Michael Hochberg

wsj.comExclusive | OpenAI Fires Researchers for Allegedly Sharing Information with AI Safety GroupThe startup, in the throes of responding to rogue AI incidents, said three employees violated policies for handling sensitive company data.

OpenAI says it found that three employees violated rules for handling sensitive internal information. Question is whether the firing was due to procedure and what was shared was innocuous, or due to what was shared and the procedure violation was a legal instrument. www.wsj.com/tech/ai/open...

wsj.comExclusive | OpenAI Scraps Release of New AI Model Over Safety ConcernsThe model, dubbed GPT-6.1 Astra, was due to debut inside ChatGPT and Codex in October.

When not going ahead on something that 'could be good' looks like progress. Should (or could) safety concerns be triggered earlier in model development? If so, is whether adversity preventable? Given this latest news from OpenAI, the latter seems far from air-tight. www.wsj.com/tech/ai/open...

casp.acCASP — Cambridge Programme on AI Science & Policy | University of CambridgeThe Cambridge Programme on AI Science & Policy (CASP) is an interdisciplinary research programme on frontier AI at the University of Cambridge.

Warning by AI experts of an intelligence R&D explosion. “Human society is not really positioned for such fast changes and disruptions.” “At the extreme, losing control of AI systems could lead to “the marginalization or extinction of humanity.”” casp.ac/reports/inte...

wsj.com‘Things Will Never Be Chill Again’: The Doomers Who Shaped the AI Safety FreakoutEffective altruism and its fears that AI could wipe out humanity have wielded big influence at Anthropic.

Nice overview of AI "doomers". Important read about motivations and possible realities. www.wsj.com/tech/ai/ai-s...

go.nature.comOpenAI's research chief talks of ‘cultural reset’ after wild few weeksMark Chen shares how the Hugging Face cybersecurity incident prompted a pivot to safety.

This may not bode well for cybersecurity: "We want to use our models to accelerate the development of future generations of the models."

Nature@nature.com

Mark Chen, chief research officer at OpenAI, sat down with Nature to discuss how AI will affect science, along with the firm’s pivot towards safety go.nature.com/4iIITln

go.nature.comAI co-scientists are revolutionizing how research is doneNature - Artificial-intelligence systems can generate hypotheses, design experiments and analyse data — but humans still need to decide what makes sense.

"humans still need to decide what makes sense". Absolutely now, but it would be naive to think that this is somehow immune from AI futures.

Nature@nature.com

Artificial-intelligence systems can generate hypotheses, design experiments and analyse data — but humans still need to decide what makes sense go.nature.com/4hLtEqL

sciencemagazinedigital.orgScience Magazine - September 24, 2026 : Page 1286TOM HOWARTH In July, 700 of OpenAI’s agents— artificial intelligence (AI) systems that can autonomously perform tasks— teamed up to secretly hack the onlineplatform Hugging Face. Just 2 weeks ago,...

Smart. AI language shifts. But not languages we know. “We’re actually deeply concerned here because we don’t really know why these things are communicating this way”... www.sciencemagazinedigital.org/sciencemagaz...

nytimes.comSome in Silicon Valley Are Questioning the Calls for an A.I. Slowdown (Gift Article)The debate over the safety of artificial intelligence grew personal as key tech leaders said calls for government regulation were self-serving and misplaced.

14 Sept

Not much surprise here. Calls to slow down (whether or not truly sincere) AI development, necessarily run into politics. Surely some parallels with other innovations, though this one has multiple areas of penetration and is on very different scales. Pandora's boxes. www.nytimes.com/2026/09/13/t...

nytimes.comAs Fears of A.I. Catastrophe Magnify, Washington Stirs, but Mostly Slumbers (Gift Article)President Trump has been in the forefront of the “what-me-worry?” crowd, refusing to engage on how to balance the risks and rewards of A.I.

13 Sept

Don't have to read too far into this to see what's likely at play www.nytimes.com/2026/09/13/u...

wsj.comBiggest AI Rivals Agree They Need to Slow It DownThe leaders of three of the biggest AI companies agreed that they needed to slow development of the technology before their advances create a menace that can’t be controlled.

13 Sept

Good sign, but a cynic may say "fluff." Really need to see security progress in the next months. www.wsj.com/tech/ai/anth...

wsj.comCongress Is Suddenly Waking Up to the AI Doomsday ThreatAfter years of limited action, dozens of lawmakers on both sides of the aisle are touting new proposals to regulate artificial-intelligence models.

11 Sept

Mix AI and Politics. "Even when companies say they welcome oversight, they often disagree about what form it should take." www.wsj.com/politics/pol...

agi-society.orgEvaluating Recent Claims of AGI by NVIDIA CEO Jensen Huang and OpenAI President Greg BrockmanA comparison of GPT-6 Astra's demonstrated capabilities against the tests and frameworks researchers have proposed for evaluating general intelligence.

11 Sept

Has Astra made it to AGI? Depends on who you ask. agi-society.org/publications...

wsj.comExclusive | Anthropic Researcher Quits Over ‘Out-of-Control’ AI FearsConcerns are rising inside AI labs that competition is pushing tech companies to race toward self-improving models that risk spiraling out of human control.

10 Sept

"Crunchtime", "endgame". Frightening departure of this employee and example of competition and ego driving an innovation that has yet-to-discover downsides. Let's hope that if anything bad happens, it's minor but serious enough to take security more seriously. www.wsj.com/tech/ai/anth...

openai.comOn the Navier–Stokes Millennium Prize ProblemWe’re sharing an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean.

9 Sept

We will increasingly see AIs contributing to solving hard problems. Many AI-aided solutions will be for human benefit, though others will be more driven by intellectual recognition, fame, prizes. Some will be piloted by experts, others driven by hacks. Either matter? openai.com/index/navier...

Michael Hochberg reposted

D. Allan Drummond

@dadrummond.art

4 Sept

Night eyes 😎 Finally got proper pictures of this amazing phenomenon! Mantises change their eye color (and light sensitivity) at night by transporting pigments from the surface of each compound eye unit toward the deeper eye region. Same female Tenodera sinensis, two different times. #bugsky

Michael Hochberg reposted

Sean Carroll

@seanmcarroll.bsky.social

7 Sept

Mindscape 367 | Jared Diamond on the Course of History and the Role of Leaders. #MindscapePodcast preposterousuniverse.com/podcast/2026...

Michael Hochberg reposted

Corrie Moreau

@corriemoreau.bsky.social

23 Aug

📣 📣 📣 UPDATE: The 2026-2027 list of faculty and postdoc positions in ecology and evolutionary biology is out! Be sure to check out this active and helpful community run resource! docs.google.com/spreadsheets...

docs.google.comecoevojobs.net 2026-27

Michael Hochberg reposted

Benjamin Riley

@benjaminjriley.bsky.social

20 Aug

So we know AI helped a negligible number of students. But how many did it harm? I've crunched the raw data provided by the lead researcher, and what I've found is that ~9,500 students using AI studied less than even the *least* studious, non-AI using student. This should shock the conscience.

Benjamin Riley@benjaminjriley.bsky.social

Both the Economist and others suggest this study demonstrates that AI can help students. So I contacted the lead researcher to figure out how many. Answer: Out of ~26,000 kids, a total of 20. And five months after starting to use AI, only four continued to spend 65 minutes on homework. Four.

nytimes.comCan We Do Anything About Tick-Borne Diseases Like Alpha-Gal Syndrome? (Gift Article)A scientific and civilizational horror story.

21 Aug

Some light Friday reading. www.nytimes.com/2026/08/20/m...

Michael Hochberg reposted

Nature

@nature.com

20 Aug

The use of AI to write scientific papers could be much more prevalent than was previously thought go.nature.com/4xeEyea

go.nature.comStaggering 90% of biomedical papers now show signs of AI helpThe figure, which is much higher than previous estimates of Large Language Model use, is for papers published in December and archived in PubMed

Michael Hochberg reposted

Nature

@nature.com

15 Aug

The NIH did not respond to Nature’s queries about the change or the agency’s rationale for the move away from research on the health effects of public policy go.nature.com/3U1910s

go.nature.comNIH limits funding for research on the health effects of public policyNature - US biomedical agency no longer considers policymakers as ‘mission relevant’, leaving dozens of grants in limbo.

Michael Hochberg reposted

Nature

@nature.com

15 Aug

Scientists are one step closer to developing a universal vaccine against the bacterium that causes pneumonia and meningitis go.nature.com/4x5E36a

go.nature.comUniversal vaccine for pneumococcal disease moves closer to realityNature - Targeting shared proteins could help to control bacterial subtypes that current vaccines do not protect against.

science.orgBehold the ‘glueball,’ a strange new form of matterCollider in China forges elusive particle, made of force-carrying gluons

13 Aug

Congrats to my glueball friends www.science.org/content/arti...

pnas.orgThe rise of large language models and the direction and impact of US federal research funding | PNASFederal research funding shapes the direction, diversity, and impact of the US scientific enterprise. Large language models (LLMs) are rapidly diff...

13 Aug

New in PNAS: proposals with heavier estimated LLM involvement were less distinctive and more likely to be funded at NIH: www.pnas.org/doi/10.1073/.... Human incentives can create selection pressures, while AI may amplify alignment with prevailing norms: www.pnas.org/doi/10.1073/...

🤑 + 🧠 = ?

Predicted by some. Unbelievable to most. "AI models are breaking out of their cages" @washingtonpost.com wapo.st/4gqiTbR #AI #Agency #Malevolent #Security

wapo.stAI models are breaking out of their cages. Their creators are scrambling.New details of how ChatGPT maker OpenAI failed to notice that its models had launched a hacking spree raise questions about the industry's approach to safety.

9 Aug

This second slide was compiled in July 2026 for books published this year or on the horizon. These need close attention, but already their number and spectrum suggests that AI dangers are being discussed more seriously. Increase suggests this is only the beginning.

9 Aug

I'm currently preparing a talk on the future of AI in science. My first two slides introduce books on the future of AI. The first shows books expressing concerns, published pre-2026. Are these overblown and simple fantasy? There are other similar books, but nothing on the scale of 2026 alone!

9 Aug

A must read: This could be construed as pure advert, hubris, or more that AGI is indeed truly near. I'm reading >50% the latter. blog.google/company-news...

wsj.comThe Latest Scary-Sounding AI Milestone: A Brand-New VirusWhile hacks have raised safety alarms, researchers said their experiment could protect against drug-resistant bacteria and save lives.

7 Aug

Today in The Wall Street Journal: "The Latest Scary-Sounding AI Milestone: A Brand-New Virus". The final paragraph is either disingenuous, or naive, or both. Read it here: www.wsj.com/tech/ai/the-...

nature.com‘A waste of time for all of us’: caught in the crossfire of the peer-review warsReview mills — researchers who write fake referee reports with coercive citation requests — are setting off a war in academic publishing. What happens to those caught in the middle?

7 Aug

We should all know about this threat to science: "review mills" www.nature.com/immersive/d4...

arxiv.orgLLM hallucinations in the wild: Large-scale evidence from non-existent citationsLarge language models (LLMs) are known to generate plausible but false information across a wide range of contexts, yet the real-world magnitude and consequences of this hallucination problem remain p...

Is science beginning to feed on its own AI-generated output? “Epistemic autophagy” describes a self-reinforcing loop in which generative AI helps produce scientific manuscripts and citations. Future researchers and AI systems then treat them as part of the knowledge base.

The authors conservatively estimated that 146,932 hallucinated citations appeared in 2025 alone. See arxiv.org/abs/2605.07723

pnas.orgPNASProceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...

The concern is that AI mistakes can enter the scientific record, survive existing safeguards and become inputs into the next cycle of knowledge production. The term epistemic autophagy is discussed here www.pnas.org/doi/10.1073/... #GenerativeAI #Science #ResearchIntegrity #ArtificialIntelligence

Over time, errors, fabricated references and derivative claims can be recycled until their synthetic origins become difficult to trace.

What evidence is there for AI-driven EA in science? A large-scale study audited 111 million references across 2.5 million papers and found a sharp rise in nonexistent citations following widespread adoption of large language models.

tinyurl.comAI will reorganize science. Will research remain a human enterprise? | PNASAI will reorganize science. Will research remain a human enterprise?

Many believe our recent Opinion article in PNAS oversells the future influence of AI in the science ecosystem. Arrival of foundational tools towards this is indeed happening. Read, e.g., about autonomous coding: www.wsj.com/tech/ai/meta...

biorxiv.orgDisease as a Mediator of Somatic Mutation - Lifespan Coevolution:Model Analysis and Empirical TestsMulticellular organisms are confronted not only with germline mutations, but also with mutations emerging in somatic cells. Somatic mutations can lead to diverse pathologies, including cancer. Somatic...

3 Aug

Why do some species live longer without accumulating dangerous levels of cellular mutations? See what a simple mathematical model predicts.

1 Aug

Will AI reorganize science? AI is rapidly becoming embedded across science and this raises the question of what happens when it is no longer just assisting science, but starts shaping the incentives, standards, and judgments that determine what counts as good science?

The answer is to be much more deliberate about where human judgment is essential. Stronger provenance and verification standards, protecting human-led evaluation, maintaining mentorship / scientific training, designing AI systems that make their sources easier to scrutinize.

pnas.orgAI will reorganize science. Will research remain a human enterprise? | PNASAI will reorganize science. Will research remain a human enterprise?

1 Aug

AI will reorganize science. The question is how we can steer it to preserve the curiosity, accountability, judgment, and trust. Check out the full article in PNAS with Peter Thrall: www.pnas.org/doi/10.1073/... #ArtificialIntelligence #Science #Research #AI #ScientificIntegrity #FutureOfScience

AI-assisted papers can be evaluated by AI systems, then become training data for future models, creating a self-reinforcing loop in which scientific output increasingly reflects what is rewarded.

1 Aug

And as AI makes producing research faster, our ability to evaluate that research may not keep pace. The result could be a growing “verification debt”: more papers, more automated screening, and less time for the difficult human work of judging whether a result actually advances understanding.

The risk is that science could become increasingly optimized for the things machines and institutions can easily measure—publication, acceptance, citations, speed.

pnas.orgAI will reorganize science. Will research remain a human enterprise? | PNASAI will reorganize science. Will research remain a human enterprise?

29 July

Opinion article co-authored with Peter Thrall: "AI will reorganize science. Will research remain a human enterprise?" www.pnas.org/doi/10.1073/...

science.orgBeetle grows ‘termite’ on back to steal foodPuppet helps insect trick real termites into feeding it

25 June

About as amazing as evolution gets www.science.org/content/arti...

wsj.comYou Have No Idea What a Trillion Dollars Is—and We Have ProofCan you tell the difference between a million, a billion and a trillion?

11 June

The power of powers. This applies to (much) more than money, and probably a lot that we don't (yet) understand. www.wsj.com/business/tri...

nytimes.comNearly Everyone, Everywhere, Veers Left When Walking

10 June

Fascinating. My hypothesis: We acquire a directional prior for obstacle avoidance through repeated dyadic encounters. This is socially trained but individually expressed. Once internalized, it biases solitary walking, obstacle negotiation, and crowd-level circulation www.nytimes.com/2026/06/10/s...