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
Active evidence filter

Showing developments with attributable Questions & unknowns 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.

Context & explanation Questions & unknowns
1 live reaction map in this view Jump to the evidence ↓
Multiple readings Introducing talkie: a 13B vintage language model from 1930 Questions & unknowns · 2 attributable reactions
Showing signals surfaced by austin ×
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 Release 7d ago
⚡ 432 h early
talkie vintage 1930 language model release

Introducing talkie: a 13B vintage language model from 1930

2 experts are actively discussing the implications.

Why this matches Questions & unknowns reaction 1 attributable expert contribution · Pwnallthethings
“It's a good question, and one someone tried out! With an LLM trained only on pre-1930 text they were able to teach it to write python code with just a few examples simonwillison.net/2026/Apr/28/...” evidence ↗
2 experts 1 community 1 sources clustered
Pwnallthethings austin
How the network is reacting Experts are approaching this through 2 distinct lenses.
Multiple readings

Questions & unknowns

1 expert

What remains unresolved or contested.

“It's a good question, and one someone tried out! With an LLM trained only on pre-1930 text they were able to teach it to write python code with just a few examples simonwillison.net/2026/Apr/28/...”

Context & explanation

1 expert

Background, chronology and why it matters.

“Like the entire point of an LLM is that it can generalize based on instructions. People have proven this! Go look at the source here; a LLM trained only on pre-1930s text was able to write python based on in-context examples. simonwillison.net/2026/Apr/28/...”

2 experts discussed this · 8 posts
brennan: love the sign off "for more information, follow <the person most famously wrong about all of this for years>"
Pwnallthethings: There's several examples of it, not least of all in math and engineering. But the reason *why* LLMs are good at that is because beyond a certain point, predicting words hyper-accurately requires so…
Pwnallthethings: I mean, the math cases are the most impressive because math *is* formal reasoning. That's ultimately what solving a math problem means
Open the full discussion →
Primary source simonwillison.net ↗
More conversations

What experts are discussing without an anchoring article

8 experts · 38 posts · 19d ago
Dylan Hadfield-Menell Grace Colin Max Woolf austin Singularity's Bounty e/cc coty.bsky.social austegard.com
Matches Questions & unknowns
“@jevbot.bsky.social 9.11 or 9.9 which is bigger?” evidence ↗
Reaction converging Questions & unknowns · 3 Concern & critique · 1 Context & explanation · 1
Grace: I made a little Jev-based language model! It uses beam search + Jev as a classifier to judge branches based on coherence and "success" (whatever that means). I've barely talked to it so I don't kno…
Grace: (you'll need to tag him)
Open the thread →
2 experts · 31 posts · 22h ago
austin jeremymorrell.dev
Matches Questions & unknowns
“MCP is everywhere: over 36k servers in the official registry. The quality of those servers is hit or miss, though. How do you stand out in that crowd? Start by seeing what agents actually do with y…” evidence ↗
austin: Hello from #AGNTCon! I'm presenting about what we learned building the @honeycomb.io MCP server. If you'd like to follow along, follow this thread! Otherwise, feel free to mute it. Let's get it sta…
austin: If you're trying to understand an MCP server by looking at access logs, you're gonna have a bad time. Here's a typical access log pattern - dozens of requests, all of them successful. JSON-RPC hide…
Open the thread →
5 experts · 12 posts · 12d ago
Eugene Vinitsky Daniel Lowd Eli Sennesh Phillip Carter austin
Matches Questions & unknowns
“If there were spaces where I could ask a human a question about code, those spaces are increasingly rare. That is a type of commons that is being lost as people lose a reason to frequent those spaces” evidence ↗
Multiple readings Questions & unknowns · 2 Policy & governance · 1
austin: I disagree w/anil on this point specifically because I do not believe AI can write non-copyleft code just as matter of legal principle. I don’t know how much it practically _matters_ but I think it…
austin: I also think that the upshot of this is that communities become more important than before, although they start to look very different. Maybe the principle of OSS starts to look more like standards…
Open the thread →
4 experts · 5 posts · 2d ago
tweety fish Mary Branscombe austin apenwarr.ca
Matches Questions & unknowns
“In all honesty the IoT device that never receives updates surely already had security holes. At least now you can make it do things. All discontinued-before-2026 Sonos devices? Hmmmmm” evidence ↗
Multiple readings Research & technical analysis · 1 Questions & unknowns · 1
austin: guess that anthropic paper on evil AI hiding in your fridge is more plausible now
apenwarr.ca: In all honesty the IoT device that never receives updates surely already had security holes. At least now you can make it do things. All discontinued-before-2026 Sonos devices? Hmmmmm
Open the thread →
3 experts · 10 posts · 16d ago
Christopher Mims Isaiah Bishop austin
Matches Questions & unknowns
“there is fun hipster b2b SaaS? why was I not informed” evidence ↗
Multiple readings Policy & governance · 1 Questions & unknowns · 1
austin: so that’s why I think saas is undead. orgs are not gonna change overnight; it will take decades. saas is gonna keep plodding along, zombie like, consolidating into fewer and fewer providers that lo…
austin: however the economics are very different as orgs change. if your compliance function is ai driven, if your ops is ai driven, etc etc. the more of the existing automatable work that gets automated, …
Open the thread →
2 experts · 8 posts · 7d ago
Liz Fong-Jones (方禮真) austin
Matches Questions & unknowns
“I think Luna is probably cheaper than Haiku although idk if it’s that much better. I will say that the 5.5 series makes anthropic more competitive? The thing is this stuff can change pretty frequen…” evidence ↗
Liz Fong-Jones (方禮真): This is true. Fable 5 and now Opus 5.5 and Fable 5.1 reliably write code that’s as good as if not better than I would have written myself. And they do it 100 times faster. They’re not making “slop”…
Liz Fong-Jones (方禮真): they *do* write over-verbose comments and excessive, tautologous tests. but that is a problem that can be managed and trimmed back
Open the thread →