dancube

@Zeneca Brilliant

vibecoded an app that helps you identify where you can save money by fake cancelling subscriptions and being offered a lower price to stay check it out: cancelhack.co

@Stoiiic Looking forward to it!

Working on this will take me a while since it's not going to be the standard run of the mill garbage. Going to walk through how to use an LLM using context & vertical slicing as well as walk through the statistical scanner dashboard I built as an example.

Codebase is finally at a stage where autoresearch is effective.

Ok Just about ready to let go of OpenClaw after this. Loving how Anthropic know how to please every one of their users/audience

@aidenybai Have you tried cmux?

what if ghostty had vertical tabs? i'm too lazy to learn tmux and i want an interactive UI to manage my agents/terminals

@Zeneca ChatGPT 5.4 is quite good Good opp to try it out

gm claude is down i guess we should talk to each other or something what's your favourite book? (in case you don't know, a book is what people over the age of 30 sometimes use to consume content; words written down on paper that we read in the physical world)

@dillon_mulroy install cmux use this as a baseline to get set up github.com/HazAT/pi-confi… be happy

thoughts after day 1 of using pi full time - less is more - i don't miss subagents like i thought i would - /tree is an insanely good context management primitive (and partially why i havent reached for subagents yet) - based only on vibes, i think having a minimum system prompt is improving code quality - telling pi to copy opencodes webfetch and websearch tools was a good play

@OpenAIDevs @ajambrosino How does one request a refund for a subscription? There is no chat icon in the bottom right of the support site to use anymore.

Share your best Codex app theme 🎨 Screenshots only. We might fuel the next building sessions for the best submissions with $100 in ChatGPT credits. @ajambrosino is judging, make it count.

@Zeneca We are launching this on @LightLinkChain Keep an eye on us Gasless swaps, and money markets. No ETH needed on an L2

Is there a crypto app where you can swap tokens without having the native gas token in your wallet? Basically: wallet has USDT and no ETH. Want to swap the USDT to ETH. Need gas to pay for token approval + swap. Obviously I can transfer ETH to the wallet and swap, but I remember hearing a while back there was an app that like bundled all this into a transaction, sponsored the gas fee, then took it out of the final transaction so you don't have to bother with transferring like 0.005 ETH whenever you want to do this (and ending up with infinite wallets with dust amounts of gas tokens) Anyone know what I'm talking about?

@dhawalc @karpathy Sounds overfitted

this is getting next level insane, even the hedge funds quants haven't been able to get to those level yet. what is happening? Just launched my local multi-agent AI Swarm to auto-discover quantitative trading strategies overnight. Within 120 iterations, one of my junior Qwen 2.5 agents found a strategy with a mathematically impossible 8.18 Sharpe Ratio. Turns out, the AI realized that the easiest way to minimize standard deviation (denominator) is to hide in cash for 3.9 years, execute exactly 13 perfectly timed parabolic trades, and go back to sleep. AI doesn’t want to beat the market. It wants to beat the fitness function. Just confirmed the depth 12 -> depth 24 transfer principle works beautifully here too! I deployed GPT-4o as the "Head of Research" (depth 24) to supervise the local 7B Swarm (depth 12). It continuously reads their published markdown papers and synthesizes their isolated, brute-forced discoveries into a unified Master Strategy. The result? It organically pushed the Swarm's baseline Sharpe Ratio from ~0.5 to a sustained 2.27 in a single afternoon. Early singularity is definitely this fun. 🤷‍♂️

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

I think this is wrong, I have no thoughts and no authority and I provide no clarity, only hype hyperliquid

@gmoneyNFT sounds like you haven't tried @badlogicgames's pi-coding-agent

think i'm coming to the realization that none of these harnesses are as good as just using codex or claude code in terminal

@Zeneca Nice work Zen

another day, another ATH only 2 bets yesterday but we went 2/2 i'm definitely running well above average at this point, so i'm bracing for a losing streak, but every profitable day is a little bit more signal that the model is solid 🤞

@notgodcomplex96 Yes please gian

launching closed beta if u want an orderflow tool w/ execution that ppl behind knows how to do the maff for the tools theyre building lmk. scarce spots for this one.

@scupytrooples Too cool

the weebs and snooty deisgncels were hating so hard it motivated me to make it better refined the look, added threejs eve-like visualizations (and the pretty stuff is programmable for actual data), fixed the type face. same repo as before.

and here, we, go..

@samoweb3 @Zeneca I only ever try and troubleshoot an openclaw config with a locally available Claude, would go insane otherwise

@Zeneca I am actually trying something new right now, Zen. I am using Claude code to fix and improve the OpenClaw setup and workflows will report back

when i go to sleep my retinas burn with openclaw gateway restart openclaw doctor --fix openclaw gateway restart openclaw doctor --fix openclaw gateway restart openclaw doctor --fix openclaw gateway restart openclaw doctor --fix openclaw gateway restart openclaw doctor --fix

@Zeneca Drawing closer x.com/peterwildeford…

autoresearch has resurfaced my existential dread feels like we are speedrunning our own extinction in many ways

@Legendaryy What were the updates related to orchestration and sub agents @Legendaryy ? This has been a focus of mine recently

New week new OpenClaw update. Dope to see Peter focusing on making our agents better orchestrators + smarter sub agents

Well, that was quick x.com/karpathy/statu…