LittlePly chess agent adds auxiliary objective predicting a move 2-9 plies ahead
andreisavu · x · 2026-11-08
A detail from the author's LittlePly chess agent training setup: a second auxiliary objective predicts one sampled move at an offset of 2–9 plies ahead. Each target contains a single move, with no intervening sequence or principal variation. Combined with the earlier note, training also includes a board objective predicting the contents of all 64 squares, but occupancy omits castling rights, en passant and repetition history — low reconstruction loss alone doesn't prove reliable state tracking.
More from coding & agent
- This PM replaced the PRD with an interactive Claude artifact — PawelHuryn · 2026-11-08
- Spacedrop lets you drag-and-drop code to deploy in seconds, no GitHub push needed — saheedniyi_02 · 2026-11-08
- Beam's technical report details its sandwich architecture — brandondamos · 2026-11-08
- LangChain's async subagents: orchestrator stays responsive while background workers run long tasks — hwchase17 · 2026-11-08
- 1122 public GitHub files leak secrets via Claude Code; auto-gitignore claim doesn't hold — PawelHuryn · 2026-11-08
- Saccadia Remote ships a read-only MCP gateway to inspect desktops without giving agents control — Plasmicon · 2026-11-08