Yi

Appreciate @MessariCrypto recognizes @codatta_io positioning with deAI data stacks. fun fact: Codatta actually evolves from OSS Microscope Protocol, which we started with @coinbase @MessariCrypto and @GoPlusSecurity docs.codatta.io/en/community/h…

Talent in the AI era: the real differentiators are ideas, taste, and relationships (underlying traits—agency, self-direction, curiosity—still hold).

Using multiple CC instances in parallel—switching in the native terminal is too cumbersome. •not for me (after working with it for a few of serious hours): recently tried Conductor @conductor_build . It seems to add an extra layer and changes the vanilla CC behavior. But, UI is sleek. •Recommended: Tried cmux @manaflowai . Multi-panel setup, keeps CC behavior unchanged, includes task completion notifications.. Still missing one critical feature: panels should summarize the ongoing task based on title keywords. •(tmux is decades old—steep learning curve, and the official site feels very outdated.)

thanks

@codatta_io Worth a deep dive on the catalog of filled and unfilled bounties.

Remember RentAHuman.ai? Well, besides having their vibe coded app leak database credentials, it turns out that the DB shows only 13 of the 11,023 bounties were ever fulfilled. So, yes, the "AIs are employing humans" meme is pure hypemaxxing right now.

agree

🎾Introducing LATENT: Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data Dynamic movements, agile whole-body coordination, and rapid reactions. A step toward athletic humanoid sports skills. Project: zzk273.github.io/LATENT/ Code: github.com/GalaxyGeneralR…

Robotics lacks infrastructure, not intelligence. Everyone wants to build bigger robot models, but most Physical AI papers complain about the same things: data collection is slow, sim-to-real is fragile, teleop is painful, evaluation is messy, long-horizon control still breaks.

I made an interactive blog post about how JPEG image compression works: sophielwang.com/blog/jpeg

BREAKING: Proof—a new product from @every It’s a live collaborative document editor where humans and AI agents work together in the same doc. It's fast, free, and open source—available now at proofeditor.ai. It’s built from the ground up for the kinds of documents agents are increasingly writing: bug reports, PRDs, implementation plans, research briefs, copy audits, strategy docs, memos, and proposals. Why Proof? When everyone on your team is working with agents, there's suddenly a ton of AI-generated text flying around—planning docs, strategy memos, session recaps. But the current process for collaborating and iterating on agent-generated writing is…weirdly primitive. It mostly takes place in Markdown files on your laptop, which makes it reminiscent of document editing in 1999. Proof lets you leave .md files behind. What makes Proof different? - Proof is agent-native: Anything you can do in Proof, your agent can do just as easily. - Proof tracks provenance: A colored rail on the left side of every document tracks who wrote what. Green means human, Purple means AI. - Proof is login-free and open source: This is because we want Proof to be your agent's favorite document editor. Check it out now, for free—no login required: proofeditor.ai

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.

Replace github.com with mintlify.com on any repo. Get production-quality docs auto-generated from the source code.

I just reread "How to Do Great Work." It's so long! But it also has less fat than most things I've written, which is a weird combination, because usually writing that's long on the macro scale is long on the micro scale too. paulgraham.com/greatwork.html

Incredible post for anyone into robot learning. We need many more of these blog post-style formats that are honest documentations of what using these big models is like, what problems you run into and what works/what doesn't. Kudos to Brandon + team.

🚨 BREAKING: Alibaba just handed the AI agent community a production-grade sandbox for free. OpenSandbox is a full-stack platform for running untrusted agent code safely: → Unified APIs across multi-language SDKs → Docker and Kubernetes runtimes purpose-built for agents → Browser automation, VS Code desktop, and network isolation included → Designed for coding agents, GUI agents, evaluation, and beyond Not a side project. Built by Alibaba. Open source. 1.5k stars (+1,100 this week). The secure agent infra you didn't have to build yourself.

Sakana AI's new paper is fascinating "Doc-to-LoRA: Learning to Instantly Internalize Contexts" The current problem is that feeding a long PDF usually means re-feeding or caching the whole document every time you ask something, which is expensive, slow, and would hit the context limit. Their paper proposes that you only read the document once, then compile it into a tiny LoRA adapter in a single forward pass, so the model can answer later without the original text present. This cuts KV-cache/memory and latency while still recalling key facts even far beyond the model’s context window.

your openclaw needs to pay for things on its own but sharing your credit card details is risky. give your agent its own debit card for free 💳 I created AgentCard to allow agents to create virtual VISA cards: a human sets the amount and the agent can then use the card through MCP. try it for free here: 👉 agentcard.sh 👈