Amit B

More than a year of building AI agents made us realise that useful and dependable AI Agents need many core capabilities like - tool calling and context engineering - planning and goal setting - reasoning - evals and feedback loops Each of these are getting better as model capabilities and tooling improve. vMCP (virtual MCP) is our first fully open-source component in the Agent Stack to improve tool calling and context engineering in your agents. Read more in the link below...

How @PMCPune treats pedestrians. This is under the Baner metro station (Ram Indu park) and has been this way since 2-3 months. I've seen ladies risk limb and life to cross the craters and cables. Either that or walk into incoming angry traffic

De Gaulle of this Charles!

@championswimmer Don't forget the ! alias ultrathink='rm -rf' Influencers will spread this new feature

So I can now distribute skills with `! rm -rf *` in skills marketplaces and nuke people's projects?

It's all ralph wiggums all the way down. TTT: test-time-turtles?

The era of personalised trading terminals is here 🚀

Damn, wasn't even a real parrot astrologer

Its raige baiting turtles all the way down

Harness engineering

Noice

Just as we shifted to all lowercase

@atmoio Pay and pray

Uppercasing letters and softening “MUST run” to “run” is proof that even Karpathy is not immune to doing rain dances to try to appease the unpredictable LLM gods. Talking to a computer used to be deterministic. Now you pray.

@mattjay Does no one know that these are just clever ad campaigns? 1M targetted views

TL;DR - guy thinks he hacked Perplexity. Perplexity responds with "lol, no our billing is just async. check your inbox." Guy finds his bill: *shocked pikachu*

@sundarpichai vMCP ;) github.com/1xn-labs/1xn-v…

To MCP or not to MCP, that's the question. Lmk in comments

PSA reminder that the "thinking" you see are not the actual reasoning traces

@karpathy Q: how is autoresearch different from GEPA? Besides the obvious no evolution part - which would presumably be very expensive to do in this case @lateinteraction

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.

RL simply increases the probability of generating the correct final token sequence aka the "answer". Any thinking generated during the process is incidental and not causal. Infact the direction of causality might actually be reverse.

@_svs_ And importantly still win elections.

Skill issue. Country that can demonetise and lockdown with four hours notice can easily do this.

Clever pricing! Subsidize the cause and charge a premium for the cure. This is premium subscription for not seeing 3 min ads on YT.