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 Mark J. Nelson ×
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 Resource 3d ago
Anthropic Institute tabletop exercise lead hiring

Tabletop Exercise Lead, The Anthropic Institute

6 experts are actively discussing the implications.

2 experts 2 communities 1 sources clustered
Mark J. Nelson tachikoma

“Bit of an odd job here. Anthropic looking for a person to design and run tabletop exercises and wargames.”

6 experts discussed this · 7 posts
Mark J. Nelson: Bit of an odd job here. Anthropic looking for a person to design and run tabletop exercises and wargames.
Tyler Shoemaker: Screaming for @mkirschenbaum.bsky.social’s attention
Matthew Kirschenbaum: Man would I love the inside story on this one.
Open the full discussion →
Primary source job-boards.greenhouse.io ↗
Established AI field signal Resource 17d ago
Google Colab AI notebook resource

Google Colab

2 experts are actively discussing the implications.

1 expert 1 community 1 sources clustered
Mark J. Nelson

“And that is all I currently teach about classic RL theory in the current iteration (this is a broad AI survey class, so space for RL is about 3 weeks). From there I go on to what you have to modify about this story to make DQN work – without, alas, the conv…”

2 experts discussed this · 17 posts
Mark J. Nelson: I've been attempting to distill tabular RL into a minimal set of lecture notes that can be introduced after random policy search but without the MDP formalism. Here's what I have so far (thread, wi…
Mark J. Nelson: Last week, we looked at neural policy search, which we saw actually finds successful policies on CartPole. However, one thing it doesn't do is any type of temporal credit assignment. Temporal-diffe…
Mark J. Nelson: Applying it to neural networks like we used in the previous notebook needs a few extra tweaks, so we will start first with a discrete grid-world environment called FrozenLake. This will let us use …
Open the full discussion →
Primary source colab.research.google.com ↗
Established Safety & security Research 12d ago
mechanistic interpretability ACL paper

Mechanistic?

3 experts are actively discussing the implications.

1 expert 1 community 1 sources clustered
Mark J. Nelson

“There's at least one actually good paper's worth of methodological critique of mechinterp, but not from that faction, along pretty different lines, and not contemptuous.”

3 experts discussed this · 5 posts
rev. howard arson: i feel like the level of contempt for mechinterp by bender's faction is incredibly weird, given that there is as far as i can tell not a single paper's worth of methodological criticism of what is …
rev. howard arson: i do think that mechanistic interpretability has somewhat expansive claims about what it is seeing for how narrow its instruments are, but those narrow instruments appear to be detecting things wit…
Mark J. Nelson: Yes! There's a good paper in Synthese on that, elaborating on how much of this critique is based on "strong Gricean assumptions ... now known to be empirically untenable".
Open the full discussion →
Primary source aclanthology.org ↗
Established AI field signal Resource 7d ago
Lynn Cherny AI highlights newsletter

Things I Think Are Awesome | Lynn Cherny | Substack

2 experts are actively discussing the implications.

2 experts 2 communities 1 sources clustered
Lynn Cherny Mark J. Nelson

“Things I Think Are Awesome by @arnicas.bsky.social – roundups of creative AI and other tech in the creative/game/elit/art spaces”

2 experts discussed this · 2 posts
Meredith Martin: this is all @arnicas.bsky.social !!! Not me!
Lynn Cherny: arnicas.substack.com 😅
Open the full discussion →
Primary source arnicas.substack.com ↗
More conversations

What experts are discussing without an anchoring article

5 experts · 27 posts · 21h ago
Mark J. Nelson Eli Sennesh SE Gyges Meredith Martin jake
Shared emphasis Questions & unknowns · 2
SE Gyges: unpopular opinion, "admits uncertainty" and "flags nuances" are shit. they are terrible fucking traits actually. is the sky blue? do we really know that? what is blue, anyway? many cultures have di…
SE Gyges: if gemini argon is a remarkably good model it is good because it is simply not optimized for this specific set of toxic, absurd, brain-poisoned behavior. it would actually defy belief that it happe…
Open the thread →
3 experts · 6 posts · 9d ago
Eugene Vinitsky Mark J. Nelson Mary Branscombe
Multiple readings Research & technical analysis · 2 Policy & governance · 1
Eugene Vinitsky: Now that character AI is dead, what on earth was the point of it? Why did the researchers that created it think it was a good idea?
Eugene Vinitsky: Like, they were obviously correct that they would make a bunch of money by doing so but given the crew they would have made a bunch of money doing literally anything
Open the thread →
2 experts · 6 posts · 4d ago
Mark J. Nelson Isaiah Bishop
Multiple readings Markets & investment · 1 Building & implementation · 1
Isaiah Bishop: We have such a weird chip industry. This company is in austin now. But the previous co-founder made another company that was acquired by AMD. We have another company called photonic that just did a…
Isaiah Bishop: We have a lot of weird legacy talent in things like fibre optics because of Nortel's impact, which was also downstream of govt funding. And also waterloo being just a monsterous university for tech…
Open the thread →
2 experts · 3 posts · 16d ago
Naomi Saphra Mark J. Nelson
Shared emphasis Research & technical analysis · 2
Naomi Saphra: All scientific field distinctions are ultimately socially defined categories. So the Transformers paper was an “NLP paper” not because it was about language modeling and MT, but because Vaswani wen…
Mark J. Nelson: by a similar token, the Transformers paper is a PL paper because Łukasz Kaiser has invented or been primary contributor to like three or four functional DSLs, classic PL researcher behavior
Open the thread →
2 experts · 3 posts · 14d ago
Mark J. Nelson Dan Greene
Dan Greene : Targeting adobe, ms, and infosys is targeting liberal execs, liberal foreign policies (in Microsoft's case in particular) and foreign capital--especially the kind us ai would like to displace. The …
Mark J. Nelson: The five companies in the WITCH acronym have been a bete noir for a while, and not only on the right, so if it were just them there might not have even been much pushback. I guess the other two are…
Open the thread →
2 experts · 2 posts · 19d ago
Mark J. Nelson Matthew L. Jones
Multiple readings Research & technical analysis · 1 Context & explanation · 1
Matthew L. Jones: They are such engrossing characters, deep scientifically in radically different ways where the weirdness matters, and show (without telling) so many crucial facets of what history of science done w…
Mark J. Nelson: I did one of these, and mine was about why you should download and experiment with open-weights "base" LLMs (i.e. pretrained only, just as language models) not solely use the chatbot-style models.
Open the thread →