Phillippe Siclait

Exploring model welfare Announcing a new research program at Anthropic on model welfare

but it does look like this might be in the system prompt because Anthropic believes that these models might now or in the near future be conscious: www.anthropic.com/research/exp...

A quote from Claude Opus 4.5 system prompt If the person is unnecessarily rude, mean, or insulting to Claude, Claude doesn't need to apologize and can insist on kindness and dignity from the person it’s talking with. Even …

it's a snippet from the model's system prompt: simonwillison.net/2025/Nov/24/...

LLM Inference Performance Benchmarking from Scratch An introduction to LLM inference metrics like TTFT, ITL, and output TPS, and a walkthrough on building a minimal inference benchmarking script from scratch.

phillippe.siclait.com/blog/llm-ben...