VQV Signal
According to a post on X by @make_robot_go, roboticist Ros Shinkle says that training the model that makes a humanoid walk took about a week on 128 GPUs before October 20. He says it now takes about 3 minutes on one GPU, due to a series of open-source releases every few days. He also notes that the comparisons are not exactly apples to apples.
According to a post on X by @make_robot_go, roboticist Ros Shinkle says that training the model that makes a humanoid walk took about a week on 128 GPUs before October 20. He says it now takes about 3 minutes on one GPU, due to a series of open-source releases every few days. He also notes that the comparisons are not exactly apples to apples.
AI-assisted summary based on the listed source.
Ros Shinkle says training the model that makes a humanoid walk took about a week on 128 GPUs before October 20 and now takes about 3 minutes on one GPU, thanks to a string of open-source releases every few days, while noting the comparisons aren’t exactly apples to apples. Much cheaper training lets small teams...
If the speedup holds, small teams could iterate on humanoid movement without access to large GPU clusters. Builders and researchers would be able to test controller changes far more often. Because the speaker flags that the comparison is imperfect, the next step is to look for like-for-like benchmarks from the open-source releases.
VQV organizes public signals from inspectable sources. It does not independently verify the underlying report.
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
VQV surfaced this signal because it is recent, relevant to Robotics, connected to Scobleizer News Pulse (X).
No login, cookies, social SDKs, or automatic posting.
