How a European AI lab post-training Chinese open models can earn credibility
alejandroll10 · x · 2026-10-28
- Responding to 'sovereign European AI labs' post-training open-weight Chinese models, the thread lays out a strategy: it's a fine start, but you don't get a seat at the adult table yet.
- Advice: release the weights immediately (nobody pays API prices anyway; earn from hosting inference); keep benchmarks spotless and avoid saturated suites where 100% scores signal something wrong; state your actual contribution explicitly, e.g. 'we took checkpoint X of Kimi and did CPT/SFT/RL to improve Y.'
- Core argument: transparency is the only path to credibility for post-training labs.
More from Companies & People
- Pancake's SF party aims for a Guinness World Record with guests from OpenAI, SpaceX, Nvidia and more — brandon_galang · 2026-10-28
- Critic coins 'cringelabs' to mock giotto_ai's dubious 'sovereign AI' pitch — tensorqt · 2026-10-28
- KHIPU 2027 opens call for workshops, proposals due November 6, 2026 — NandoDF · 2026-10-28
- IEEE Spectrum interviews Google DeepMind's resident philosopher on AI autonomy, safety, human-AI relations — dioscuri · 2026-10-28
- 13 Weeks, 1,701 Posts: OpenAI Sentiment Stuck at 4.9/10 Until Terra Launch — kraios · 2026-10-28
- UvA Amsterdam opens funded AI4Science PhD/Postdoc positions via ELLIS, deadline Oct 31 — windx0303 · 2026-10-28