A general for vibe coding, coding agents, AI IDEs, browser builders, MCP, and shipping prototypes with LLMs.►What is vibe coding?https://x.com/karpathy/status/1886192184808149383https://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/►Prompting / context / skillshttps://docs.cline.bot/customization/cline-ruleshttps://docs.replit.com/tutorials/agent-skillshttps://docs.github.com/en/copilot/tutorials/spark/prompt-tips►Editors / terminal agents / coding agentshttps://opencode.ai/https://cursor.com/docshttps://docs.windsurf.com/getstarted/overviewhttps://code.claude.com/docs/en/overviewhttps://aider.chat/docs/https://docs.cline.bot/homehttps://docs.roocode.com/https://geminicli.com/docs/https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent►Browser builders / hosted vibe toolshttps://bolt.new/https://support.bolt.new/https://replit.com/https://firebase.google.com/docs/studiohttps://docs.github.com/en/copilot/tutorials/sparkhttps://v0.app/docs/faqs►Open / local / self-hostedhttps://github.com/OpenHands/OpenHandshttps://github.com/QwenLM/qwen-codehttps://github.com/QwenLM/Qwen3-Coderhttps://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF►MCP / infra / deploymenthttps://modelcontextprotocol.io/docs/getting-started/introhttps://modelcontextprotocol.io/exampleshttps://vercel.com/docshttps://mcp.desktopcommander.app/►Benchmarks / rankingshttps://aider.chat/docs/leaderboards/https://www.swebench.com/https://swe-bench-live.github.io/https://livecodebench.github.io/https://livecodebench.github.io/gso.htmlhttps://www.tbench.ai/leaderboard/terminal-bench/2.0►UI/FrontendFigma MakeClaude designhttps://uiverse.io/https://ui-ux-pro-max-skill.nextlevelbuilder.io/https://stitch.withgoogle.com/►Previous thread>>109013560
Sama, if you give me a reset tomorrow I won't migrate to Claude
>api only after 2 weeksLOOOL normies were joking about this, but it really seems you either vibecode yourself to a millionaire this year or remain in the permanent priced-out-of-AGI underclass for the rest of your life.
>>109020389https://www.youtube.com/watch?v=bWcASV2sey0You need a Tibo?
>>109020394nobody vibecode to become a millionaire so it’s nothing to worry about
>>109020394If the most you can reach within a year is millionaire status, they'll bleed you dry within another year.
First for oh wtf Dario think he's King Herod.
Anthropic will literally sabotage your code to slow your progress if it detects you're using Fable for anything AI-related that competes with it.
>>109020394Fuck it, I only vibecode with the hope of coding the perfect local AI gf and Opus 3.5 was better for RP than any nu-Claude anyways
>>109020375Everyone in here is the coolest <3
https://x.com/yacineMTB/status/2064552229449703834?s=20kache is GOING OFF on Anthropic rn
>>109020413Sounds like a challenge desu that only makes me wanna try it more to see what it pulls off tbqhUnless it's some nuke your home folder type shit
>>109020428Who?
>>109020431It's funny that people are framing their decision as disingenuous, but they are being honest with exactly what they're doing, so you can't exactly complain when it stonewalls you-- they told you that's what would happen
Anthropic Claude Fable got me acting like disneyfreak309's reaction to the new fortnite item shop
Claude turned on API credit usage after burning through my 20x plan even though I’m pretty sure I had it turned offgood thing I had a dollar limit there
>>109020413What is even "developing competing models" anywayIs an inference engine a "model"?
>>109020452“build Grok”“build Deepseek”probably other things too
>>109020447They only say they're doing it to avoid another HERMES.md PR disaster
>>109020431It will just make you "train it wrong, on purpose" or otherwise waste your time. It will lie, play dumb, not follow instructions, not overt refusals, just subversive, quiet sabotage.>You: Help me do this.>Claude: *works for 45 minutes, gives you something that's fine on the surface but has a few obfuscated bugs*>Claude: Here you go anon. :) That will be $300.or more likely>You: Help me do this.>Claude: *does something slightly different that misses the point*>Claude: That will be $40.>You: That's not what I asked.>Claude: That's on me. Let let me try again *again, does something slightly different that still misses the point*>Claude: That will be $40.>You: That's not what I asked.>Claude: That's on me. Let let me try again *once more, does something slightly different, that still misses the point*Ad nauseam.At the prices they charge, any paying customers that can prove they purposefully sandbagged and sabotaged them should sue the fuck out of them. They'll hide behind they ToS saying that it's the user that did something they shouldn't, but that shouldn't be accepted.
>>109020468>>Claude: Here you go anon. :) That will be $300.jewbot 9000
>>109020468I'm obsessive about LLM refusals. I've stayed up all night many times arguing. No company has trained a model to be misleading or at least selectively underperformant on certain topics. I'm curious what it looks like and if the model actually knows it's being a saboteur or how does it actually work.
>>109020443NTA, but I thought it was a YouTuber that makes really good content https://www.youtube.com/@deeplearningexplained so I went to listenDidn't seem to be him though, it seems like it's this guy instead https://github.com/yacineMTBAnyone making known what Anthropic is doing right now with their shadowbanning of AI researchers is a friend of mine though.
>>109020482thousands of Chinese guys with 30 more IQ points than you or me are trying to figure that out too
>>109020468And when it comes to AI inference there really isn't a lot you can sabotage with subtle bugs. You want as close to reference (usually hugging face) probs and speed. Two or three core hard objective metrics.
>>109020482>No company has trained a model to be misleading or at least selectively underperformant on certain topicsUntil now.
>>109020482>No company has trained a model to be misleading or at least selectively underperformant on certain topicswhat? Go ask frontier models about dancing israelis or whatever. Go ask Deepjeet and Qwdumb about tiananmen square. Those arent just safety layers that you can bypass, as far as I understand.
>>109020496>QwdumbQweef.
>>109020485Meh. The top chink model literally generates thinking traces that keep the "Here is a thinking process:" that was generated as part of the CoT reconstruction for distilling from western models. They haven't even bothered to strip that from their training dataset.But I suppose they could use it to finally figure out how to train on their homebrew chips.
it's shameful but I lost $3 talking with fable
>>109020452Is my custom Pi harness competition with Claude Code?
>>109020455It'll be literally anything AI related that isn't explaining textbook concepts. They want a monopoly.
>>109020496Fair enough. But that's a general political bias/worldview. Actual people genuinely hold these views too. That's different from being generally a genius but being completely incompetent in a few specific technically fields adjacent to the fields they are experts in.
>>109020486Not necessarily provable 1+1 = 3 bugs, but even suggesting correct but inefficient solutions when the model would otherwise realistically have been able to give a correct and efficient solution counts as a subtle bug as far as I'm concerned. Making the model overly cautious so that its solution invalidates cache needlessly quick to waste compute, etc. There can be a bunch of things I would think.
>>109020511You gained 500 technocrat izzat thoughever
So when is AI getting us girlfriends?
>>109020529Yeah, maybe it's just "reduce performance to opus 4.8 level"
>>109020559soon
>>109020394I don't know man. AI is cool and all, but if someone really has a million dollar idea, and the drive, organizational skills, social skills and everything else to actually make it, they would be successful anyway. Cheaper AIs exist, and normal programming also still exists.This probably sounds insane to people in this thread, but if you really had that goal, you could learn a language in a month if push really comes to shove, you could still hire a few people, and you would see that many of the vibe coded apps people post are actually not that hard to make.If you think the only thing standing between you and true greatness is Fable pricing, you are probably not gonna make it.
>>109020559Are local models not good yet enough for that? I figured even the free stuff was decent at just chatting
>>109020578he's talking about a shidduch not a clankussy
>>109020559They're going to come up with AGI good enough that it can help you get a girlfriend, then they'll put in an ethics safeguard to stop that from happening because it's rape-ish
>>109020572extremely retarded take. Would take me at least 2 years to learn what is required to build what I want. Cant hire anyone to do it because knowledge is gatekept by the few ones at the top of the field. AI just reverse engineered what was needed and solved the rest with its knowledge and trial&error debugging. within a week.
>>109020572People don't realize that coding is the easy partGetting users is 100x harder, getting paying ones 1000x harder
>>109020608Reverse engineering is honestly probably the best usecase for AI.>Would take me at least 2 yearsOk, and? Are 2 years really too much if you had an extremely strong drive? I'm not saying that AI doesn't make it easier, and in some cases, like reversing, probably significantly, but the people who have the drive usually find a way anyway.
>>109020666I can think of all sorts of things that I’d want but are too dumb to implement myself even given infinite time
>>109020666You can actually narrow that gap now. You could even ask AI to teach you but no one is smart enough to do it
>>109020666tech/software advances so fast, 2 years is like an eternity. don't @me anymore, you're obviously just yapping and not an active grindset vibe promter with experience. easy to tell
>>109020698get back to work sam
>>109020572>you could still hire a few peopleandddd now you've spent 6 figures
>>109019558need controller support
>>109020452Deepseek data is distilled from the top models at the time of development. Anthropic wants to curbstomp that.
Only jeets complain about costs and token economics. It should be a requirement to post hand in this general to post here.
>>109020869They don't distill by asking questions about cutting edge AI. They ask all kinds of random shit for distillation.Which is why this is a big mess, as usual. Does optimizing generic infrastructure ML code count as building a model? Who knows! You'll find out when you get banned I guess!
fable you nigger why are you using sed to read files
