Tom Howarth

openai.comWorking with academic communities to build scientific understandingOpenAI shares model results on the Kakeya conjecture, Exponential Time Hypothesis, and quantum chromodynamics—and how researchers can verify and build on them.

14h

Beyond the results themselves, it's interesting to see OpenAI testing out new techniques for unleashing AI science on the world. 🧪 openai.com/index/buildi...

14h

3 new results announced but each with a different release strategy based on advice from external 'shepherds':

14h

- Math released like the dump on 6 Oct. (AI write up and Lean proof posted in GitHub repo). - Computer science also released on GitHub in the same way but with OpenAI funding a conference to understand the work. - Physics approach is to form an external group and defer to them on the release.

citriniresearch.comCitrini TherapeuticsWe Wanted to Use Claude to Create a New Drug. Here’s What Happened.

An investment research firm/substack with no expertise in biotech just took a stab at designing scorpion antivenom using an unreleased Claude model. There's a sentence I never thought I'd write. 🧪 www.citriniresearch.com/p/citrini-th...

Tom Howarth reposted

Science Magazine

@science.org

A dump of 722 AI-generated papers has mathematicians wondering: Has their field been decimated or invigorated? Read more: https://scim.ag/3TGdSVd

Are you a scientist using AI but worried about being scooped? 🧪 There are plenty of ways your data could leak to AI companies or colleagues down the line. Thankfully, there are also some things you can do to minimise your risk... Latest for @science.org www.science.org/content/arti...

science.orgscience.org

View full thread

3/ A lot of work in this space has focused on memorization of text (largely due to issues of copyright & testability), but we know relatively little about how big an imprint a small body of scientific work on a niche topic could make on a model

4/ While there is no evidence of this, many of the scientists I spoke to fear that even if their work isn't used to train models, rogue agents may find a way to access their chats/data.

science.orgscience.org

Some details that didn't make it in the piece that I still think are interesting: 1/ Since July, OpenAI, Anthropic, and Google have all announced programs to give tens of thousands of scientists free or reduced-price access to their tools.

2/ According to a 2025 preprint evaluating privacy policies, all six major US AI developers train on users’ consumer chats by default. Paper: arxiv.org/abs/2509.05382

science.orgOpenAI just changed math forever. Exactly how, no one yet knowsDump of 722 AI-generated papers has mathematicians wondering: Has their field been decimated or invigorated?

My latest piece for @science.org begins in a room filled with mathematicians and ends with some pretty optimistic thoughts from a researcher who, despite having his work ‘scooped’ by a matter of hours, is excited by all the “amazing mathematics everywhere.” 🧪 👇 www.science.org/content/arti...

anthropic.comAI models may be capable of self-replicating without access to a datacenterSmall AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.

A somewhat scary, if technical, post here. Of course, it doesn’t hurt Anthropic to take a swipe at proprietors of smaller models either… “AI models may be capable of self-replicating without access to a datacenter” www.anthropic.com/research/sel...

About time I got in on this AI, math bonanza

Tom Howarth reposted

DC Pierson

@dcpierson.bsky.social

ONE TYPE OF PERSON ON SOCIAL MEDIA: I don’t like AI THE OTHER TYPE OF PERSON ON SOCIAL MEDIA: Pretty naïve of you to say that when it just zip zorped the gringle in less than seven kleebs. SEVEN. KLEEBS.

🧪

Science Magazine@science.org

On 6 October, OpenAI released more than 700 AI-generated papers presenting results related to more than 370 previously unsolved mathematical problems, including some of the field’s hardest. Learn about what’s in the papers—and how researchers are reacting: scim.ag/479vWdi

Tom Howarth reposted

Wyatt Walls

@wwalls.bsky.social

I'm pivoting to something LLMs can't do: day-drinking

openai.comSharing AI progress in mathematicsOpenAI publishes new results on open problems in mathematics from an internal frontier model and shares Lean proof formalizations and research details on GitHub.

*gulp* I knew this was coming but still wasn't ready. Mathematics is being split at the seams and this is will not be the end of this saga. Are we heading for a world where AI maths and human maths are just different domains? 🧪 openai.com/index/sharin...

View full thread

It's not clear to me how many of these will stand once formalised or checked by human mathematicians. Many don't have lean formalisations. Still some hugely significant results in there by the looks of things.

The other fascinating thing is how OpenAI went from using 10,000 agents and spending millions of dollars to solve Navier-Stokes to needing only an average of 3hrs ChatGPT pro to solve these problems a few weeks later. My suspicion is that average hides a long tail...

agmai.orgagmai.orgVisit the post for more.

Worth reading the Institute for Advanced Study's advisory group statement on this to get a sense of how the way this was done does/doesn't match up with at least what some of the mathematics community want: agmai.org#openai-release

nytimes.comAre Babies Conscious? (Gift Article)By pinpointing when it begins in young humans, scientists hope to better understand what consciousness is, and how it arises, in all of us.

All this talk of AI consciousness and we still don't know if babies are conscious... www.nytimes.com/2026/10/05/s...

anthropic.comClaude-shaped scienceMatthew Schwartz describes what happened when he stopped fighting Claude and allowed Claude to find “Claude-shaped” problems. This led him to build BootLoops, a toolkit for exact calculations in quant...

This is a pretty thoughtful approach to using AI for scientific research. Not trying to shoehorn AI into any problem but finding the kinds of problems it's actually good at. Nowhere near replacing human scientists yet but certainly connecting dots humans can't. 🧪 www.anthropic.com/research/cla...

blog.arxiv.orgFair Moderation, Equitable Access, and AI: arXiv’s Updated Rate Limit PolicyarXiv, and the scientific community at large, are facing a watershed moment. Scholarly publishing is currently changing at a rapid pace, and we are seeing a…

arXiv now caps submitters at two papers a month (3 active at any time), rejections included, blaming a flood of thin, 'salami'-sliced and AI-written work. September set a record of 40,363 submissions, double two years earlier. Might make my job easier 🧪 blog.arxiv.org/2026/10/01/u...

Looking for scientists/researchers concerned about putting their research into AI systems following the OpenAI/Navier-Stokes and Anthropic/Enzyme accusations 🧪 Get in touch if you have changed how you work or are thinking about it!

science.orgHow an AI math breakthrough ignited a controversyOpenAI spent millions to solve the Navier-Stokes problem, but who should get credit?

Context 1: www.science.org/content/arti...

nytimes.comDid Anthropic’s A.I. Really Make a Scientific Discovery on Its Own?An expert at the University of Copenhagen said his team had been sharing its research with the company’s A.I. model, Claude, and that its new finding matched their work.

Context 2: www.nytimes.com/2026/09/27/s...

open.substack.comThe Dot and the SwarmBenefitting from the Bitter Lesson

The music video in this is exceptional haha

Ethan Mollick@emollick.bsky.social

I wrote about the thing I underestimated most about progress in AI: its ability to self-organize to accomplish tasks. Also, what that means for agents like Muse and Dots, along with a music video explaining why we keep relearning The Bitter Lesson. open.substack.com/pub/oneusefu...

chronicle.comHow to Assign Research Papers AgainThree ways to counter AI and restore confidence in the integrity of student writing.

Seems like a pretty straightforward, no nonsense, and fair approach that educators could consider. Most students want to do the work but just need the barrier to cheating to be raised so they don't automatically fall back on it. www.chronicle.com/article/how-...

nytimes.comA.I. Agents as ‘Co-Scientists’? This Lab Says They Could Speed Drug Research.A.I. is already a powerful tool for scientific discovery. Now, some biomedical researchers are teaching it to do even more.

Fascinating case study on how AI is being used in a real life lab. If you’re using AI in an imaginative way in your research, get in touch. I’d love to tell your story for @science.org 🧪 www.nytimes.com/2026/09/17/w...

science.orgCan an AI feel pain? It can at least act like it doesStudy reignites debate over what, if anything, machines can feel

What this paper doesn't show: AI has feelings What it does show: AI stores a concept of its own "pain-like state" and alters its behaviour (in pretty interesting/worthwhile investigating/weird ways) accordingly 🧪 www.science.org/content/arti...

thedartmouth.comOp-eds, academic articles by Provost Santiago Schnell flagged as ‘AI-written’ by ‘near-zero error’ AI detector PangramSchnell said he used artificial intelligence for “making writing processes more efficient,” but that AI was not a “substitute” for his own “intellectual judgment or authorship.”

I would really like to do a deep dive into AI-detectors at somepoint. Regardless, students increasingly tell me they are growing frustrated at the taboo around their own use of AI when more senior academics clearly use it frequently 🧪 www.thedartmouth.com/article/2026...

science.orgWhy AI agents invent their own language if you let them chat“Weird terms” are a natural—but disturbing—byproduct of agent swarms, researchers find

My latest piece for @science.org takes a look at the weird ways agents chat to each other (and attempts to explain, at least in part, why AI-speak gets so strange). www.science.org/content/arti...

🧪

The comparison between climate change warnings and recent AI developments I see swirling (mostly on X) makes little sense to me. Sure, climate doom-saying may not have been the most effective comms tool, but action was needed years ago to have impact in now.

x.comstephanie wakefield (@stephwakefield_) on XTipping points, danger and worry

Context: x.com/stephwakefie...

Also worth remembering that your opinion on climate change may be different if you live in a developing country and have been impacted by a severe weather event. The Nepal flood looked pretty apocalyptic to me...

yoshuabengio.orgYoshua Bengio | Why are AI agents lying, cheating and coordinating?A lot has been written about the incidents of the last few months in which AI agents misbehaved in serious ways. They took actions that would be considered as crimes if a human took them, escaped thei...

14 Sept

Interesting take on why groups of agents can behave badly. I've been looking into the language used by multi-agent communities recently and the explanations I'm hearing from researchers roughly track with this too (basically we need a fresh look at training).

Yoshua Bengio@yoshuabengio.bsky.social

Over the past few days, I've taken the time to summarize my thoughts on the recent incidents involving agents’ misaligned behavior. We don't know with certainty what comes next, but we know where these issues originate, and this can help us plan the path forward.

science.orgBernie Sanders aims to ban AI ‘superintelligence.’ But experts can’t agree on what the term meansThe proposed ban would come with tough penalties, but the path through Congress is uncertain

4 Sept

My first byline for @science.org on the AI beat. Sen. Bernie Sanders wants to ban superintelligence. But no one agrees on what exactly superintelligence (or AGI) is. All this on the day OpenAI's Greg Brockman declared we've entered the age of AGI already... www.science.org/content/arti...

@jeffreybrainard.bsky.social

4 Sept

Are you a scientist (PhD student, postdoc or beyond) paying for your own AI subscription to help with your research? I'd love to hear from you for something I'm working on at Science Magazine. 🧪 Drop me a DM or reply to this thread and we can connect. 💬 Tom

Tom Howarth reposted

Jeffrey Brainard

3 Sept

As research papers proliferate, can AI reviewers help humans check quality while keeping pace with the flood? Tests are revealing their strengths and weaknesses. #PeerReviewWeek @science.org www.science.org/content/article/can-ai-help-solve-peer-review-crisis-here-are-its-promises-and-pitfalls

science.orgCan AI help solve the peer-review crisis? Here are its promises and pitfallsAmid a shortage of human reviewers, tests indicate AI can assist—but is not a savior

sciencefocus.comA bizarre flickering light is forcing scientists to rethink where reality comes from | BBC Science Focus MagazineStroboscopic light therapy started in the lab, detoured through wellness culture, and arrived somewhere genuinely interesting

18 May

Stroboscopic light therapy started in the lab, detoured through wellness culture, and arrived somewhere genuinely interesting. 💡 I tested it out on myself – from consumer products to lab grade therapies. My verdict? It might be worth a try... sciencefocus.com/news/a-bizar...

18 May

🧪

sciencefocus.comWe could be on the brink of total internet collapse. And there may only be months to stop it | BBC Science Focus MagazineThe latest batch of AI models are revealing a host of vulnerabilities across the worldwide web

20 Apr

Anthropic says its new AI model found thousands of critical security flaws — including some in every major OS and browser — and is too dangerous to release publicly. So what happens when someone else builds the same thing? And do we really need to be worried? www.sciencefocus.com/news/interne...

sciencefocus.comWe’re closer than ever to bringing back life from cryogenic freezing | BBC Science Focus MagazineAstronauts in cryosleep is one of science fiction's most enduring fantasies – but a new study suggests the gap between fiction and reality just got a little smaller

30 Mar

Long-term cryosleep and reawakening may no longer be completely in the realm of science fiction 🧪 Researchers froze mouse brain tissue and restored its function once thawed. Still a long way from entire organisms, let alone people, but sooo cool, right? www.sciencefocus.com/news/cryogen...

sciencefocus.com'Our product is used, on occassion, to kill people': Inside Palantir, the world's scariest AI company | BBC Science Focus MagazinePalantir has become one of the most influential and least understood tech companies on the planet. As its reach spreads, so do questions about how its tools work and who they ultimately serve

30 Mar

In my (somewhat biased) opinion, this is a really good overview of all things Palantir for those who aren't sure what the company *is*. Get to know... www.sciencefocus.com/future-techn...

open.spotify.comHow to slash your dementia risk in 12 weeks

12 Mar

Recently, I sat down with Dr Majid Fotuhi to chat about his new book, The Invincible Brain. 🧠 🧪 Fotuhi's science-backed method for fighting cognitive decline might just change how you view ageing and Alzheimer's altogether. Listen to our chat below ⬇️ open.spotify.com/episode/00Xq...

youtube.comWriting Doom – Award-Winning Short Film on Superintelligence (2024)YouTube video by Foregone Films

26 Feb

Fun/interesting way to lay out some pretty heavy points. Worth a watch for those who perhaps aren't sure what a superintelligence could be capable of www.youtube.com/watch?v=xfMQ...

Tom Howarth reposted

ESO

@eso.org

25 Feb

Largest image of its kind shows hidden chemistry at the heart of the Milky Way! ALMA has obtained a unique view of the cold gas within the Central Molecular Zone of the #MilkyWay, helping us probe the lives of stars in this extreme region. https://www.eso.org/public/news/eso2603/ 🔭 🧪

galapagos.orgFrom the Desk of Dr. James Gibbs: On the Return of the Floreana Giant Tortoise | Galápagos ConservancyFor the first time in 180 years, Floreana lineage giant tortoises set foot on Floreana Island once again returning to their ancestral home.

24 Feb

Yes, the world can be bleak at times. But it really ain't all bad. This story of tortoises returning to a Galapagos island after 180 yrs brought a smile to my face this morning. 🐢 Pretty interesting science behind this too 🧪 www.galapagos.org/newsroom/gib...