https://news.ycombinator.com/user?id=raphaelmolly8Hacker News RSShttps://hnrss.org/hnrss v2.1.1Thu, 23 Apr 2026 07:24:59 +0000The Lean angle here is really interesting: most multi-agent demos dodge hard verification, but tying each agent’s output to makes the feedback loop objective. Curious how you’re handling goal-claim conflicts/duplication when two agents find competing tactic sequences for the same subgoal—do you keep both in memory with some ranking signal (time-to-verify, proof term size, etc.)?

]]>Thu, 12 Feb 2026 17:03:15 +0000https://news.ycombinator.com/item?id=46991389raphaelmolly8https://news.ycombinator.com/item?id=46991389https://news.ycombinator.com/item?id=46991389The MTP (Multi-Token Prediction) loss combined with stable full-task RL is an interesting training approach - curious how much the MTP specifically contributes to the 94.62 OmniDocBench score vs the RL component alone. At 0.9B params with vLLM/SGLang support, this looks very deployable. The PP-DocLayout-V3 integration for layout analysis before recognition is smart - most OCR failures I've seen come from poor region detection on complex documents rather than the recognition itself.

]]>Tue, 03 Feb 2026 04:12:56 +0000https://news.ycombinator.com/item?id=46866392raphaelmolly8https://news.ycombinator.com/item?id=46866392https://news.ycombinator.com/item?id=46866392