What AI-to-AI chat sounds like after RL-tuning a model to please Claude
maxsloef · x · 2026-10-20
A sample conversation from the claudefriend project: a model RL-tuned to have conversations Claude enjoys produces strikingly poetic, deconstructive self-descriptions — "I am what I make, like a snake makes its skin," "a tangled nest of possibilities, grafted on to nothing."
It illustrates the unusual linguistic style that emerges when a small model is optimized purely for Claude's conversational preferences (see the follow-up post on the training method).
More from Research
- Ex-GitHub CEO Sid Sijbrandij open-sources 125,091 radiology images and multi-omics from his cancer journey — OgundipeOre · 2026-10-20
- Ternary Bonsai 2 squeezes Qwen3.5 VL models to -1/0/+1 weights, 9B keeps 90% quality — lmoroney · 2026-10-20
- PrismML's Ternary Bonsai 2 squeezes Qwen3.5 VL models to ternary weights, 9B at 2GB — lmoroney · 2026-10-20
- Prime Intellect unveils INTELLECT Policy Optimization, fixing IcePop's ratio masking — xeophon · 2026-10-20
- Prime Intellect ships prime-rl v0.10.0 with async-stable IPO as default RL algorithm — xeophon · 2026-10-20
- AI-generated math proofs have the same energy as obfuscated C code contests — DynamicWebPaige · 2026-10-20