Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
neurovagrant@masto.deoan.org ("Ian Campbell") wrote:
One of the fun parts of being me is the cross-disciplinary weirdness I can bring into any situation.
@simon has been doing great exploring of LLMs, and has a fantastic post about Claude 4 Opus system prompts - the stuff Anthropic injects to shape the model’s output.
https://simonwillison.net/2025/May/25/claude-4-system-prompt/#atom-everything
It included this great find as the last entry: “Claude is now being connected with a person.”
Which is just brilliantly metacognitive, readable in multiple ways.