'10% challenge' closes: match a model with 10% of its size and no weight training, $1k prize
tokenbender · x · 2026-11-13
tokenbender's '10% challenge' has closed per its original deadline; the author will review recent submissions and pick qualifying winners. Rules: start from a trained model (>0.5B params), keep at most 10% of its fp16 size, no weight training, match the original on a commercial-class task, and pass a held-out exam locked away until the end. $1,000 for the best complete working proof. Repo: DeusExperiments/ten-percent-challenge (16 stars); results pending.
More from Research
- PhD thesis explores intra-turn interaction: AI interfaces beyond turn-based dialogue — OfirPress · 2026-11-13
- CMU trains memory writing with RL: Qwen3.5-4B rivals GPT-5.4 on continual learning — aviral_kumar2 · 2026-11-13
- NTU paper: decision info lives in middle layers, last six layers removable loss-free — Minh-Anh Nguyen · 2026-11-13
- Inkling 2 Small's width-4 sconv forces a tiny rolling 3-token conv cache per layer at inference — andrew_n_carr · 2026-11-13
- Open-source legal conflict detector runs on encoder models only, 133-clause contracts in under 20s — ladyofcode · 2026-11-13
- MIT's AI copilot for structural design boosts buckling load 39% with 15 seconds extra — MIT News AI · 2026-11-13