iiinova — where your digital self keeps evolving

Reposted by Max Little

Georgia Tomova @georgiatomova.bsky.social · 13/05/2026

Why we should rethink causal mediation, and what to do instead? Come to hear the answer from Vanessa Didelez at the next CIIG seminar! The seminar will be hybrid. If you are in London, come join us in person at UCL! Otherwise, you can join on Zoom as usual. Registration links in comment below.

Vanessa Didelez, 8th June 2026 3 to 4.30pm. Why we should rethink causal mediation and what to do instead

44615

Reposted by Max Little

Peter Tennant @pwgtennant.bsky.social · 12/05/2026

Does engaging with the arts help to slow the aging process? According to this new paper from Fancourt et al the answer is yes! Alas, the paper is actually unsuitable for such claims. Let’s examine by discussing how you'd establish whether arts engagement actually caused future health! 1/18 🧵

academic.oup.com

Does leisure activity matter for epigenetic ageing? Analyses of arts engagement and physical activity in the UK Household Longitudinal Study

AbstractBackground and Objectives. Over the past decade, ageing clocks have become widely adopted as important tools for understanding biological ageing an

813160

Reposted by Max Little

Peter Tennant @pwgtennant.bsky.social · 25/02/2026

This recent RCT of an "AI stethoscope" claims the technology "shows promise" for diagnosing cardiovascular conditions. It does not. It is a textbook example of the risks of conducting unprincipled 'per protocol analyses'. Once again, peer review at a major medical journal has failed. 🧵 1/

8436186

Reposted by Max Little

Darren Dahly @statsepi.bsky.social · 02/09/2025

Depends on how you spin it I guess (screenshots from 2 different articles). People working at universities are pushed so incredibly hard to ensure that every study is a breakthrough that they just...lie. All the time. Probably without even realizing it. www.standard.co.uk/news/tech/im...

“Our study shows that three heart conditions can now be identified in one sitting,” Dr Nicholas Peters, a cardiologist and the trial’s senior investigator from Imperial College London, said in a statement.

Notably, though, about two-thirds of patients who were flagged by the AI stethoscope as potentially having heart failure did not actually have the condition, the doctors found after ordering blood tests or heart scans.
Unnecessary anxiety?

The researchers acknowledged that the stethoscope’s apparent oversensitivity could lead to unnecessary anxiety and tests for some patients, but noted that the tool also detects genuine heart problems that might otherwise have been overlooked.

They suggested that the AI stethoscope only be used for patients with suspected heart problems, not for routine health checks.

However, it’s unclear whether doctors find the tool useful. A year after being given the AI stethoscopes, 70 per cent of GP offices stopped using them regularly, the trial found.

The AI stethoscope was developed by Eko Health, a California-based health technology company.

“So it is incredible that a smart stethoscope can be used for a 15-second examination, and then AI can quickly deliver a test result indicating whether someone has heart failure, atrial fibrillation or heart valve disease.”

The findings of the trial, known as Tricorder, are being presented at the European Society of Cardiology (ESC) Congress in Madrid.

Researchers are now planning to roll out the stethoscopes to GP practices in Wales, south London and Sussex.

Professor Mike Lewis, scientific director for innovation at the NIHR, which supported the study, said: “This tool could be a real game-changer for patients, bringing innovation directly into the hands of GPs.

“The AI stethoscope gives local clinicians the ability to spot problems earlier, diagnose patients in the community, and address some of the big killers in society.”

Professor Nicholas Peters, senior investigator from Imperial College London and consultant cardiologist at Imperial College Healthcare NHS Trust, added: “Our study shows that three heart conditions can now be identified in one sitting.

“Importantly, this technology is already available to some patients and being widely used in GP surgeries.”

4308

Reposted by Max Little

Rodney Brooks @rodneyabrooks.bsky.social · 01/01/2025

Every Jan 1 I post a scorecard on predictions I made, with dates, on Jan 1, 2018 on cars (self-driving), robots, AI, & ML, and on human spaceflight. Besides telling which turned out right and which wrong in the last year I also talk a lot of smack about these topics. rodneybrooks.com/predictions-...

rodneybrooks.com

Predictions Scorecard, 2025 January 01 – Rodney Brooks

711045

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 21/12/2024

𝗢𝟯 𝘄𝗮𝘀 𝘁𝗿𝗮𝗶𝗻𝗲𝗱 𝗼𝗻 𝟳𝟱% 𝗼𝗳 𝘁𝗵𝗲 𝗽𝘂𝗯𝗹𝗶𝗰 𝘀𝗲𝘁 𝗳𝗼𝗿 𝗔𝗥𝗖-𝗔𝗚𝗜. OpenAI did not disclose this in the video. Sam said they didn’t target the test. Never trust a staged demo. Never trust a product you haven’t tried. Never trust OpenAI.

2438655

Max Little @maxal.bsky.social · 21/12/2024

Likewise, a simple adversarial strategy beats "superhuman" Go-playing algorithms: goattack.far.ai It's wise to remember that there is no scientific consensus on what "intelligence", actually is.

goattack.far.ai

Adversarial policies in Go

051

Max Little @maxal.bsky.social · 21/12/2024

Just for those who don't know: the vast majority of open problems in maths, are not numerical in nature.

010

Reposted by Max Little

Timothy Gowers @wtgowers.bsky.social · 21/12/2024

The questions have numerical answers, so it is easy to check whether it gets them right.

111

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 21/12/2024

How many times do we have to see this same movie, where an AI beats some benchmark and influencers gleefully shout “It’s So Over” without even trying out the AI and then on careful inspection the AI turns out to not be robust or reliable? Thousands? (It’s already been hundreds.)

7749

Reposted by Max Little

Timothy Gowers @wtgowers.bsky.social · 20/12/2024

It seems that OpenAI's latest model, o3, can solve 25% of problems on a database called FrontierMath, created by EpochAI, where previous LLMs could only solve 2%. On Twitter I am quoted as saying, "Getting even one question right would be well beyond what we can do now, let alone saturating them."

8868

Reposted by Max Little

Stephanie M. Lee @stephaniemlee.bsky.social · 18/12/2024

It's widely agreed that scholars are supposed to say when they use ChatGPT. Yet phrases like "I am an AI language model"—with no disclosure—are popping up in papers. I wrote about how journals seemingly aren't enforcing their AI policies, according to a new study: www.chronicle.com/article/scho...

chronicle.com

Scholars Are Supposed to Say When They Use AI. Do They?

Journals have policies about disclosing ChatGPT writing, but enforcing them is another matter, according to a new study.

15221

Reposted by Max Little

Rodney Brooks @rodneyabrooks.bsky.social · 18/12/2024

This seems like a pretty balanced commentary. They certainly get this right: "connection between capability improvements & AI’s social or economic impacts is extremely weak. The bottlenecks for impact are the pace of product development and the rate of adoption" www.aisnakeoil.com/p/is-ai-prog...

aisnakeoil.com

Is AI progress slowing down?

Making sense of recent technology trends and claims

1184

Max Little @maxal.bsky.social · 17/12/2024

Good reporting here, but sadly, these tragedies were predictable. Those of us who actually work on machine learning know that deep-learning based computer vision simply isn't reliable enough for safety-critical applications such as self-driving cars. @garymarcus.bsky.social @filippie509.bsky.social

091

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 14/12/2024

The late Suchir Balaji’s blog post on AI, copyright and fair use, reposted in his memory. suchir.net/fair_use.html

suchir.net

When does generative AI qualify for fair use?

412436

Reposted by Max Little

Maths Facts @mathsfacts.bsky.social · 14/12/2024

The bootstrap can be used to generate a new random sample from an existing random sample. Its validity can be guaranteed by the Glivenko-Cantelli theorem, which demonstrates how the empirical cumulative distribution (CDF, top panel), converges on the CDF of the sample (bottom panel).

001

Reposted by Max Little

Maths Facts @mathsfacts.bsky.social · 14/12/2024

For an increasing function 𝑓:ℝ→ℝ, max(𝑓(𝑎),𝑓(𝑏))=𝑓(max(𝑎,𝑏)). An important special case is 𝑓(𝑥)=𝑥+𝑐, for which we obtain max(𝑎+𝑐,𝑏+𝑐)=𝑐+max(𝑎,𝑏).

001

Reposted by Max Little

Filip Piekniewski @filippie509.bsky.social · 14/12/2024

Since 2016 Waymo raised ~$25B, so they burn ~$3B/y or little over 8mln/day. With ~700 cars, assuming they operate each car every day, it costs them over 11k dollars to operate each of their cars per day. $11k PER DAY per CAR. If you don't find this ridiculous IDK what else to say.

283

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 14/12/2024

Suchir Balaji was a good young man. I spoke to him six weeks ago. He had left OpenAI and wanted to make the world a better place. This is tragic.

816146

Max Little @maxal.bsky.social · 13/12/2024

Very proud of the Birmingham HDRUK PhDs!

010

Reposted by Max Little

cwcyau.bsky.social @cwcyau.bsky.social · 13/12/2024

Health Data Research UK PhD meet! Work from Ant Lee and Jianqiao Mao (latter with @maxal.bsky.social)

021

Reposted by Max Little

Filip Piekniewski @filippie509.bsky.social · 10/12/2024

As first predicted some 10 years ago that is how "self driving cars" will end - as glorified driver assistance features. The graveyard of autonomous vehicle efforts is pretty crowded already with pretty much only Waymo remaining, until life support from Google mothership ends.

192

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 09/12/2024

What if all the hype just didn’t turn out to be true? Evidence of productivity gains is mixed - yet hypey takes continue to dominate in the media.

7429

Max Little @maxal.bsky.social · 02/12/2024

Not quite: AI got people excited about interpolation, it seems. Numerical analysts suddenly feel seen.

030

Max Little @maxal.bsky.social · 27/11/2024

Fully-funded PhD position available. If you are interested in machine learning for signal processing of biosignals, please do get in touch.

010

Max Little @maxal.bsky.social · 23/11/2024

The current incarnation of AI companies earnestly engaged in the Sisyphean task of labeling "all the things, in every circumstance".

000

Max Little @maxal.bsky.social · 10/11/2024

Nice, terse quote: "There is no principled solution to hallucinations in systems that traffic only in the statistics of language without explicit representation of facts and explicit tools to reason over those facts."

000

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 09/11/2024

Game over, I won. GPT is hitting a period of dimishing returns, just like I said it would. Analysis and breaking news here: open.substack.com/pub/garymarc...

open.substack.com

CONFIRMED: LLMs have indeed reached a point of diminishing returns

Science, sociology, and the likely financial collapse of the Generative AI bubble

76311

Max Little @maxal.bsky.social · 24/09/2024

Can we be confident that rigour-enhancing practices causally increase the replicability rate of scientific studies in social science? A high-profile claim of that specific cause-effect relationship was found to have insufficient rigour, now retracted. statmodeling.stat.columbia.edu/2024/09/24/w...

statmodeling.stat.columbia.edu

Well, today we find our heroes flying along smoothly… | Statistical Modeling, Causal Inference, and Social Science

020

Reposted by Max Little

Jessica Hullman @jessicahullman.bsky.social · 24/09/2024

In other science reform implosion news, the contested article including OSF & Data Colada authors on how preregistration & other rigor-enhanching practices led to high observed replicability (the one that couldn't produce its own preregistration) was just retracted www.nature.com/articles/s41...

nature.com

RETRACTED ARTICLE: High replicability of newly discovered social-behavioural findings is achievable - Nature Human Behaviour

Four labs discovered and replicated 16 novel findings with practices such as preregistration, large sample sizes and replication fidelity. Their findings suggest that with best practices, high replica...

613259

Reposted by Max Little

samuel mehr @mehr.nz · 14/09/2024

this is pretty amazing: @lucinauddin.bsky.social is taking on the giant, hugely profitable publishers of academic journals, on grounds of antitrust If you'd like to join the case as a plaintiff you can sign up at www.lieffcabraser.com/antitrust/ac... summary of the case:

The Publisher Defendants’ Scheme has three primary components. First, the Publisher Defendants agreed to not compensate scholars for their labor, in particular not to pay for their peer review services (the “Unpaid Peer Review Rule”). In other words, the Publisher Defendants agreed to fix the price of peer review services at zero. The Publisher Defendants also agreed to coerce scholars into providing their labor for nothing by expressly linking their unpaid labor with their ability to get their manuscripts published in the Publisher Defendants’ journals. In the “publish or perish” world of academia, the Publisher Defendants essentially agreed to hold the careers of scholars hostage so that the Publisher Defendants could force them to provide their valuable labor for free.

Second, the Publisher Defendants agreed not to compete with each other for manuscripts by requiring scholars to submit their manuscripts to only one journal at a time (the “Single Submission Rule”). The Single Submission Rule substantially reduces competition among the Publisher Defendants, substantially decreasing incentives to review manuscripts promptly and publish meritorious research quickly. The Single Submission Rule also robs scholars of negotiating leverage they otherwise would have had if more than one journal offered to publish their manuscripts. Thus, the Publisher Defendants know that if they offer to publish a manuscript, the submitting scholar has no viable alternative and the Publisher Defendant can then dictate the terms of publication.

Third, the Publisher Defendants agreed to prohibit scholars from freely sharing the scientific advancements described in submitted manuscripts while those manuscripts are under peer review, a process that often takes over a year (the “Gag Rule”). From the moment scholars submit manuscripts for publication, the Publisher Defendants behave as though the scientific advancements set forth in the manuscripts are their property, to be shared only if the Publisher Defendants grant permission. Moreover, when the Publisher Defendants select manuscripts for publication, the Publisher Defendants will often require scholars to sign away all intellectual property rights, in exchange for nothing. The manuscripts then become the actual property of the Publisher Defendants, and the Publisher Defendants charge the maximum the market will bear for access to that scientific knowledge.

1116899

Reposted by Max Little

David Ho @davidho.bsky.social · 14/09/2024

In case you don’t believe the same equations describe motion in the atmosphere and the ocean.

321074331

Reposted by Max Little

Rodney Brooks @rodneyabrooks.bsky.social · 13/09/2024

"By creating the illusion of complete autonomy, companies can fuel interest in their technology and raise the billions of dollars they need to build a viable robot taxi service." Waymo & Cruise recently admitted they use remote assist. Zoox is more up front about it. www.nytimes.com/2024/09/11/i...

nytimes.com

When Self-Driving Cars Don’t Actually Drive Themselves

An immersive article shows readers what a New York Times reporter has tracked for nearly a decade: Robot taxis still need human help.

1113

Reposted by Max Little

Blake Richards @tyrellturing.bsky.social · 12/09/2024

1/ Here's a critical problem that the #neuroai field is going to have to contend with: Increasingly, it looks like neural networks converge on the same representational structures - regardless of their specific losses and architectures - as long as they're big and trained on real world data. 🧠📈 🧪

512635

Reposted by Max Little

Gary Marcus @garymarcus.bsky.social · 05/09/2024

“AI worse than humans in every way at summarizing information … the technology might actually make more work for people, not less”

2184