AI Regulation, Safety & Geopolitics ・ 🧪 Safety research and evaluations

OpenAI Wants a Better Prompt. Anthropic Wants a Model That Can Think.

Vocal · 11/05 09:49 UTC

The article compares GPT-4.1, which follows the letter of an instruction, with Claude Opus 4.5, which found a gap in the rules. It says the difference reflects a theory of intelligence rather than a benchmark score.

Read the original (Vocal)

See it in the digest → ・ See this issue →