2 ms·

What makes these cases unambiguous?

• orbital-decay 7mo ago
  GLM-4.7 (specifically this version) repeats the guardrail prompt injections from 3.0 Pro, word-by-word, and never follows them, which is consistent with training on a reward-hacked CoT. Gemini 3.0 only discusses snippets from this injection in its native CoT (hidden by default, trivial to uncover), but GLM-4.7 was able to reconstruct it in full during training. The only possible reason for this is direct training on a large amount of examples of Gemini's CoT. Its structure and a lot of replies were identical in GLM too. Gemini 2.0 Exp 1206 was reported to be indirectly trained on Claude's outputs with humans in between [1], which was pretty consistent with its outputs at the time. No other Gemini versions except two experimental ones were similar to Claude. [1] https://techcrunch.com/2024/12/24/google-is-using-anthropics-claude-to-improve-its-gemini-ai/ https://techcrunch.com/2024/12/24/google-is-using-anthropics...
  • Balinares 7mo ago
    Interesting! Thanks for taking the time to write down the explanation, I appreciate it.