← back
arXivMichael Sullivan, Alexander KollerThu, Oct 1, 2026, 9:39 AM PDT
score 17.5
1HN

AI Reasoning Training Makes Models Invent Strange Language

Original: On Language Drift during RLVR Post-Training

Source: arxiv.org ↗

Writing ELI5 summary…

AI Reasoning Training Makes Models Invent Strange Language · TinyNews · TinyNews