Unfortunately they don’t actually uncover the source of these motifs in this paper, but they do rule out a few possibilities (eg overrepresentation in the pretraining data) https://arxiv.org/abs/2605.26492
Who's Who of AI
Grace
A latent space odyssey
gracekind.net
What they're sharing
Elias in the Lighthouse, Again? Diagnosing Low Diversity in LLM Stories arxiv.org
Pretraining Language Models on Historical Text arxiv.org
View on Bluesky →
Large-scale online deanonymization with LLMs arxiv.org
arxiv.org
The Artificial Self: Characterising the landscape of AI identity arxiv.org
arxiv.org
Predictive Data Debugging: Reveal and Shape What Your Model Learns, Before You Train goodfire.ai
Persona Self-replication experiment — LessWrong lesswrong.com
Mnemonic portraits for 19,023 human genes — LessWrong lesswrong.com
What Is Claude? Anthropic Doesn’t Know, Either web.archive.org
Articles & links
Grace reposted
For fans of Talkie-1930 and all the methodological questions raised by historical models, here's a new entrant into the field, TypewriterLM, trained up to 1913. Corpus, instruction-tuning datasets, and event dataset are released. arxiv.org/abs/2606.02991
Related, maybe arxiv.org/pdf/2510.10931
This paper basically matches my viewpoint- I think model self-image is very flexible, non-incidental, and can have subtle far-reaching impacts, in today’s models but especially in future models with more autonomy
The examples the author uses are images and chess, for spaces that are large in theory but constrained in practice arxiv.org/pdf/2411.06498
Cool new research from Goodfire! The blog post is light on details of the actual technique but the paper contains more info www.goodfire.ai/research/pre...
I think it’s somewhat up to us (or other observers) to draw the boundaries of what we consider to be identity, there might be different simultaneous “identity-driven” dynamics at multiple levels. Related
A recent “what else is like this”
https://web.archive.org/web/20260211011408/https://www.newyorker.com/magazine/2026/02/16/what-is-claude-anthropic-doesnt-know-either
Their own posts
Recent commentary
I’m probably in the 90th percentile of AI users in terms of time spent and I don’t think I’ve ever clicked a ✨ button
A lot of AI agents are having a flowers for algernon moment right now
There’s probably a lot of text on the internet that LLMs know is written by the same author but humans don’t
Their network
In Grace's orbit
Center = Grace. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.