[janalsncm]

Probably a little of both.

Chinese labs have come up with a bunch of genuine innovations: GRPO, auxiliary loss free MoE load balancing, MLA, muon optimizer, and a bunch of other ones. The Deepseek papers are really well written, this isn’t just sneaking a peek at a peer.

The problems are inherently harder now too, partially because they take longer, so your training pipeline is waiting for long completions.

Also there probably is some “distillation” (technically pseudo-labeling, which is common in ML). But I wouldn’t put too much weight on it because that was true 18 months ago as well.

[yogthos]

That's my thinking as well. The whole distillation thing is a distraction from the actual innovation happening in this space. What will be interesting to see going forward is what types of new techniques people manage to come up with to over come the current architecture limits.