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
Filter the conversation Who is saying what?

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

Clear all
Active evidence filter

Showing developments with attributable Research & technical analysis reactions.

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.

Showing signals surfaced by Andrea Lathrop ×

The developments commanding sustained expert attention

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

Established AI research Analysis 9d ago
⚡ 3 h early
neural networks already clean essay

Neural Networks Are Already Clean

13 experts across 6 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 2 attributable expert contributions · Tim Kellogg, Shahab Bakhtiari
“Goodfire: interpretability may be solved soon A new discovery shows that LLMs are directly interpretable without first “untying” them already-clean.goodfire.com” evidence ↗
13 experts 6 communities 1 sources clustered
How the network is reacting 2 experts are emphasizing research & technical analysis.
Shared emphasis

Research & technical analysis

2 experts

Evidence, methods and technical implications.

“Quite a claim in this new work from Goodfire. Among many points, they argue that some of the main examples of superposition and polysemanticity in large models were misunderstood. This definitely deserves some deep, rigorous peer review. already-clean.goodf…”

4 experts discussed this · 14 posts
Ryan Moulton: Haven't read this yet, but I'm posting it because it seems important and I want to. already-clean.goodfire.com
Ryan Moulton: Claim, as they say.
Ryan Moulton: @norvid-studies.bsky.social some sparkling fresh juicy claims for you.
Open the full discussion →
Developing AI research Research 2d ago
⚡ 42 h early
Pew Research AI survey takers study

Can AI Stand In for Human Survey-Takers? Not Really

7 experts across 3 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · J. Nathan Matias
“In case you've been wondering about the viability of AI simulated surveys, Pew has offered in-depth reporting on their experiment with simulated survey participants. By Athena Chapekis, Arnold Lau, Samuel Bestvater, Sono Shah, Andrew Mercer, Aaron Smith www…” evidence ↗
7 experts 3 communities 1 sources clustered
How the network is reacting Experts are approaching this through 2 distinct lenses.
Multiple readings

Policy & governance

1 expert

Rules, institutions and accountability.

“We just had a discussion on this in our editorial meeting of Internet Policy Review. It would be great to hear from other academic journal editors what are their policies around researchers using synthetic data. -> www.pewresearch.org/data-labs/20...”

Research & technical analysis

1 expert

Evidence, methods and technical implications.

“In case you've been wondering about the viability of AI simulated surveys, Pew has offered in-depth reporting on their experiment with simulated survey participants. By Athena Chapekis, Arnold Lau, Samuel Bestvater, Sono Shah, Andrew Mercer, Aaron Smith www…”

Established Evaluation & benchmarks Research 7d ago
LLMs human attributes AoE2 paper

If LLMs Have Human-Like Attributes, Then So Does Age of Empires II

2 directory members surfaced this signal.

Why this matches Research & technical analysis reaction 2 attributable expert contributions · Mark Riedl, Andrea Lathrop
“Everyone should read this, just to see how they implemented a perceptron (a very basic neural network) in Age of Empires II arxiv.org/abs/2605.31514 The artificial neurons pass activation energy in the form of goats.” evidence ↗
2 experts 2 communities 1 sources clustered
How the network is reacting 2 experts are emphasizing research & technical analysis.
Shared emphasis

Research & technical analysis

2 experts

Evidence, methods and technical implications.

“Everyone should read this, just to see how they implemented a perceptron (a very basic neural network) in Age of Empires II arxiv.org/abs/2605.31514 The artificial neurons pass activation energy in the form of goats.”

Established Evaluation & benchmarks Research 9d ago
⚡ 3780 h early
LLM introspection reality check paper

Can LLMs Introspect? A Reality Check

2 directory members surfaced this signal.

Why this matches Research & technical analysis reaction 2 attributable expert contributions · Shauli Ravfogel, Andrea Lathrop
“12/ We just think the evidence so far doesn't quite support a strong interpretation of the introspection findings in previous work. And, as always, extraordinary claims require extraordinary evidence. Paper: arxiv.org/abs/2605.26242” evidence ↗
2 experts 2 communities 1 sources clustered
How the network is reacting 2 experts are emphasizing research & technical analysis.
Shared emphasis

Research & technical analysis

2 experts

Evidence, methods and technical implications.

“12/ We just think the evidence so far doesn't quite support a strong interpretation of the introspection findings in previous work. And, as always, extraordinary claims require extraordinary evidence. Paper: arxiv.org/abs/2605.26242”

New AI research Research 12h ago
Platonic representation JEPA proof paper

Proving and Improving the PlatonicRepresentation Hypothesis with JEPAs

1 directory member surfaced this signal.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Andrea Lathrop
“Or... LeCun and co-author recently proved that next word prediction is the equivalent of bad, inefficient JEPA: arxiv.org/html/2610.39...” evidence ↗
1 expert 1 community 1 sources clustered

“Or... LeCun and co-author recently proved that next word prediction is the equivalent of bad, inefficient JEPA: arxiv.org/html/2610.39...”

Developing AI field signal Signal 2d ago

Counterfactual World Models:JEPAs That Reason About “What If”

1 directory member surfaced this signal.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Andrea Lathrop
“Bookmarking to read later, if I don't forget: arxiv.org/html/2611.24...” evidence ↗
1 expert 1 community 1 sources clustered

“Bookmarking to read later, if I don't forget: arxiv.org/html/2611.24...”

Established Models & releases Analysis 7d ago
defeating LLM nondeterminism essay

Defeating Nondeterminism in LLM Inference

4 experts are actively discussing the implications.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Andrea Lathrop
“Inference on an LLM embedded in an API is not deterministic, in practice, but the forward inference pass in a raw LLM *should theoretically be* but is mostly not, for kernel and GPU parallelism reasons, but the determinism can be recovered, if one is very, …” evidence ↗
2 experts 1 community 1 sources clustered

“Inference on an LLM embedded in an API is not deterministic, in practice, but the forward inference pass in a raw LLM *should theoretically be* but is mostly not, for kernel and GPU parallelism reasons, but the determinism can be recovered, if one is very, …”

“thinkingmachines.ai/blog/defeati... from Horace He”

4 experts discussed this · 12 posts
Adverb: Yeah, he's a fraud. Could pick apart predictions, arguments, technical knowledge but his repping for the companies he supposedly hates is the cleanest of giveaways.
Adverb: He thinks LLMs can't be deterministic which like, listening to someone who knows their subject -- even if not with terms like greedy sampling & batch invariant kernels -- could save people's time.
Adverb: I actually took like 5 minutes to look over one of his posts & stopped after the first embarrassing technical inaccuracy.
Open the full discussion →

What experts are discussing without an anchoring article

4 experts · 23 posts · 7d ago
Matches Research & technical analysis
“They are not the same thing, which is why we find the punchline on that comic to be humorous.” evidence ↗
Multiple readings Research & technical analysis · 1 Questions & unknowns · 1
Andrew D Wilson: I see why these guys are impressed with a technology that just fires off words without any thought
Andrew D Wilson: Gen AI does not think. They also don’t even solve maths problems - people solve maths problems with them. Left alone, no LLM goes looking for things to do And it’s only unoriginal because it has be…
Open the thread →