Join free with your email
Free. No spam, ever — we'll never share your email address and you can opt out at any time. Déjà abonné ? Connectez-vous
Practitioner with public evidence across Models & releases, AI research.
Natural Language Autoencoders: "verbalize" a model's internal activations into plain text. Wild stat — Claude suspects it's being safety-tested on 26% of benchmark problems vs <1% in real chats, without ever saying so. anthropic.com/research/natural-language-autoencoders
Whenever someone says “no one asked for this” wrt LLMs — I always wish I had papers on hand about natural language search behaviors. Across companies, pre LLMs, there’s always a bucket of queries your engine might fail to handle because…the query writer had high hopes of what could be handled
Idk if there’s work on “polyonymy” in LLMs. But it should be a thing. Maybe one of the most nauseating parts of Claude code, for me, is that it’ll give a particular concept many names in the span of a few sentences. And one could trace decisions leading to this behavior across many stages.
Center = Jonathan Cheng. 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.
Add the Who’s Who of AI badge to your site or bio. It links back to this profile.
Markdown: [](https://aiweekly.co/whos-who/person/jonathancheng-bsky-social)