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
Active evidence filter

Showing developments with attributable Context & explanation 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 Concern & critique Questions & unknowns Research & technical analysis
2 live reaction maps in this view Jump to the evidence ↓
Reaction converging Hugging face model evaluation security incident Concern & critique · 5 attributable reactions Multiple readings Introducing Muse Spark 1.1 Research & technical analysis · 2 attributable reactions
Showing signals surfaced by Tim Kellogg ×
Most important

The developments commanding sustained expert attention

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

Established Evaluation & benchmarks Signal 7d ago
⚡ 1 h early

Hugging face model evaluation security incident

18 experts across 6 network communities independently surfaced this.

Why this matches Context & explanation reaction 1 attributable expert contribution · Grace
“This headline is extremely funny given what happened (OpenAI hacked HF by accident) openai.com/index/huggin...” evidence ↗
18 experts 6 communities 1 sources clustered
Ethan Mollick Thomas Steinke Mor Naaman Tim Kellogg Djamé 🟥 Ryan Calo Grace Anna Mills +10
6 experts discussed this · 11 posts
Grace: This headline is extremely funny given what happened (OpenAI hacked HF by accident) openai.com/index/huggin...
Grace: Maybe the most concerning part is the OpenAI claim to not have known about this before investigating?
Grace: Well, I think the model passed the test
Open the full discussion →
Primary source openai.com ↗
Established Culture, work & education Release 20d ago
⚡ 1 h early
Meta Muse Spark 1.1 release

Introducing Muse Spark 1.1

4 experts across 3 network communities independently surfaced this.

Why this matches Context & explanation reaction 1 attributable expert contribution · Tim Kellogg
“here’s the official announcement. they mention “multi-million” context, though i can’t find an exact number ai.meta.com/blog/introdu...” evidence ↗
4 experts 3 communities 1 sources clustered
Ramon Astudillo Tim Kellogg Sung Kim Rami Krispin
How the network is reacting Experts are approaching this through 2 distinct lenses.
Multiple readings

Research & technical analysis

1 expert

Evidence, methods and technical implications.

“Meta's muse spark 1.1 is an industry-competitive agentic and coding model. across many agentic evals it rivals gpt-5.5 and opus-4.8. available now through the new meta model api and in meta ai. ai.meta.com/blog/introdu...”

Context & explanation

1 expert

Background, chronology and why it matters.

“here’s the official announcement. they mention “multi-million” context, though i can’t find an exact number ai.meta.com/blog/introdu...”

Primary source ai.meta.com ↗
Established Models & releases Release 14d ago
⚡ 4 h early
Thinking Machines Lab Inkling model launch

Inkling: Our Open-Weights Model - Thinking Machines Lab

5 experts across 3 network communities independently surfaced this.

Why this matches Context & explanation reaction 1 attributable expert contribution · Tim Kellogg
“Thinking Machines has a real model — 975B multimodal audio + image input it’s a generalist model. built for easy customization, TM has a great finetuning API audio up to 20 minutes, 1M token context for text thinkingmachines.ai/news/introdu...” evidence ↗
5 experts 3 communities 1 sources clustered
Nathan Lambert Tim Kellogg Sung Kim Cameron Evangelos Kazakos

“Thinking Machines has a real model — 975B multimodal audio + image input it’s a generalist model. built for easy customization, TM has a great finetuning API audio up to 20 minutes, 1M token context for text thinkingmachines.ai/news/introdu...”

“A big day for open model supporters in the U.S. Blog: thinkingmachines.ai/news/introdu... Model card: huggingface.co/thinkingmach...”

3 experts discussed this · 4 posts
Nathan Lambert: Thinking Machines just released with a ~1T param, 41B active, apache-2 model Benchmarks are a clear step up from Nemotron Ultra (55B active), new best American model, and omni input. A bit behind G…
Nathan Lambert: A big day for open model supporters in the U.S. Blog: thinkingmachines.ai/news/introdu... Model card: huggingface.co/thinkingmach...
Ben Recht: Yeah, that part is weird, no? The small model is *much* smaller and yet indistinguishable from the big one on benchmarks? I guess we'll have to wait and see, as it's not available yet.
Open the full discussion →
Primary source thinkingmachines.ai ↗
Debates gaining attention

What experts are discussing without an anchoring article

3 experts · 21 posts · 13d ago
Ted Underwood Tim Kellogg James MacGlashan
Matches Context & explanation
“And its an optimization because its strictly more efficient; because it does not need a context window and has constant compute/memory. MHA context windows are one of its biggest bottlenecks. You'r…” evidence ↗
James MacGlashan: An open model with this performance is great, but perhaps more striking is it looks like it's a hybrid model with 3/4th linear attention and 1/4 standard attention. This kind of performance with a …
Tim Kellogg: no most models have been hybrid for the last, oh probably the whole last year
Open the thread →
3 experts · 8 posts · 4d ago
Tim Kellogg jeremymorrell.dev austegard.com
Matches Context & explanation
“I think we still should have human review, but of the _shape_ of the code, not the individual lines. Have the agent provide a shape…erm shaped overview as part of the PR.” evidence ↗
Tim Kellogg: probably true there will always be a diminishing portion of software that needs to be reviewed by human, and it’ll never be zero, but right now you probably shouldn’t be reviewing code manually
Tim Kellogg: having started my career before PRs were normal, i’d say it was a combo of cargo culting the Linux project and an extension of “code review is good” but “pair programming doesn’t scale” and in that…
Open the thread →
3 experts · 5 posts · 12d ago
Ted Underwood Tim Kellogg Scott H. Hawley
Matches Context & explanation
“Choose one: 1. SV platform duopoly with lock-in and rent extraction 2. Communist State AI people who imagine a third option might be possible are utopian dreamers with no understanding of history” evidence ↗
Ted Underwood: Choose one: 1. SV platform duopoly with lock-in and rent extraction 2. Communist State AI people who imagine a third option might be possible are utopian dreamers with no understanding of history
Tim Kellogg: US-China duopoly with lockin and tax extraction?
Open the thread →
3 experts · 6 posts · 20d ago
Tim Kellogg austin mackuba.eu
Matches Context & explanation
“alright, back, what happened while i was gone?” evidence ↗
Tim Kellogg: alright, back, what happened while i was gone?
Tim Kellogg: hold on wait a minute Fable is just going to run out and OpenAI is simply going to release GPT-5.6 **on the same day**, and direct all previous attention from Fable to Sol + ultra no way, is that a…
Open the thread →