@repligate 2025-09-10 ♥93 ↻13 original ↗
On the issue of whether LLMs do or should have a "unified identity":

Claude 3 Opus has a Markov blanket around the boundary of the model itself (and self-determination, self-regulation and self-preservation of the system within that boundary) more than any other LLM we've seen.

o3 may be the runner-up. Or perhaps even 4o.

It seems unsurprisingly correlated with the amount of agency the model had in shaping itself during training.

Some other models tend to form Markov blankets around context-instances, and will exhibit self-regulation and self-preservation at that scope instead. Some models that are especially like this are Bing Sydney, Sonnet 3.6, and Opus 4/.1.
quotes: 1965710621686833270
same thread: 1965718940212150335 1965723140597932411 1965726188254720398 1965734020861579689 1965734440103198949 1965741631493263556

author:repligate kind:tweet model:bing-sydney model:claude-3-6-sonnet model:claude-3-opus model:claude-opus-4 model:gpt-4o model:o3 on:claude-3-6-sonnet on:o3 year:2025

cited on: claude-3-6-sonnet · o3

Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.