New
AI field signal
Analysis
2h ago
verification is not knowledge essay
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Matt Goldrick
“I'd also recommend @jessicahullman.bsky.social's recent piece for thinking about the deeper questions of what peer review should be for -- beyond simply finding errors substack.com/home/post/p-...”
evidence ↗
2 experts
1 community
1 sources clustered
“I'd also recommend @jessicahullman.bsky.social's recent piece for thinking about the deeper questions of what peer review should be for -- beyond simply finding errors substack.com/home/post/p-...”
“Understanding is often pitted against output in AI-for-science debates. But seeing understanding as an optional mental state revives old ideals of value-free science & mechanical objectivity that philosophers reject as insufficient Thanks Cosmos for publish…”
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Al Merose
“With my friend @realcromwellian.bsky.social’s help (and his Mac Studio), we got a 126M param NanoGPT model training in ~165s — primarily written in SQL. github.com/xqlsystems/x...”
evidence ↗
1 expert
1 community
1 sources clustered
“With my friend @realcromwellian.bsky.social’s help (and his Mac Studio), we got a 126M param NanoGPT model training in ~165s — primarily written in SQL. github.com/xqlsystems/x...”
AI enclosure scientific commons
3 experts across 2 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Jeffrey Vagle
““Many observers judge enclosure of the commons to have been theft at a societal scale. This article argues that AI provides precisely the tools needed to enable a previously-infeasible enclosure of the information commons.” arxiv.org/abs/2610.27251”
evidence ↗
3 experts
2 communities
1 sources clustered
““Many observers judge enclosure of the commons to have been theft at a societal scale. This article argues that AI provides precisely the tools needed to enable a previously-infeasible enclosure of the information commons.” arxiv.org/abs/2610.27251”
AI education meta-analyses audit finds flaws
4 experts across 3 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Ben Williamson
“The research record on AI in education is already pretty appalling. And researchers just keep cranking up AI to do yet more meta-reviews that then recycle the findings of sloppy studies like this one. hechingerreport.org/proof-points...”
evidence ↗
4 experts
3 communities
1 sources clustered
“The research record on AI in education is already pretty appalling. And researchers just keep cranking up AI to do yet more meta-reviews that then recycle the findings of sloppy studies like this one. hechingerreport.org/proof-points...”
ODE sampler discrete diffusion paper
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Shubhendu Trivedi
“Looks cool arxiv.org/abs/2610.40618”
evidence ↗
2 experts
1 community
1 sources clustered
“Looks cool arxiv.org/abs/2610.40618”
agentic AI myth set in motion essay
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Frank Pasquale
““We weight sources by trust. We discount cranks, privilege people with good track records, and know that one careful monograph can outweigh ten thousand blog posts…A model’s starting map weights things largely by how often they appear in the corpus” braddel…”
evidence ↗
2 experts
2 communities
1 sources clustered
““We weight sources by trust. We discount cranks, privilege people with good track records, and know that one careful monograph can outweigh ten thousand blog posts…A model’s starting map weights things largely by how often they appear in the corpus” braddel…”
researchers using AI for peer review
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Stephen Turner
“More than half of researchers now use AI for peer review — often against guidance (Dec 2025): www.nature.com/articles/d41...”
evidence ↗
1 expert
1 community
1 sources clustered
“More than half of researchers now use AI for peer review — often against guidance (Dec 2025): www.nature.com/articles/d41...”
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Stephen Turner
“More than half of researchers now use AI for peer review — often against guidance (Dec 2025): www.nature.com/articles/d41...”
evidence ↗
1 expert
1 community
1 sources clustered
“More than half of researchers now use AI for peer review — often against guidance (Dec 2025): www.nature.com/articles/d41...”
EEF seeks AI edtech evaluator researchers
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Ben Williamson
“"The EdTech Testbed Programme will run from 2026 to 2030... The first stage, led by a research consortium headed by Ipsos, will involve up to 75 rapid evaluations of education technology products." That's a lot of live edtech and AIed testing in schools www…”
evidence ↗
1 expert
1 community
1 sources clustered
“"The EdTech Testbed Programme will run from 2026 to 2030... The first stage, led by a research consortium headed by Ipsos, will involve up to 75 rapid evaluations of education technology products." That's a lot of live edtech and AIed testing in schools www…”
AI coding agents more code less software
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Mor Naaman
“Weirdly, same trend and similar takeaways (and methods!) in the research summarized here: AI coding agents generate more code, but not more software - Ars Technica share.google/tqH0TzAJJpeR...”
evidence ↗
2 experts
2 communities
1 sources clustered
“Weirdly, same trend and similar takeaways (and methods!) in the research summarized here: AI coding agents generate more code, but not more software - Ars Technica share.google/tqH0TzAJJpeR...”
AI copyright memorization generative paper
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Nicole Hennig
“The Files are in the Computer: On Copyright, Memorization, and Generative AI arxiv.org/abs/2404.12590… #AI #copyright #memorization”
evidence ↗
1 expert
1 community
1 sources clustered
“The Files are in the Computer: On Copyright, Memorization, and Generative AI arxiv.org/abs/2404.12590… #AI #copyright #memorization”
New
AI field signal
Signal
17h ago
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Nicole Hennig
“Making it easier for websites to express AI training preferences - OpenAI openai.com/index/helping-… #AI #training”
evidence ↗
1 expert
1 community
1 sources clustered
“Making it easier for websites to express AI training preferences - OpenAI openai.com/index/helping-… #AI #training”
Developing
AI field signal
Resource
1d ago
⚡ 215 h early
America.gov US government resource
25 experts across 5 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Lightspeed
“What if AI could handle the paperwork you’ve been putting off? “Being able to file a passport online and have it be done through a chatbot just feels like the future,” says @Machiz. That’s part of the longer-term ambition for https://t.co/jzCcQ8Xa7i, which …”
evidence ↗
25 experts
5 communities
1 sources clustered
Policy & governance
2 experts
Rules, institutions and accountability.
“America.gov does not do pronouns. Any pronouns. (Courtesy of my daughter!) aifails.substack.com/p/pronouns”
Concern & critique
1 expert
Risks, limits and unintended consequences.
“Jessica Ji weighed in on the new #AI-powered https://t.co/sEHIhq7mvb website for @USATODAY, saying that it can be a useful resource for people trying to navigate U.S. government websites. She also notes its risks: “When this kind of AI system is empowered t…”
Building & implementation
1 expert
How teams are shipping and applying it.
“The chatbot comes as an expansion of the Office of Personnel Management’s work to modernize access to public federal workforce data. Integration with America.gov is planned. www.youtube.com/watch?v=3SUx... Read more by @madialder.bsky.social: fedscoop.com/o…”
2 experts discussed this · 3 posts
Julian Sanchez: Ok wow. This is both extensive and deeply surreal. Tell America.gov to Play Minecraft.
Julian Sanchez: Amusingly, the chatbot appears to be unable to recognize that it said anything weird. It “thinks” it just responded that it can’t play minecraft.
Open the full discussion →
LLM distillation teacher bias paper
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· AI Firehose
“Iterative Calibration and Learning (IterCaL) innovatively addresses teacher bias in LLM distillation, showing polynomial convergence to optimal student performance without needing target-domain feedback, potentially redefining language model training. https…”
evidence ↗
1 expert
1 community
1 sources clustered
“Iterative Calibration and Learning (IterCaL) innovatively addresses teacher bias in LLM distillation, showing polynomial convergence to optimal student performance without needing target-domain feedback, potentially redefining language model training. https…”
advantage clipped policy optimization RL paper
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· AI Firehose
“Researchers introduced Advantage Clipped Policy Optimization (ACPO), a method that stabilizes gradient estimates via clipping. This innovation boosts training efficiency and accuracy for language models in reasoning tasks, yielding significant gains on benc…”
evidence ↗
1 expert
1 community
1 sources clustered
“Researchers introduced Advantage Clipped Policy Optimization (ACPO), a method that stabilizes gradient estimates via clipping. This innovation boosts training efficiency and accuracy for language models in reasoning tasks, yielding significant gains on benc…”
AI theoretical tool cognitive science paper
4 experts across 2 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Nicole Rust
“One paper I thought a lot while writing was this by @irisvanrooij.bsky.social et al (linked in my piece; I recommend). There's a role for AI for "assessing whether something is computationally possible or not, in principle and in practice." link.springer.co…”
evidence ↗
4 experts
2 communities
1 sources clustered
“One paper I thought a lot while writing was this by @irisvanrooij.bsky.social et al (linked in my piece; I recommend). There's a role for AI for "assessing whether something is computationally possible or not, in principle and in practice." link.springer.co…”
“Theory is more important now than ever! link.springer.com/article/10.1...”
2 experts discussed this · 21 posts
Iris van Rooij: Footnote 23 in Reclaiming AI: link.springer.com/article/10.1...
Iris van Rooij: "no finite set of ‘impressive’ observations about AIs/LLMs establishes that they exhibit a (human) cognitive capacity. (...) if we propose that a machine has a cognitive capacity X, we must also fo…
Olivia Guest: Footnote 23 in Reclaiming AI: link.springer.com/article/10.1...
Open the full discussion →
inference cooperative AI governance
3 experts across 3 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
2 attributable expert contributions
· Nathan Schneider, B! 🐝 Cavello (they/them)
“There are now hand-made doodles on the Inference Cooperative home page. It's a small, superficial thing, but to me it matters: To make clear that the heart of this co-op is the people who join, not the AI tools we share. https://inference.coop”
evidence ↗
3 experts
3 communities
1 sources clustered
Building & implementation
1 expert
How teams are shipping and applying it.
“Maybe relevant @ntnsndr.in in context of inference.coop (congrats on the launch!)”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“There are now hand-made doodles on the Inference Cooperative home page. It's a small, superficial thing, but to me it matters: To make clear that the heart of this co-op is the people who join, not the AI tools we share. https://inference.coop”
2 experts discussed this · 6 posts
Nathan Schneider: A month ago, I found myself in the unexpected position of founding an AI startup. Today, the Inference Co-op adopted its first real governance charter. We can't make AI good, but we can own our cho…
B! 🐝 Cavello (they/them): A month ago, I found myself in the unexpected position of founding an AI startup. Today, the Inference Co-op adopted its first real governance charter. We can't make AI good, but we can own our cho…
Marco Z: where does it run ? is that Tinfoil?
Open the full discussion →
Para-Gaussian neural network approximations
1 directory member surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Mihai Nica
“This is the Edgeworth expansion idea from the classical CLT applied to deep networks. The para-Gaussian achieves an error rate of O(1/n²) instead of the Gaussian’s O(1/n), which matters more and more as depth increases. Check out arxiv.org/abs/2610.37388 fo…”
evidence ↗
1 expert
1 community
1 sources clustered
“This is the Edgeworth expansion idea from the classical CLT applied to deep networks. The para-Gaussian achieves an error rate of O(1/n²) instead of the Gaussian’s O(1/n), which matters more and more as depth increases. Check out arxiv.org/abs/2610.37388 fo…”
