What Happens After Post-Training Data Collection?
Since launching Axis V2, human-gated DAgger correction has been a core part of our data engine. In this thread, we share a series of experiments exploring how to best process and use post-training data.
The core finding: not every human intervention helps. But if you verify which specific actions actually change the outcome — and train only on those — the improvements are real and they scale.
Results:
- 660 corrections collected → 161 (24.4%) entered training
- Each training snippet is ~0.8 seconds
- Naively imitating full human trajectories dropped success from 40.0% to 36.7%, while 161 verified snippets lifted the three-seed average to 48.3% on the same evaluation set — and reached 48.8–52.5% across three seeds in a separate paired evaluation.
Read the full blog:
Details below ⬇️
SHAKIRA AUSTIN FOR THE WIN 🔥
She finishes through contact to hit the game-tying layup with 12.8 seconds left, then sank the game-winning free throw for the @WashMystics!
NASA used a supercomputer to build a mind-bending simulation of falling into a black hole.
Once you cross the point of no return, you have 12.8 seconds left to exist.
To visualize these final seconds, NASA scientists ran the math of general relativity on a supercomputer, generating 10 terabytes of data in 5 days.
The simulation shows a camera plunging into a supermassive black hole where light warps and space flows inward. The actual descent takes 3 hours, but to a distant observer, your image would freeze at the edge forever.
A century of Einstein's equations, finally playing as video.