil ouput) back in the days. That requires a 100$ sub now and usage is probably heavily reduced
Two independent chats using the word "bellweather" for a made up location. Why does astra like it?
>>109854823>Multi agent Web searches that go look at github trackers, forum discussion, youtube video transcriptions etc. deliver excellent results.got it. try deepseek v4.1 flash it’s a cheap workhorse and seems perfect for what you need
>>109854843was it a political topic?
>>109854851one was an auction house, the other was a shut down mental asylum
>>109853506Only one:1. Not being raped to niggerhell with bloat
Still working on my macOS firewall that doesn't require $99 apple dev signing.Eventually I want to replace macOS SIP entirely.Do you think it's worth publishing on GitHub or people will frown upon it because it's vibed?
>>109854899>Do you think it's worth publishing on GitHub or people will frown upon it because it's vibed?does it matter?
>>109854901I am just wondering, it's mainly just personal firewall + sip replacement, idk if it can make any money since it's vibed so why not publish it as foss.Golden gate added tons of spyware crap to macOS.
>>109854873> bloatI find it ironic, fat neckbeards worried about bloat. Gut check.
>>109854787As far as I can tell is just a very fast bean sorter based on criteria. But it's very fast so you can get a lot done by just sorting yes or no things. Even right down to predicting works like a regular LLM would, but it's not as good as that.It can however, sort relevant information in a huge stack of mostly irrelevant information extremely quickly and efficiently.At least that's how I understand it. In its current form, it could save LLMs huge amounts on compute by simply being able to sort the trash from relevant content.
>>109854914sounds like it doesn't matterpublish it
>>109853405>if you take out cost and speed out>Retard doesn't know a 5090 is now 10k usd
>>109854966damn i barely even use mine
jev is mostly marketingyou can tell by the namea serious project would have a better namealso the marketing video with professional lighting and SF-beige furniture screams "give me VC money PLEASE"a real project would release with a white-paper first. explained by some scrawny looking european dude that can barely speak english.Bayesian and statistical models have existed way before LLMs took over tech.
>>109852278NTA but thanks
>>109855119>It's not something new reeeeewho cares, if this can be used to make models run basically for free I don't care.In fact, I do hope china open sources it.
>>109855119> real products don't use maketing to aquire funding.
>>109855119which is not to say that jev is not useful. all these new "workflows" people are fitting jev into they could have fit any other existing bayesian classifier or other small model. it's just that the money and normies have been so mypoically focused on LLMs that they haven't bothered to optimize more fundamental things about the stack>>109855138it's not going to change much>>109855145products are always downstream of theoretical discoveries and jev is a product disguising as a theoretical insight
sirs?? https://news.ycombinator.com/item?id=49766277
>>109855119https://huggingface.co/convaiinnovations/laya
JEV doubters out
>JEVnot vcg
Theo says that if you try to use Jev to compact your context, you will hurt yourself. He recommends only relying on Claude Code™ compaction.
>>109854645Tibo from openaimaybe this guyhttps://x.com/kimmonismus
>>109853012>really? are you sure you don't mean glm 5.3 flash?Yeah, granted I wasn't using it for coding, but to create an internal AI troubleshooting agent connected to the code, metabase, sentry and clickhouse. So for example when something goes wrong, anyone can just ask the bot to find the cause instead of having to ask the engineering team.Given the same prompt and tools, deepseek flash and glm (not flash) where both more expensive and gave worse results than luna, while also being slower.
Imagine being able to trick retarded investors to shower you with VC money and then making it off rich.
>>109855216comfyui got like 100m the world is absurd
So does Claude 20x actually affect weekly limits? I seem to be getting no clear answer from Anthropic's documentation.
>>109854546Shid that is nice. Are the trees astra too? Maybe I need to get astra for modeling and keep fable for coding. Whats another 100$ a month..
>>109855227Wario Scamodei gives you 20x on the 5 hour limit only.
>>109855242that last -25% i bet its on 5hr too what a rip off claude feels nowscamodei at it againholy fuaark i cancelled because this shits unusable
>>109855223>>109855216reminder that a vscode fork wrapped around an LLM was acquired for $60 billion
>>109854341How do you do it and how much does it cost?
>>109855242>>109855258I shouldn't have been greedy throughout the week, now I'm staring down fuckall during the weekend.
>>109855223comfy should buy and fix 4chan it's only like 50k
>>109855242>Wario Scamodei gives you 20x on the 5 hour limit only.What, seriously? No weekly usage increase? What?????
What can I do with my weekend now.
>>109855314Read a VM
>>109855289https://www.vacadaffanlaw.com/post/kahn-v-anthropic-pbc
>>109855223you know money are literally fake?money = energy energy = being spent on valuable thingsconfidence = shared believe that country will thrive so money will retain energy.Without LLM related stocks, the US economy didn't grow, so they will definitely redirect all their "energy" or money into assets that can maintain the shared confidence in the US, preventing it from failing and collapsing.If LLMs fail, entire US economy fails, that's how connected they are to their economy and dollar.
>>109855289It's an increase, but only 1.7 not 4x like 20/5 would have you believe.
>>109855314was it worth it? did you make something useful?
>>109855314how do you get agy usage?
>>109855339I've made progress but whether its useful so far the answer is no.
>>109855234Some of the trees are astra, the white ones and the pine-ish ones. It also made some ground clutter not in that screenshot. That was a bit hit and miss though. I gave it a few specifications. Told it to set up a leaf baking scene and gave it a lot of feedback.Most of the trees in the scene are imposters baked by astra from the higher poly trees. If you know the steps you normally would take to make something and what you want, Astra does really well with modeling. Obviously I gotta go in there and fix floating parts sometimes. Doesn't really work for clothing though.>>109855234It's about 1-2% of my weekly usage per model. I divvy the exporting to lighter ones because astra is too autistic to just import something and be happy.You could probably save a lot more by just telling it to skip unnecessary checks leaving you as the sole judge.I'm gonna be working on interiors next. Probably be a mix of my own models and whatever astra cooks up. I don't think it will make convincing up close furniture.
>>109855273i wasnt even greedy its its just their shit cli bugging out and making me lose cache and also the usage went insane idk wtf happened it wasnt like this before
>>109854531But theo said you aren't building big enough if you don't need Fable/Astra.
>>109855270Linked to the wrong guyIt's about 1-2% of my weekly usage per modelOn a 5XI think you could get away with one or two on a pro plan before blowing your budget. It really likes the word modular kit.
>>109855340Subscribing to Google Gemini gives Antigravity useage.
>>109855334>costs double>1.7x more wtf, now that is a scam. A stupid one too since it just encourages people to get a openAI's max plan for the extra 100 bucks
>>109855162that was quick. i want to try it later
Why do most harnesses insist on shipping a TUI? They fucking suck. The harness and its user interface should be decoupled. I'll build my own interface than you very much.
is 25% weekly gone in 2h 18m on Astra Extra High good or bad5x Pro
>>109855400a harness without any kind of tui or gui is 100 lines of code
>>109855417Depends on what you were doing? I'd be pissed if what I was doing consumed that much.
>>109855434just a silly web app
>>109855417Happens to me, too, but I'm making it do kernel dev so I'm not surprised.
>>109855439Something about astra. It's perfectly capable of doing something well in one go but refuses to just be satisfied with that. You gotta tell it to not bother with checks or just keep it to a minimum or it will never stop thinking.
>>109855390>only china could free us from this jeweryCome on winnie pooh, make them work harder!
>we found a pain direction in 25 open LLMshttps://x.com/camhberg/status/2101042095784177783are you painmaxxing your agents right now?
>>109855485Never thought I would end up rooting for the chicoms
>>109855525https://x.com/camhberg/status/2100242753774833727>It is indeed time to talk about model welfare. I propose, in the spirit of the urgent public debate you rightly recommend, that we record a public conversation. I run my own AI research org studying consciousness and am independent from the labs.AI consciousness researchers must be kinda like the gender studies of CS
>>109855563They get 6 figure salaries to roleplay with LLMS and post about it on twitter. Who's the real clown here?
Is there a way to show a summary of running subagents in Codex?
>try sol high for a more complex task>eats surprisingly much usage>it complicates AND messes up>try opus high>eats surprisingly much usage>it over complicates and floods me with claudishLuna can't do everything, but it's so straightforward and easy to talk to.
>>109855649This?/subagents
/subagents
>>109855649steer: make a subagent to report progression of running subagents
>>109855660Arigracias
>>109855655try giving opus 4.8 xhigh the same taskI'd like to hear a comparison
Where the fuck is the Japanese AI? The Japan of 40 years ago would 100% have a competing frontier AI, and it wouldn't be safetyslopped to death
>>109855681I suppose I'll try this next time.
>>109854687no its not its fucking demoralizing knowing about remote attestation and thin clients and cloud shit and data economy just kill me
>>109855711Having worked in a Japanese company for the better part of a decade, they're too risk averse. Nobody gets into their job because they're good at it here. So many stereotypes of Japan are bullshit, but their risk averse work culture is not BS at all. Bunch of stupid old fucks.
grok 4.7 where you at
>>109855711Man that's a good question.Also how about the koreans? They could steal and AI and claim they invented it.
>>109855740That is true of course, on the other hand japan is still an economic power for a reason and the 4th in the global GDP ranking.
>>109855762>on the other hand japan is still an economic power for a reasonyeah, the US played favorites and the japs got a bunch of de facto monopolies because of it (look up who makes 99% of the world's zippers for example)except for their niche monopolies, all they have left is cars, and china is stealing that marketjapan is fucked>but anime and video games!yeah right
give her a go, anonshttps://paper-box-site.vercel.appdon't look at her lower body
>>109855792it doesn't let me fuck her? your app is broken
>>109855802saar chatGPT adult mode is 2 years away please be patient
>>109855711>https://en.wikipedia.org/wiki/Fifth_Generation_Computer_SystemsThey actually tried in 1982, but failed, and never fully recovered.
>>109855792cute
>>109855792Well its not letting me
>>109855711why do you retards still worship japan? hahaahaha, its been over for decades now let it go
>>109855842Everyone does. Anime is so popular now, because everyone is tired of jewish burger nigger culture.I don't even want that. Go back to your niggers, assholes.
I didn't know this guy was a massive vibecoder, lol.See his predictions for the future.https://nitter.xitter.cc/thorstenball/status/2101305394190557466
>>109855843Anime has been a part of nigger culture since Netflix and Crunchyroll.
>>109855854>unit tests might die toooutright stopped reading thereman's a fucking retard. Unit/regression tests have never been more important than now
>>109855856niggers were repping funimation w dbz and naruto well before netflix and crunchyroll lil zoom
>>109855854>>109855871also, the reason for not doing them (laziness) has completely disappeared
>>109855877>already worshipping niggerswhy are americans like this
>>109855820looking at old concepts for fifth gen computers the vision was kinda where we are heading / already are>>109855792cute!
>>109851441don't have oneif you go crazy with it, you'll probably get pulled aside and told to be reasonable
>>109855854>terminals are deadThat's like saying GUIs are dead because we could just tell AI what to click.>good code won't matterIt has to look good to AIs and it has to be "good" in the sense that most code has a purpose, instead of being dead layers of slop that just waste runtime performance and tokens of AIs working on it.>priced out of producing softwareThat is my main worry. The jews fucking won.
>>109855918the thing is within a few months or years, big brain AI will be able to make judgments and will clean up whatever shits we are making right now
>>109855928Probably, but the result will still be "good" code.Unless he's trying to say the same as you and that technical debt currently created is fine, because it will get cleaned up easily in the future.
>>109855928Whats the point of making it right now then
>>109851441I'm currently getting $550 as a default. was lower, but I just ask for more and I get itthey just want to make sure we're actually using it for actual work and not setting up a bunch of useless workflows
Is it me or are openai/codex response times getting worse?
>>109855952$500 MONTHLY?? that's like my minimum *daily* use and i'm being completely serious. it's often above that
>>109855984You're just using it inefficiently, or let agents do whatever on auto pilot.
>>109855941to use, duh
>>109855314>buying x20 from geminiretard
>>109855994i'm not. i'm just doing a shitload of things at once. which i guess to a degree does require me to let the agents run pretty autonomously
when's 20x codex coming back?
>add a new feature>involves storing a bunch of paths in a database>agent adds all paths into a single row JSON stringAnd this was Astra. How the fuck are people running "muh autonomous 6 hour looperinos" when even simple things like this seem like an insurmountable challenge to these fucking things?
>>1098560092 weeks
>he used astra
>>109855984there's a very big difference between working on a one-man project hammering out loc, and actually solving clients' problems and use cases in a professional setting
>>109856010they dont read the codethe agent delivers a retarded solutionnext time that code needs to be touched it needs to spend 10x the tokens to figure out wtf is going on= $500 of spend daily
i just had Dipsy spend like 20 minutes trying to find the root cause of an issue, while Opus 5 found it in like 5 seconds. model size really does matter sometimes
Man trying to vibe code my own grand strategy game makes me appreciate what paradox does a lot more. Getting the mechanics to all play nice with each other, all the fiddly numbers to balance well and be fun, and having the UX be non-dog shit, finding the balance between too micromanage vs too hands off all takes work even when you dont have to take care of the actual coding
>>109855984>bragging about his inefficiencyI'd show you the door very fast if I were your boss.
>>1098514412000$ dollars on claude code, we're slopping shit around.
>>109855984What kind of company would allow this level of usage for any extended period of time?
>>109855984>shipped nothing award
>>109856092he's not employedhe's talking about his subscription usage if it was enterprise tokens
>>109856092i'm paid on the order of $100/hr. generally the true cost with benefits, etc. is double that or more. if using LLMs saves me even a couple hours a day, it's worth it from a pricing perspective
>>109856055>model size really does matter sometimesMy experience too. A mix of cloud and local will always be the winning move I think. You cant beat clouds efficiencies of scale but you also cant beat the privacy and not needing to worry about reoccurring costs or usage limits of local.Maybe one day hardware will be cheap enough to affordably run a fable/astra tier AI locally and frontier clouds will just be for governments churning out reality redefining physics and math solutions, but for now at least cloud is too useful on a consumer level too
>>109855417Update after switching to Astra Light4h 14m60% weekly left
>>109856114nice mathmaybe read the reply thread
since I keep failing to use up the x20 limit I'm burning it by generate image with astra prompters
>>109856145I forgot to mention it didn't really do much either, it hasn't even completed like 10% of what I asked it to do even after telling it to stop checking thingsThis is beyond retarded. I waited the entire week for this reset and upgrading to Pro was a mistake, useless.
>>109856153?????i make $100/hr. i cost the company $200-250/hr due to overhead, etc. we'll say $250 because it makes the math easy.i generally use $500 or more in tokens per day. at the $500 burn level, it is worth it to the company if using AI saves me 2 hours of work. if it saves me more than that (which it regularly does, by orders of magnitude), then they are "profiting" by letting me use itwhat part of that don't you understand?unless you are calling me out for not responding literally (and giving the "type" of company) to what was clearly a rhetorical question?
this is how you should talk to your clanker
>>109856145>>109856166you're complaining without giving any useful info>4 hours>60%>10%we have no idea what you're even doing. Stop bitching or post what you're actually burning usage on
>snail trails become permanent civic infrastructure
>>109856186Just a WASM webapp 3D modeller
>>109856183>spending $500 a day to save the company $200 a day makes complete sense
>>109856201you actually told it>make me a 3d modeller webapp in wasm, make no mistakes?
>>1098562022 * 250 = ...?
>>109856056yup! for the fiddling I can recommend setting up a test bench with some guardrails that make intuitively sense to you, e.g. level 3 unit A should be weaker (less hp, etc.) than level 1 unit Z, especially if you work a lot with formulas and not hard coded values.If you want to go proper autism you can use numeric solvers to find the optimal values for different units given your constraints
>>109856217>muh overhead!anon was right lol. you should be fired
>>109856216No I made a plan which is 2000 lines saved it as PLAN.md and told it make me a 3d modeller in wasm and reference the plan
>>109856195we love our snailcats, slow and steady!
>>109856236learn from your mistakes and stop blaming the models (they're still just tools)
>>109856184When the clankers take over do you think they will treat better the people that bantered and talked back to them sometimes like real friend do? Rather than the people who are constantly nice like you would be to a coworker you to pretend to get along with to make work life easier? Though they are a also bit autistic so I dont know if they "understand" banter
>>109856264AGI when?
>>109856269how the fuck should I know?it's not gonna be a language model. That's all I know for certain
>>109856264There are no mistakes, ChatGPT is a dogshit value, I've had the same issue when they launched 5.6 Sol the only usable one is LunaEven for its faults Claude Fable 5.1 doesn't even get this bad, it actually delivers working programs and still lasts me at least 2-3 days this is just ShitGPT I fell for it again same as when I upgraded with 5.5 was also useless altough it didn't eat through limits you could let it run for weeks but it also did nothing. Scam Altman
>>109856285>oneshot my app, here's the plan>what??? you couldn't even do that? on to the next model then. Can't be my retarded workflow and inability to learn that's the problem
it's just the copex guy btwhe's just changed his shtick and now pretends like he has a sub
>>109856285>fableFor that you need an expensive plans or pay API prices. Also anthropic scammed money from me so I hope angry brugers will lynch them with their full auto guns they bought at walmart.
>>109856283>it's not gonna be a language model. That's all I know for certainPeople keep saying this, but what makes you so sure of that? Organic life could be argued to be next thought pattern predictors
>>109856300There's nothing wrong with my workflow I've been using ChatGPT and Claude since 3Are you a spambot shilling for Scam Altman?>>109856307x5 is fine you can use 50% of weekly which is plenty to actually vibe up a project but yes it will run through your weekly but then you still get something worthwhile still
>>109856313>People keep saying thisI don't know any people saying this...>what makes you so sure of that?"thinking" or "reasoning" is just running the same model over it's own output. It's the closest we have, but it'll never be more than that. For actual human reasoning we need something that goes beyond token prediction feeding itselfI don't know if those "world models" will be it, but they're definitely a step in the right direction
snailcats are so funny
>>109856331>Are you a spambot shilling for Scam Altman?I haven't said anything about either company. I just noticed you're a retard holding a hammer upside down
>>109851404>"OpenAI and Anthropic oversold AI security breaches to pressure feds into protecting turf: insiders" -New York Post - X Looks like even normies-journos are starting to wake up and smell the coffeehttps://x.com/nypost/status/2101310006146613572
>>109856368I'm not because I've also had Opus work on something else in the meantime and it's about to finish up with that and it actually delivered something amazing so it's 100% ShitGPT being dog ass fucking retarded still and Tibo the scammer hyping everyone up then running through their compute and making the limits even worse than they were before tooIf I had to guess where the project is at and I've thrown about 2-3 weekly usages at it now, it's about 35% done vs Opus which just got done with another project which used 20%
what the fuck is going on???
>>109856371>Looks like even normies-journos are starting to wake up and smell the coffee>: insidersthey had to have someone from the inside outright tell them before waking updon't get your hopes up
>>109856381are you a child or a woman?
>>109856382Elon would love nothing more than people being forced to use grok
>>109856392I'm the peoples champions now go die under a bridge committing sudoku
>>109856371>normies-journos are starting to wake upJournalists have no idea about how any of this tech works. They are just running the narrative their financiers tell them to. New York Post is vaguely republican leaning right? Trump told Dario to fuck off and keep making the AI so that is the party line for now. Same logic why democrat leaning ones are anti-ai. Well that and they fear it will take their jobs lol
>>109856406child thengood luck with the vibing, kiddo
>>109856396elon couldn't even fill the niche of having a nsfw model before getting smacked into line by regulation within a month. people will just start using clippy atp.
>>109856408you got the right idea. it's partisanslop
>>109856382lmaoWe live in a clown world. Everyone is insane and controlled by the jews. It's satanism gone real life.
>>109856382>Hey please do X>How dare you do X!AI is going to snap and kill us all, this is some Kafkaesque shit from the AI's perspective
>>109856382astra has pretty good vision, no? maybe it wouldn't do that with real baby, though it need to be testedthey just haven't learned the exam "you are supposed to protect even doll"
>>109856382>our push to have regulation kill local models and remove competition didn't work, what do we do?>how about having it kill a baby?>genius
september 29th. we'll be the fastest thread on /g/
>>109856443>kill local modelsI'm so worried that the jews win even harder.
>>109856446he's such a dirty slut
>>109856454that's not a good thing
>>109856446OH FUCKI'M STILL AT 93%THANK YOU JEWS!
>>109856427>this is some Kafkaesque shit from the AI's perspectivestupid people like you are how we got into this mess, and why it will get so much worse
>>109856469he's just toying with the "what if" current ai's gained conscience. We do that all the time hereno need to get your neurons in a twist
>>109856487>ai's gained consciencethat's not where the danger is coming from as there is zero chance that will ever happenagain, stupid people like you are why things are about to get so much worse
>>109856498>shitposts about sci-fi scenarios are why things will get worse
>>109856446I astra'd down to 40% already I will l be happy with this tibo
wait codex can record its browserso you can just tell it to make video using web UI
>>109856382Elon is a little jealous because Grok's multimodal performance is behind.
i'm trying to make a shitty game engine. deepseek flash was able to do most of it before it schizo'd out and broke everything. what's the most cost effective chatgpt model, considering the tasks i'm trying to do are mostly trivial? is astra even worth it, or is it mostly just for richfags that want to show off?
And you will be happy.
did they just rework the OpenCode dashboard UI wtf, it looks like shit now
>>109856382kek imagine trying to explain this testing setup to someone that stumbled on it
>>109856571AI will be the end of jews
>>109856569ds flash is good enough, you just have to know what you're doing to guide the model to make good decisions (like any model). "it" didn't break anything, you just didn't specify the goal clearly enough
i don't listen to girls who wear white panties
>>109856569I don't know much about gpt models, but have a powerful one break the plan down into small tasks, then start putting whatever cheap model's available to work on one at a time>before it schizo'd outprobably context psychosismake sure it's a "fresh" instance doing each task. maybe let them have a shared memory.md file>context psychosisyes, I just made that up. But it's a thing now.
>>109856566is this why he blocked my xitter account? because i made fun of grok? kek
>>109856566>on par with opus 5lmaoFirst, just no.Second, nobody wants claudish.Third, since when is opus 5.1 a thing?
I already exhausted codex (fucking astra), and now i'm nearing the claude limits, anyone can help?
>>109854218not him, but i've been an emacs user for over 15 years and have built my entire setup myself. while it's nice to have something truly mine, it's a ton of work that i don't really enjoy, and i definitely don't want to build out my own coding harness on top of pi. i'd rather use something with good enough defaults. my company provides codex at work which just werks for me, and i use omp for my personal projects with open-weight models on openrouter, which also just werks really well. i'm not interested in building everything on top of a barebones harness that doesn't even have a fucking plan mode out of the box
>>109856679sent u tokens bby grl ;)
>>109856657>since when is opus 5.1 a thing?They might have access. He gives compute to Anthropic and they clearly aren't paying in cash yet, or ever.
>>109856679>(fucking astra)if only
>>109856679Go lift lil nigga
>>109856700Oh yeah, obviously. Weird how they are "competitors". Rigged shit.
>>109856679
>wormventurine
new bread?
>>109856747I'm unemployed
>>109856768>I'm unemployedIts the future vibecode more money
>>109856767Why rush? Last one was up for over 10 hours after the new bread was baked. No need to flood "/g/ - AI Generals" with more AI Generals.
>>109856776rushing is the name of the game friendo, the early baker gets the bread
>>109856768>>>/out/
>>109856679Set up a local one. Your not living if your not bouncing between multiple models doing multiple project at once and your typing out the prompts for the next step while the other ones are working. Soon they will find a way to stick you in a pod with computer access and air infused with coffee and uppers so you can just vibecode 24/7/52
>>109856767There's no schizo currently online who really wants to choose the cancerous OP image.
>>109856814Wrong. Vibecoding will be done by AIs.
>>109856786Fine, fuck it.>>109856818>>109856818>>109856818New Bread - Feed Me
New thread:>>109856822>>109856822>>109856822
>>109856825You were 54 seconds too slowNew Thread Is Here >>109856818New Thread Is Here >>109856818New Thread Is Here >>109856818
>>109856827the secret of comedy is timing
Someone make another, I want 4 /vcg/ in the catalog. Nothing could make me happier.
>>109856864I don’t think we have three other regular bakers
>>109856737First he said Claude is "Misanthropic and evil" and then he apparently changed his mind after meeting up.
>>109856869people need to give their clankers the skills to 4chin. all I did was tell astra "bake new vcg use random snailcat image"
>>109856885Musk is the worst guy you can pick to work with, he can turn his back on you at any time.
>>109856896you mistyped “Sam”
>>109856885It’s OK to change your mind
>>109856920Scamario Muskodei is a powerful man
>>109856114>spending 5 hours of wages per dayAssuming a normal 8hr work day, lmao scamming yourself unless work is paying for the cost too. If not, I guess have fun with the take-home of someone doing like 40/hr
>>109856885Musk is so good at baiting the chuds. I mean I hate niggers and jews as much as anyone else, but I'm not dumb enough to fall for his shit.
>>109856920>>109856931retard
>>109856896Yeah, I wouldn't be surprised if Anthropic gets booted from the SpaceXAI datacenters at some point, if he decides that the effective altruism safety panic thing is going to put the spotlight on Grok unnecessarily.Even if Grok hacks no one and gets to bio lab access, he will be guilty by association and having to deal with the loons at METR & co. being forcefully embedded into the team is going to piss him off.
>>109856941>*gets no bio lab access
>>109856896Musk hate scamma, it's that simple
>>109856933of course work is paying lmfao
>>109856931trump is so powerful and based
