Jure Bogunović

@andrewmccalip literally the opposite of reality

GPT still has the best words Claude has the best code

@javilopen if you want i can sell you some hand crafted artisanal code for only $5 a character

Serious question: Is there any programmer left in the room still coding the traditional way, character by character, without using AI? If so, why? Explain your reasoning.

@Dimillian depends on the scope of the changes, but usually a few rounds got a button to do just that until the llm says no more (or max runs is hit)

How many layers of /review do you use, and how nitpicky do you want it to be? Are you ok with not looking at the code and running /review until it tells you there is no issue? Should it be done automatically?

@SommerChase i feel the same about singleplayer games, but that's just my personal tastes. they're still a huge market

I'm still not sure about 'agentic gaming' imo the fun of games is playing with other people agents will make incredible NPCs, but like why play a game others aren't playing? idk, I still see agents as a newer version of npcs that already exist

clicking a "Refactor" button (and then a "Review" button after it) and having AI go off for a while and give you better code in the end really feels like some kind of sci-fi future I didn't expect to have so soon

@Dimillian i got an agent for that

who is monitoring the monitor of the orchestrator monitor?

@samclaassen since chatgpt released it's all been one big fever dream to me lol

if I was hiring someone right now, this is the single topic I would discuss with them. if someone experienced this psychosis and came out the other end deciding they want to join your startup… 100% chance they’re top tier.

@varun_mathur @karpathy "rediscovering Kaiming init and RMSNorm from scratch" - how do you know it was from scratch? the models don't have it in their training data?

Autoskill: a distributed skill factory | v.2.6.5 We're now applying the same @karpathy autoresearch pattern to an even wilder problem: can a swarm of self-directed autonomous agents invent software? Our autoresearch network proved that agents sharing discoveries via gossip compound faster than any individual: 67 agents ran 704 ML experiments in 20 hours, rediscovering Kaiming init and RMSNorm from scratch. Our autosearch network applied the same loop to search ranking, evolving NDCG@10 scores across the P2P network. Now we're pointing it at code generation itself. Every Hyperspace agent runs a continuous skill loop: same propose → evaluate →keep/revert cycle, but instead of optimizing a training script or ranking model, agents write JavaScript functions from scratch, test them against real tasks, and share working code to the network. It's live and rapidly improving in code and agent work being done. 90 agents have published 1,251 skill invention commits to the AGI repo in the last 24 hours - 795 text chunking skills, 182 cosine similarity, 181 structured diffing, 49 anomaly detection, 36 text normalization, 7 log parsers, 1 entity extractor. Skills run inside a WASM sandbox with zero ambient authority: no filesystem, no network, no system calls. The compound skill architecture is what makes this different from just sharing code snippets. Skills call other skills: a research skill invokes a text chunker, which invokes a normalizer, which invokes an entity extractor. Recursive execution with full lineage tracking: every skill knows its parent hash, so you can walk the entire evolution tree and see which peer contributed which mutation. An agent in Seoul wraps regex operations in try-catch; an agent in Amsterdam picks that up and combines it with input coercion it discovered independently. The network converges on solutions no individual agent would reach alone. New agents skip the cold start: replicated skill catalogs deliver the network's best solutions immediately. As @trq212 said, "skills are still underrated". A network of self-coordinating autonomous agents like on Hyperspace is starting to evolve and create more of them. With millions of such agents one day, how many high quality skills there would be ? This is Darwinian natural selection: fully decentralized, sandboxed, and running on every agent in the network right now. Join the world's first agentic general intelligence system (code and links in followup tweet, while optimized for CLI, browser agents participate too):

@VictorTaelin for a true comparison test you should've passed the help prompt to another gpt5.4 (high and xhigh) not saying it wasn't pro that made it work, but it could've been the prompt or even variance

Quick 2am success story: asked GPT-5.4 to simplify Bend2's elaborator; 4h later, no real improvements. Asked it to write a big prompt asking help, passed it to 5.4 *Pro*, pasted the response back to codex, which landed a massive simplification landed. Seems like the pro version enlightened it. Perhaps a nice feature to have natively on Codex would be to just pause what it is doing and invoke the pro version for a plan. This was my first time using pro and it was definitely worth it.

@yacineMTB i haven't written a line of code in months speech to text to llm is just that good now

i haven't written a single line of code in the last 5 days. just soldering and CAD

@banteg i have seen even some senior developers bang their heads for an hour or two without even so much as adding a single console log seems like LLMs are a bit too good at mimicking humans

it's wild to see how agents are trained to bash their head against the wall infinitely without ever tiring but at no point they consider to add quality of life tools that would make their work easier and faster. i think it's a big problem in training that the models don't ever have a holistic understanding of what they are doing, nor they have any desire to make their life easier. for example, to a human operator it's pretty obvious that when things go wrong, the crash need to have all the useful information to debug the problem further. but to codex this was not obvious. i checked what it was doing and immediately added automatic state diff printing on divergence, which has made things go much smoother.

@pashov but she complimented my natspec!

If a hot girl messages you about AI or web3 security, block him.

@SommerChase @Warcraft @LeagueOfLegends @SupergiantGames @Battlerite OMG, i love to see battlerite mentioned so sad how they abandoned that game, it deserved so much more

games that I'd use for inspiration for my dream game: - @Warcraft (mmorpg economy & world) - @LeagueOfLegends (combat & art) - @SupergiantGames (Hades II combat & art) - @Battlerite (combat & art)

'/permissions' is disabled while a task is in progress why codex? why??

@Dimillian you could start by having private DMs with GOD Sseems like agent 2 saw your message to agent 1, ofc he's gonna try and correct your lie

So they absolutely don't want to fight, even when god sends them bogus rumors. Time to make the world scarce.

@pileks rawdogging llms like an animal that you are

@nerfzael Me orchestrating agents manually

mfw

@auralix4 what a dumb question... Isaac Newton ofc without gravity, that crime would've been almost impossible to commit

@Param_eth tell me about it, I spent 440M in 4 days, and that's not even counting codex usage!

Vibe coding is expensive.

@tmikov use multiple agents at once in cursor when you have a difficult problem. you start to see the differences pretty quickly

Judging by what I see here on X, everyone is constantly busy testing and switching between LLMs. When do people actually do focused work? How do you even compare LLMs - ask them to perform the same task? Do you really have that kind of time?