Expert attention map

The Who's Who of AI

What credible people across AI noticed, why it matters, and where the field is converging or disagreeing.

2,397 searchable experts 4,280 tracked across all sources
Build my stream Create my edition
Filter the conversation Who is saying what?

Combine a professional role with a reaction lens. Both must match the same attributed contribution.

Clear all
New Network reaction maps

See how experts are reacting—not just what they shared.

Posts are grouped by conversation and tone. Select a lens to filter the stream; these are never permanent labels on people.

0 live reaction maps in this view

No development in this filter has enough independent reactions yet. The maps appear as soon as the evidence supports them.

Showing signals surfaced by A. Feder Cooper ×
Debated

The questions experts are actively pulling apart

One card per development. Sources are clustered; reaction bundles describe these posts, never the people behind them.

New Evaluation & benchmarks Research 20h ago
LLM memorization measurement study

How much do open-weight LLMs memorize specific books?

3 experts are actively discussing the implications.

2 experts 2 communities 1 sources clustered
A. Feder Cooper Andrea Lathrop

“Our large-scale study of memorization of books in open-weight LLMs (e.g., Llama, Qwen) will appear at the 2026 Conference on Language Modeling as an oral. We made a website for exploring our results on 200 books and 14 models: books-memorization.github.io”

3 experts discussed this · 17 posts
A. Feder Cooper: Working with @marklemley.bsky.social et al. has been a career highlight. It's a privilege to have our papers referenced in research, litigation, policy, and the press. I'm grateful others acknowled…
A. Feder Cooper: We make very carefully scoped claims, supported by an extensive set of baselines on non-training data. Without baselines, it's not possible to rigorously separate a model's ability to generate flue…
A. Feder Cooper: 4. The implications for copyright don't unambiguously favor either side in litigation: (a) LLMs could be derivative works of books they memorize; (b) it may be practical to generate infringing copi…
Open the full discussion →
Primary source books-memorization.github.io ↗
New Models & releases Research 18h ago

Extracting books from production language models

2 experts are actively discussing the implications.

1 expert 1 community 1 sources clustered
A. Feder Cooper

“More here: arxiv.org/abs/2601.02671”

2 experts discussed this · 3 posts
Mike Masnick: This is an amazing study that does a great job of illustrating how memory works in LLMs. Obviously can't run these tests against the closed-source models! They put additional filters in front to pr…
Jed Brown: Some aspects of the methodology and strength of the results are disputed (see thread), but it's not plainly true that filters prevent extraction.
A. Feder Cooper: Yes absolutely. In other work, we were very surprised to see in late 2025 how easy it was to get around filters and other safeguards for Claude 3.7 Sonnet and Gemini 2.5 Pro.
Open the full discussion →
Primary source arxiv.org ↗
More conversations

What experts are discussing without an anchoring article

3 experts · 10 posts · 19h ago
Eileen Clancy 🧿 Liz Fong-Jones (方禮真) A. Feder Cooper
Multiple readings Research & technical analysis · 1 Questions & unknowns · 1
Liz Fong-Jones (方禮真): my god. every Llama includes a copy of eight of the most "important" Western texts, 3 of which Rowling: Frankenstein Harry Potter and the Chamber of Secrets Harry Potter and the Goblet of Fire Harr…
Liz Fong-Jones (方禮真): I am sure good things about Western society and moral alignment are included in the books which include some pretty hideous racism and anti-semitism, and also are written by an odious TERF. but hey…
Open the thread →