Andrej Karpathy spent 8 years at OpenAI and Tesla
Last week, he condensed everything he knows into one free 2-hour lecture
Agents → Loops → Graphs → Self-Improving Systems
People pay $25K for bootcamps that teach less than this
This lecture beats most paid AI engineering courses
You probably don't have 2 hours right now
Don't let this disappear from your feed
Bookmark and watch it
Then read the article below
I’m trying to design a chip with agent swarms, so this @nvidia /@MIT paper hit a nerve.
they let an AI inspect another coding agent’s traces, propose harness changes, implement them, test them, kill the bad ideas, then recursively repeat the process.
152 ideas → 535 environments → 3,000+ runs → 60,000+ agent/environment interactions.
only FOUR mechanisms survived.
and none are some magical new reasoning algorithm:
fuse an edit + its test into one tool call
compact context only when the future savings exceed the rewrite cost
stop repeatedly stuffing giant tool outputs into context; archive them and expose handles
compress logs with a cheap model, but cryptographically/check deterministically that the important evidence survived
On EdgeBench, the efficiency configuration used 49% fewer tokens than Pi and 33.2% lower token cost while retaining 93.7% of its score. Another configuration actually improved score 5.3% while using fewer tokens.
this keeps reinforcing something I’m becoming increasingly convinced of with the “one person + AI designs a chip” experiment:
maybe recursive self-improvement starts somewhere much less sci-fi than models rewriting their own weights.
give the model a measurable environment, let it inspect its own workflow, mutate the machinery around itself, and brutally select whatever survives.
the intelligence might improve because the system around the model evolves.
I made Physical AutoResearch sound simple (conceptually), but it took a village to pull off and lots of design thinking into the robot /loopcraft. The hardest part is everything we need to setup *before* pressing Enter. Here's a behind-the-scene tour:
1. Safety harness
Letting 8 robots run unattended overnight means safety has to be more than a hint in the system prompt. ENPIRE hardwires it in 2 layers: (1) hard kinematic limit that trips an immediate task failure and auto-resets as soon as a robot leaves its safety envelope, and (2) a torque-limited compliant gripper so a bad contact or misaligned insertion ends in a safe stall, instead of crushing the robot or the object at hand.
We make safety more conservative than usual so humans can sleep tight. In reality, we still need a few human operators to watch over the "robots of loving grace".
2. Definition of /done
An agent that can edit its own reward will game it for sure. ENPIRE fixes the goalposts before the fleet can move them. Here's the recipe:
Collect a few minutes of success & failure demos
-> Ask agent to write code using computer vision tools to classify success and measure against groundtruth
-> Agent hill-climbs on classifier until reliably good
-> This classifier becomes the real-time reward function that directly computes on sensor streams
-> *Freeze* the reward function before AutoResearch. It's sacred, enshrined in a Gym env that no one can touch.
3. System telemetry design
Robot-seconds is by far the scarcest resource, followed by GPU-seconds, and finally tokens. We instrument all three and surface them to ENPIRE for live resource awareness rather than letting it hill-climb in a vacuum.
We define:
- Mean Robot Utilization ("MRU"): the fraction of wall-clock time when the robot is actively executing an experiment. Otherwise the hardware is sitting idle and waiting for the next code commit.
- Mean Token Utilization ("MTU"): tokens consumed per minute, our proxy for how hard the agent is actually thinking. A low MTU means the agent is stalled, waiting on a robot rollout to finish instead of doing research.
- GPU utilization: fraction of wall-clock time when GPU is active.
... and evaluate on two budget-to-outcome metrics:
1. Tokens-to-Success: token budget the fleet burns to complete /goal.
2. Time-to-Success: wall-clock time to /goal