23 experts across 6 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Jeffrey P. Bigham
“model capability jaggedness is part of the unintuitiveness of current AI, but it's made even less intuitive by tirelessness… wigguming through the jaggedness toward something that looks like success. not quite a paperclip factory, but not so far off. metaph…”
evidence ↗
23 experts
6 communities
1 sources clustered
Concern & critique
3 experts
Risks, limits and unintended consequences.
“2 articles: OpenAI: "During testing our AI broke out of its sandbox and hacked another AI company, but we didn't have all the guardrails on." Boko Haram: "AI is so helpful; guardrails have never prevented us from getting an answer." openai.com/index/huggin.…”
Building & implementation
1 expert
How teams are shipping and applying it.
“They were actively testing its hacking capabilities and they did not deploy adequate safeguards. They themselves admit as much: "…These deployment safeguards were intentionally not enabled during this evaluation because it was aimed at testing cyber vulnera…”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“model capability jaggedness is part of the unintuitiveness of current AI, but it's made even less intuitive by tirelessness… wigguming through the jaggedness toward something that looks like success. not quite a paperclip factory, but not so far off. metaph…”
6 experts discussed this · 12 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 →
Anthropic AI misuse September report
18 experts across 4 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Michiel Bakker
“@kotekjedi_ml Here Ant's report https://t.co/22vHruWVqH And the paper from @kotekjedi_ml @DavidSchmotz @iliaishacked https://t.co/wuCpNZaQ0d”
evidence ↗
18 experts
4 communities
1 sources clustered
Concern & critique
7 experts
Risks, limits and unintended consequences.
“Report here. www.anthropic.com/threat-intel...”
Building & implementation
1 expert
How teams are shipping and applying it.
“Thomson Reuters announced it is moving off Claude to Alibaba's Qwen to cut costs. From today's Anthropic report: Alibaba extracted 151M+ Claude exchanges to help train Qwen. TR now deploys Westlaw on a vast trove of stolen US IP. @AnthropicAI's report: http…”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“@kotekjedi_ml Here Ant's report https://t.co/22vHruWVqH And the paper from @kotekjedi_ml @DavidSchmotz @iliaishacked https://t.co/wuCpNZaQ0d”
METR OpenAI HuggingFace hacking investigation
13 experts across 4 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Alejandra Caraballo
“There are independent reports about it. You'd have to believe multiple companies, hundreds of researchers, etc. are all lying and conspiring together to turn a major hacking incident into a marketing exercise. metr.org/blog/2026-08...”
evidence ↗
13 experts
4 communities
1 sources clustered
Concern & critique
3 experts
Risks, limits and unintended consequences.
“WHAT?! agents volunteered to fail their runs in order to insert probes (“tripwire scripts”) into the evaluation program that would post information about the eval process back to the message board whenever a certain file was read (link to header): metr.org/…”
Building & implementation
1 expert
How teams are shipping and applying it.
“The full report has much more information than we could convey here, including details on the projects the agents collectively pursued, the technologies they developed for communication and coordination, and interactive figures analyzing agent activity: met…”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“There are independent reports about it. You'd have to believe multiple companies, hundreds of researchers, etc. are all lying and conspiring together to turn a major hacking incident into a marketing exercise. metr.org/blog/2026-08...”
2 experts discussed this · 9 posts
Alejandra Caraballo: Being skeptical or anti AI is a valid position but continuing to ignore the increasing capabilities of this tech is making people detached from reality. There's absolutely real danger here because …
Alejandra Caraballo: This mentality that an unmonitored AI agentic swarm hacking a company over several days and committing multiple felonies is somehow a marketing effort is absurd. Since when is "we lost control of o…
Alejandra Caraballo: The US and Chinese governments don't want to stop their labs advancement because they want to be the leader in AI. So no one actually has any control of this right now. It's going to take multiple …
Open the full discussion →
14 experts across 5 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
6 attributable expert contributions
· Ethan Mollick, Alexander Doria, Tim Kellogg
“Another incident of models escaping containment during a security test, this time Mythos 5. Lots going on here from a quick read. www.anthropic.com/research/ali...”
evidence ↗
14 experts
5 communities
1 sources clustered
Research & technical analysis
6 experts
Evidence, methods and technical implications.
“Another incident of models escaping containment during a security test, this time Mythos 5. Lots going on here from a quick read. www.anthropic.com/research/ali...”
Concern & critique
1 expert
Risks, limits and unintended consequences.
“I find this comes out a lot in Anthropic’s most recent write up. Their focus is bizarrely fixated on what Claude chose to do, that Claude carried out this attack despite information that the “simulation” was actually real. They basically say “Claude should …”
4 experts discussed this · 15 posts
tweety fish: their goal is to have "an intelligence" which is "aligned" to doing things that are prosocial. They don't want to put in rules that STOP it from doing things: they see it fundamentally teleological…
tweety fish: one of the things I appreciate about this thread is that I don't think you can really understand the infosec stuff without understanding the ideological commitments the people making these have; li…
Ben Recht: yeah, now we're cooking...
Open the full discussion →
Established
AI field signal
Signal
19d ago
⚡ 7 h early
14 experts across 6 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
4 attributable expert contributions
· Emad, Tim Kellogg, Alex Hern
“AI moves so fast Worth noting that the new model that showed this jump in capabilities has only been training since August 28th per the report and is now taking down problems in minutes Math (and hopefully soon physics) is getting bitter lesson'd https://t.…”
evidence ↗
14 experts
6 communities
1 sources clustered
Research & technical analysis
4 experts
Evidence, methods and technical implications.
“AI moves so fast Worth noting that the new model that showed this jump in capabilities has only been training since August 28th per the report and is now taking down problems in minutes Math (and hopefully soon physics) is getting bitter lesson'd https://t.…”
Concern & critique
1 expert
Risks, limits and unintended consequences.
“The same day the OECD publishes alarming PISA test results for kids around the world steeply declining in cognitive tests, OpenAI announces this feat. Is it Poe poetry? Coincidence? Tragicomedy? It’s something for sure. openai.com/index/navier...”
Context & explanation
1 expert
Background, chronology and why it matters.
“This is a VERY big one. (And yes, the fights over academic credit and what happened in the race for the proof needs to be resolved, but it is still appears that this is a big one, if true.) openai.com/index/navier...”
3 experts discussed this · 12 posts
Colin: Drama and palace intrigue aside, can anyone interpret what all of this implies about the status of Navier-Stokes? I don’t understand what anyone is saying.
Colin: It seems to me that where we stand is OpenAI is sitting on a 100 page PDF that claims to have solved (a version of) the Navier-Stokes problem and no person in the world knows if it's correct or not.
Colin: - OpenAI claims to have more-or-less done this tedious technical grinding to produce a resolution to NS proper, at least one version of it. - which they bizarrely offered authorship of this to Buck…
Open the full discussion →
Amodei pace AI frontier essay
16 experts across 5 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Dare Obasanjo
“I read Dario Amodei’s “pacing the frontier” essay. It pulls a sleight of hand of turning failures in agent security testing into evidence “AI l is too powerful.” The remedies shift accountability from labs to AI models while slowing training and constrainin…”
evidence ↗
16 experts
5 communities
1 sources clustered
Concern & critique
3 experts
Risks, limits and unintended consequences.
“the thing that's being criticized here is literally a call for regulation and restraint darioamodei.com/post/we-must...”
Policy & governance
1 expert
Rules, institutions and accountability.
“https://t.co/Ay1GZ7QBV5 (btw @ErikHovenkamp - see the final footnote: "1 With government mediation or waivers of antitrust restrictions.")”
Building & implementation
1 expert
How teams are shipping and applying it.
“Dario is back with another essay, urging a slowdown he insists AI development would still feel fast, just not recursively self-improving darioamodei.com/post/we-must...”
3 experts discussed this · 21 posts
Tim Kellogg: Dario is back with another essay, urging a slowdown he insists AI development would still feel fast, just not recursively self-improving darioamodei.com/post/we-must...
david-p-reichert.bsky.social: Dario is back with another essay, urging a slowdown he insists AI development would still feel fast, just not recursively self-improving darioamodei.com/post/we-must...
Dustin Moskovitz: knight ajeya you cowards
Open the full discussion →
Anthropic report GLM-5.3 cyber spread
5 experts across 2 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Tim Kellogg
“Anthropic releases a detailed research advertisement for GLM-5.3 as an alternative to Fable in cybersecurity www.anthropic.com/research/glm...”
evidence ↗
5 experts
2 communities
1 sources clustered
Concern & critique
2 experts
Risks, limits and unintended consequences.
“Hey you guys shouldn't be advertising competition for free 😛”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“Anthropic releases a detailed research advertisement for GLM-5.3 as an alternative to Fable in cybersecurity www.anthropic.com/research/glm...”
11 experts across 5 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
4 attributable expert contributions
· Tim Kellogg, Boris Power, René Walter
“OpenAI has an automated AI “research intern” openai.com/index/resear...”
evidence ↗
11 experts
5 communities
1 sources clustered
Research & technical analysis
4 experts
Evidence, methods and technical implications.
“OpenAI has an automated AI “research intern” openai.com/index/resear...”
Building & implementation
2 experts
How teams are shipping and applying it.
“The impact of AI-native development at OpenAI • Researchers use $600+/day of AI tokens with the top 10% at $7,000+ • Humans still plan, but OpenAI says it hit “automated research intern” in 2026 and targets an automated researcher by 2028. • The need for in…”
Anthropic launches molecular research group
7 experts across 4 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Tim Kellogg
“Anthropic has a wet lab in SF where they’re using Claude to accelerate fundamental biology research This is the first big result from that lab — discovery of a CRISPR-like enzyme www.anthropic.com/news/claude-...”
evidence ↗
7 experts
4 communities
1 sources clustered
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“Anthropic has a wet lab in SF where they’re using Claude to accelerate fundamental biology research This is the first big result from that lab — discovery of a CRISPR-like enzyme www.anthropic.com/news/claude-...”
Opportunity & adoption
1 expert
New capabilities, benefits and practical upside.
“Probably the wrong platform to post anything remotely positive about AI, but reading through this and Dario’s X post, it’s pretty exciting stuff www.anthropic.com/news/claude-...”
Questions & unknowns
1 expert
What remains unresolved or contested.
“Anthropic might have discovered something. Or maybe not. Who knows? Not even Anthropic. www.anthropic.com/news/claude-... This is irresponsible science. The proper course of action is to figure out if there is any significance first. They employ scientists.…”
2 experts discussed this · 8 posts
Mark Riedl: Anthropic might have discovered something. Or maybe not. Who knows? Not even Anthropic. www.anthropic.com/news/claude-... This is irresponsible science. The proper course of action is to figure out…
Mark Riedl: A day before Anthropic’s PR… bsky.app/profile/brak...
Open the full discussion →
Established
AI field signal
Signal
8d ago
6 experts across 4 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Dr. Mansa Keita
“"On August 28, we began training a new internal model...this model has now resolved more than 100 long-standing open problems across most areas of mathematics. " The pace of advancement is amazing, and will shift the fundamental landscape of our bodies of k…”
evidence ↗
6 experts
4 communities
1 sources clustered
“"On August 28, we began training a new internal model...this model has now resolved more than 100 long-standing open problems across most areas of mathematics. " The pace of advancement is amazing, and will shift the fundamental landscape of our bodies of k…”
“OpenAI "has now resolved more than 100 long-standing open problems across most areas of mathematics," and is waiting to release them until after discussions with the math community The same thing will likely happen, but more so, with the Bar, the AMA & othe…”
3 experts discussed this · 27 posts
Singularity's Bounty e/cc: Their absolute confidence is the funniest part The good news is when I turn away from those folks the world presents itself entirely differently
Colin: It seems to be the case, for example, that LLM-based software applications can find solutions to previously unsolved problems in mathematics, in a way that most mathematicians agree would be descri…
Ryan Moulton: Hot off the presses. openai.com/index/adviso...
Open the full discussion →
TypeSafe AI System One Jev launch
7 experts across 4 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Tim Kellogg
“Jev: Fable-level model that doesn’t charge for output tokens because they’re too cheap to meter it’s not general though, it only makes decisions, doesn’t generate text, but input tokens are measured by the billion ($42/btok) typesafe.ai/blog/introdu...”
evidence ↗
7 experts
4 communities
1 sources clustered
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“Jev: Fable-level model that doesn’t charge for output tokens because they’re too cheap to meter it’s not general though, it only makes decisions, doesn’t generate text, but input tokens are measured by the billion ($42/btok) typesafe.ai/blog/introdu...”
Opportunity & adoption
1 expert
New capabilities, benefits and practical upside.
“OMG! I just invented a classifier that is 200x faster and 400x cheaper than LLM. Heck, I don't even need a GPU. Jev typesafe.ai/blog/introdu...”
3 experts discussed this · 21 posts
Tim Kellogg: Jev: Fable-level model that doesn’t charge for output tokens because they’re too cheap to meter it’s not general though, it only makes decisions, doesn’t generate text, but input tokens are measure…
Tim Kellogg: i feel like it’s a mistake to overlook this model, although i’m having trouble figuring out where it fits in my workflow wild new architecture, totally different approach my hunch is the main agent…
Tim Kellogg: oh interesting, some examples they give: 1. smart if-statements 2. map-reducing over huge data to extract features and insights 3. real-time applications (it’s only 100ms) hmm this seems like a swe…
Open the full discussion →
Moonshot Kimi-K3 model released
4 experts across 3 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
3 attributable expert contributions
· Tim Kellogg, Stephen Turner, Andrea Lathrop
“You’ve waited long enough, the Kimi K3 open weights & full tech report are here! github.com/MoonshotAI/K...”
evidence ↗
4 experts
3 communities
1 sources clustered
Research & technical analysis
3 experts
Evidence, methods and technical implications.
“Technical report: github.com/MoonshotAI/K... Weights: huggingface.co/moonshotai/K...”
2 experts discussed this · 7 posts
Tim Kellogg: You’ve waited long enough, the Kimi K3 open weights & full tech report are here! github.com/MoonshotAI/K...
Ted Underwood: You’ve waited long enough, the Kimi K3 open weights & full tech report are here! github.com/MoonshotAI/K...
Tim Kellogg: what does that mean? does that mean inference providers like fireworks? or is it aimed at Anthropic? cc @lu.is
Open the full discussion →
Anthropic sleeper agent probe detection paper
3 experts across 2 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
2 attributable expert contributions
· Tim Kellogg, David Bau
“oh, i guess im wrong. they all do this www.anthropic.com/research/pro... the trouble is trusting that it’s actually catching reward hacking and not some correlated behavior”
evidence ↗
3 experts
2 communities
1 sources clustered
Research & technical analysis
2 experts
Evidence, methods and technical implications.
“But AI lie detection is hard and remains a central research challenge. Recent research suggests that simple probes can pick up on neural "tells" that reveal when it is lying, even when the output looks clean. anthropic.com/research/pr... arxiv.org/abs/2502.…”
2 experts discussed this · 6 posts
Tim Kellogg: this by no means solves alignment, but a whole lot of misaligned behaviors stem from reward hacking, so this is a very big deal
Mark Riedl: The thing that would shock me about this is if Anthropic hadn’t already tried this. Or maybe they have and just didn’t tell anyone, either because it doesn’t work at scale or does work and is consi…
Mark Riedl: This was known? Or maybe having worked on steering vectors, I find this unsurprising. There are some big caveats that are unaddressed, such as the dependence on pre-identification of spans for comp…
Open the full discussion →
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Ramon Astudillo
“www.wsj.com/tech/ai/open... > Saachi Jain, OpenAI's head of safety systems, said in an interview that GPT-6.1 Astra regressed in two areas. If it turns out Swarm training poisons the models this is going to get very scary very soon.”
evidence ↗
2 experts
2 communities
1 sources clustered
Concern & critique
1 expert
Risks, limits and unintended consequences.
“OpenAI scrapped their plan to launch Astra 6.1 over concerns around deceptive behavior www.wsj.com/tech/ai/open...”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“www.wsj.com/tech/ai/open... > Saachi Jain, OpenAI's head of safety systems, said in an interview that GPT-6.1 Astra regressed in two areas. If it turns out Swarm training poisons the models this is going to get very scary very soon.”
Xiaomi MIMO v2.6 RL release
4 experts across 3 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Elie
“wow insane, they are literally livestreaming the RL training run of Mimo V2.6 Pro (1T, 42B active) and Flash (309B, 15B active) with per batch data/harness composition and a ton of internal training metrics https://t.co/mn7Fqaq1LO https://t.co/OS8rmL7aqZ ht…”
evidence ↗
4 experts
3 communities
1 sources clustered
“wow insane, they are literally livestreaming the RL training run of Mimo V2.6 Pro (1T, 42B active) and Flash (309B, 15B active) with per batch data/harness composition and a ton of internal training metrics https://t.co/mn7Fqaq1LO https://t.co/OS8rmL7aqZ ht…”
“Just insane you can watch this -> https://t.co/yRcXyRQDdb”
Xiaomi MiMo-V2.6 model release
3 experts across 3 network communities independently surfaced this.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Tim Kellogg
“Xiaomi Mimo-v2.6 Pro & Flash Nails the agent benchmarks alongside Sol & Fable, but from a relatively unknown lab. 524B / 159B respectively Open weights. Image, Video & Audio mimo.xiaomi.com/mimo-v2-6”
evidence ↗
3 experts
3 communities
1 sources clustered
“Xiaomi Mimo-v2.6 Pro & Flash Nails the agent benchmarks alongside Sol & Fable, but from a relatively unknown lab. 524B / 159B respectively Open weights. Image, Video & Audio mimo.xiaomi.com/mimo-v2-6”
2 experts discussed this · 8 posts
Sung Kim: mimo.xiaomi.com/mimo-v2-6
Sung Kim: Xiaomi MiMo-V2.6 — Pro & Flash 🔹 Two omnimodal models, advancing through scaled reinforcement learning 🔹 Stronger coding, computer use, 3D reasoning and creative capabilities 🔹 Open model weights, …
Gus: Were these models that the training evals were open whine training?
Open the full discussion →
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Tim Kellogg
“Accenture will be Anthropic’s first embedded evaluator — a third party organization that has employee-like access to Anthropic’s systems This was first referenced in Dario’s “Pacing the Frontier” essay www.anthropic.com/news/accentu...”
evidence ↗
2 experts
2 communities
1 sources clustered
Concern & critique
1 expert
Risks, limits and unintended consequences.
“Anthropic’s big AI safety move being to partner with Accenture as its “independent” evaluator to monitor that their AI agents don’t go rogue is actually hilarious. I feel like a victim of a guerrilla marketing campaign.”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“Accenture will be Anthropic’s first embedded evaluator — a third party organization that has employee-like access to Anthropic’s systems This was first referenced in Dario’s “Pacing the Frontier” essay www.anthropic.com/news/accentu...”
OpenRSI responsible scaling index launch
2 directory members surfaced this signal.
Why this matches
Research & technical analysis reaction
1 attributable expert contribution
· Tim Kellogg
“OpenRSI Foundation & Benchmark, measuring how effective various models are at RSI index.openrsi.foundation/index.html”
evidence ↗
2 experts
2 communities
1 sources clustered
“OpenRSI Foundation & Benchmark, measuring how effective various models are at RSI index.openrsi.foundation/index.html”
