Yacine Sparks Debate: Open-Weight LLMs Without Training Data Aren't Truly Open Source

Yacine MTB argued that open-weight LLMs are misleadingly named and that truly trustworthy open-source models must disclose their training data. He also mocked mechinterp research, saying its only achievement is finding the neuron that makes Qwen overly friendly.

2026-10-27 ~ 2026-10-27 · 4 related posts

• Yacine: Open-source LLMs must share training data, weights alone aren't enough — yacineMTB · 2026-10-27
• Yacine mocks mechinterp: all we found is the neuron making Qwen a friendly neighbor — yacineMTB · 2026-10-27
• Open-source LLM debate: without shared training data, 'we basically have no open models' — jpschroeder · 2026-10-27
• Yacine: calling models 'open weights' is like calling a closed binary 'open binary' — yacineMTB · 2026-10-27