A reverse-engineered model mimics Jev, TypeSafe's commercial system for selecting from multiple text options in a single pass. The repository includes implementations for Doom and chess games, with training and evaluation tools using attention-based scoring across option-context pairs.
MetaRSI-v1 is a recursive self-improvement system that extends beyond formal benchmarks to operate across diverse scientific and engineering domains by composing three typed operators—Data-RSI, Harness-RSI, and Model-RSI—over a unified paradigm, enabling both parameter-level and interface-level improvements without external supervision.
Anthropic's commercial products (Claude for Work, API, Gov) do not use customer inputs for model training by default, but explicit feedback via thumbs up/down buttons may be retained for up to 5 years and used for research and model improvement. Organization owners can disable feedback submission through privacy settings.