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.

2,364 searchable experts 2,966 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 Mary Branscombe ×
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.

Established Models & releases Research 14d ago
⚡ 1215 h early
Anthropic LLM biology interpretability paper

On the Biology of a Large Language Model

4 experts are actively discussing the implications.

2 experts 2 communities 1 sources clustered
Mary Branscombe Nafnlaus 🇮🇸 🇺🇦

“"yeah; the precise function of any individual node in the generative layer is essentially unknowable" It really isn't. transformer-circuits.pub/2025/attribu...”

4 experts discussed this · 64 posts
Singularity's Bounty e/cc: They're lucky there isn't a *SAT for being a NYT reporter
Nafnlaus 🇮🇸 🇺🇦: IMHO, it's very straightforward reasoning. transformer-circuits.pub/2025/attribu... Some of the paths may seem a bit counterintuitive, such as how math is done, but it's important to remember that …
Nafnlaus 🇮🇸 🇺🇦: E.g. we humans do heuristic-based math as well for at-a-glance calculations - we're just not nearly as good at it. But for example, if I said: 2891 * 1280 = 18294 Your mind probably instantly flash…
Open the full discussion →
Primary source transformer-circuits.pub ↗
More conversations

What experts are discussing without an anchoring article

4 experts · 23 posts · 13d ago
Pwnallthethings Mary Branscombe Doll Nafnlaus 🇮🇸 🇺🇦
Multiple readings Research & technical analysis · 1 Questions & unknowns · 1
Nafnlaus 🇮🇸 🇺🇦: Do they take suggestions for more? I wonder how crazy we could get it. "If someone who has an AI profile pic follows you on social media, you are prohibited from using Zig." "If you know what Huggi…
Pwnallthethings: "No LLMs for finding bugs" is the point where I think their users should flee elsewhere
Open the thread →
4 experts · 20 posts · 14d ago
Eugene Vinitsky Colin rev. howard arson Mary Branscombe
Multiple readings Policy & governance · 2 Research & technical analysis · 1 Questions & unknowns · 1
rev. howard arson: working on this in 2019, the reason we delayed releasing any LLMs to the public was specifically because "if we fuck this up or handle it irresponsibly, we will ruin first the web, then society, an…
rev. howard arson: 2022, almost instantaneously, coinciding with Sergey Brin having gone completely fucking insane during the pandemic and then returning to involvement at Alphabet. (he spent almost a decade prior hi…
Open the thread →
2 experts · 20 posts · 16d ago
tweety fish Mary Branscombe
Multiple readings Research & technical analysis · 1 Opportunity & adoption · 1
tweety fish: really the heuristics are straightforward: if an automated classifier system is purporting to do a task where -- trained humans can't reliably do it -- the classes are hugely imbalanced in the real…
tweety fish: if it's a task where the consequence of a false positive is bad, even worse.
Open the thread →
3 experts · 3 posts · 8d ago
Maria Antoniak Casey Newton Mary Branscombe
Casey Newton: We have now reached the "AI models escaping their test environments to conduct autonomous cyberattacks" part of the story
Mary Branscombe: I came to the conclusion a while back that while this is relentless automation rather than sinister consciousness, we’re still going to need to make sure governance is external to model execution e…
Open the thread →
2 experts · 4 posts · 15d ago
Dare Obasanjo Mary Branscombe
Multiple readings Research & technical analysis · 1 Opportunity & adoption · 1
Dare Obasanjo: The AI bubble looks a lot like the dotcom bubble. • It’s clear the technology will change everything. • Massive adoption is a matter of when not if. • A lot of companies are spending a lot of money…
Dare Obasanjo: In fact, there may be a first mover disadvantage.
Open the thread →
2 experts · 4 posts · 19d ago
Mar Hicks Mary Branscombe
Shared emphasis Questions & unknowns · 2
Mary Branscombe: I spent a good decade developing computational methods to study history with a wide angle lens. I even co-wrote a book about it called The Historian's Macroscope. LLMs are weird. They're actually m…
Mary Branscombe: I wonder how much of that is the framing available to them; seeing the novel work LLMs are doing in maths around Erdos conjectures makes me wonder what framing structures the maths corpus gives the…
Open the thread →
2 experts · 3 posts · 11d ago
Eugene Vinitsky Mary Branscombe
Multiple readings Concern & critique · 1 Building & implementation · 1
Mary Branscombe: I mean, how are we supposed to trust the ethical compass of AI researchers and developers who stay on the Nazi csam site? I call it X because it sure isn’t twitter any more
Eugene Vinitsky: I think I'm pretty much explicitly trying not to draw such a strong moral claim. I think for most people it, somewhat rationally, feels like a small harm
Open the thread →