What do you do after you’re done jumping the shark? Whatever it is, Nature Careers is all in.

Scientists and scholars in AI and its social impacts call on von der Leyen to retract #AIHype statement. @olivia.science @abeba.bsky.social @irisvanrooij.bsky.social @alexhanna.bsky.social @rocher.lc @danmcquillan.bsky.social @robin.berjon.com & many others have signed www.iccl.ie/press-releas...

Russia’s success in poisoning LLMs with lies, and the effects it has on both AI and politics, reflects Russia’s much deeper understanding of how societies operates than much of Silicon Valley - and how important social sciences are in understanding and waging information warfare

arXiv will no longer accept review articles and position papers unless they have been accepted at a journal or a conference and complete successful peer review. This is due to being overwhelmed by a hundreds of AI generated papers a month. Yet another open submission process killed by LLMs.

Pretends to be shocked www.bbc.co.uk/mediacentre/...

The viral "Definition of AGI" paper tells you to read fake references which do not exist! Proof: different articles present at the specified journal/volume/page number, and their titles exist nowhere on any searchable repository. Take this as a warning to not use LMs to generate your references!

imho — anyone who equates a human with an app or a machine today is just dehumanizing people and stripping people of their (dwindling, already eroding, not well respected) rights.

Keynote at #COLM2025: Nicholas Carlini from Anthropic "Are language models worth it?" Explains that the prior decade of his work on adversarial images, while it taught us a lot, isn't very applied; it's unlikely anyone is actually altering images of cats in scary ways.

I said a thing :).

OpenAI's rapid rush into education has been achieved by habituating users through training programs, institutional lock-ins, strategic marketing partnerships, and third party integrations that together are helping it become infrastructural to teaching and learning. It's going to be hard to get out.

"We are told that AI is inevitable, that we must adapt or be left behind. But universities are not tech companies. Our role is to foster critical thinking, not to follow industry trends uncritically." www.ru.nl/en/research/...

After 2 years in press, it's published! "Talkin' 'Bout AI Generation: Copyright and the Generative-AI Supply Chain," is out in the 72nd volume of the Journal of the Copyright Society copyrightsociety.org/journal-entr... written with @katherinelee.bsky.social & @jtlg.bsky.social (2023)

In a new paper, I try to resolve the counterintuitive evidence of Meehl’s “clinical vs statistical prediction” problems: Statistics only wins because the game is rigged.

If kids’ schools trained them for work based on what everyone thought the hot new technology was going to be, both my kids would have spent the past several years learning about the blockchain. This is why schools don’t attempt to do workplace training: life is pretty long.

There are many similarities with the AI discourse now and the early web. There too were some utopian visions of the future like in this famous Barlow's declaration. 20 years after, the www is a great technology, yet here we are now with all the dopamine-driven design and social polarization.

From claims of "Ph.D. level" intelligence about generative models and calls for knowledge work to be replaced by these models, to Musk's claim that there is no research only engineering, to — all are manifestations of anti-intelluctualism.

This is absolutely what I expect as well.

Excellent post @aarontay.bsky.social on how there are so many LLMs in modern library search pipelines, each applying content moderation filters (or subject to cloud providers'), leading to unexpected & unwanted censorship on topics like Gaza and race rioting. aarontay.substack.com/p/the-ai-pow...

Rachel L. Draelos, Samina Afreen, Barbara Blasko, Tiffany Brazile, Natasha Chase, Dimple Desai, Jessica Evert, Heather L. Gardner, Lauren Herrmann, Aswathy Vaikom House, ... Large language models provide unsafe answers to patient-posed medical questions https://arxiv.org/abs/2507.18905

firefox even does this for you

The whole "Grok can be re-tuned at Elon's will to spout off like Hitler" thing sorta punctures the "LLM chatbots are AGI" discourse just a little bit, doesn't it.

New preprint on the most precise as of yet mapping between differential privacy and common operational notions of privacy risk used in practice:

Some ✨ personal news ✨: I'm starting my independent consultancy, focused on helping organizations do good things with privacy-enhancing technology 🎉 It's called Hiding Nemo, and you can read all about it on our website ➡️ https://hiding-nemo.com 🪸

Ah, the AI for Good Summit, complete with a cybertruck display.

ExplainableAI has long frustrated me by lacking a clear theory of what an explanation should do. Improve use of a model for what? How? Given a task what's max effect explanation could have? It's complicated bc most methods are functions of features & prediction but not true state being predicted 1/

I love using m-dashes and it's quite sad that people often associate them with generated text now. They're a great writing tool.

Original Article by S. Shekar et al.: People Overtrust AI-Generated Medical Advice despite Low Accuracy nejm.ai/43xKDG0 #AI #MedSky #MLSky

This is incredible. The deployed AI models have finally managed the impossible: becoming utterly worse at image recognition than anything from the past 15 years.

You would think it can't get worse than Venmo's public feed of payments.

If your business model doesn’t work without breaking the law, then you’re not in business. You’re in organized crime.

Denied a loan, an interview, or an insurance claim by machine learning models? You may be entitled to a list of reasons. In our latest w @anniewernerfelt.bsky.social @berkustun.bsky.social @friedler.net, we show how existing explanation frameworks fail and present an alternative for recourse

1. LLM-generated code tries to run code from online software packages. Which is normal but 2. The packages don’t exist. Which would normally cause an error but 3. Nefarious people have made malware under the package names that LLMs make up most often. So 4. Now the LLM code points to malware.

The Computing Research Association (@cra-hq.bsky.social) has a new trustworthy AI research fellowship. Intended for folks 1-3 years out of their PhD and with interdisciplinary experience in social sciences. cra.org/cra-trustwor...

Science is not collection of findings. Progress happens through theories.As we move from findings to theories things r less amenable to automation. Proliferation of scientific findings based on AI hasn't accelerated—& might even have inhibited—higher levels of progress www.nature.com/articles/d41...

Would you present your next NeurIPS paper in Europe instead of traveling to San Diego (US) if this was an option? Søren Hauberg (DTU) and I would love to hear the answer through this poll: (1/6)

I had the pleasure to give an interview about synthetic data www.wired.com/story/nvidia... by @laurengoode.bsky.social for Wired. I'd like to add a caveat: the promises of synthetic data to solve privacy and data scarcity issues are not yet backed up by scientific evidence.

despite popularised beliefs, LLMs are not fit for medical applications. SoTA models produce "non-trivial levels of hallucinations" even w inference techniques like CoT & search augmented generation: arxiv.org/pdf/2503.05777 of surveyed clinicians, 53% use LLMs daily & 91% encountered hallucinations

91% of medical professionals using LLMs have encountered hallucinations and 84% believe they could impact patient health arxiv.org/abs/2503.05777

$( varepsilon, δ)$ Considered Harmful: Best Practices for Reporting Differential Privacy Guarantees Juan Felipe Gomez, Bogdan Kulynych, Georgios Kaissis, Jamie Hayes, Borja Balle, Antti Honkela http://arxiv.org/abs/2503.10945

All the silly hysterical media debates of 2023 about how AI may pose an existential threat to humankind really missed the mark. The harms are here, & they are not caused by AI autonomy. They are the result of how bosses & autocrats use machine learning systems to discipline & destroy.

I have no idea how this works but it's really cool

Still thinking about this young mathematician. Her name was Yulia Zdanovska. She could have done so much for her country and for our shared world. We all lost her. Killed by Putin’s Russia in Kharkiv almost exactly three years ago. Memorial from the ACM: cacm.acm.org/news/in-memo...

The US voted *with* Russia against the UN resolution condemning the invasion of Ukraine and demanding a withdrawal. Utter betrayal of country, allies, and principles.

The weaponization of US technology companies means we should treat our dependence on them as a national security threat ↘️

Elon Musk’s terrifying vision for AI; short, urgent new piece at Marcus on AI. I can’t remain silent - and hope you won’t, either. open.substack.com/pub/garymarc...

🚨 Do LLMs fall for spin in medical literature? 🤔 In our new preprint, we find that LLMs are susceptible to biased reporting of clinical treatment benefits in abstracts—more so than human experts. 📄🔍 [1/7] Full Paper: arxiv.org/abs/2502.07963 🧵👇

how much lighter touch can you get than applying standards to high risk systems without any need for anyone to check your working, followed by a prohibition on member states from further regulating the placing on the market or use of AI in any way?

Jack Posobiec has boosted not one, not two, but *three* Russian intelligence-backed operations. One of them, a website called SouthFront that U.S. Treasury sanctioned in 2021 and said "receives taskings from the FSB," regularly published disinformation on the war in Ukraine.

I tried some of the prompts from this paper and found that most of the time "preferences" are fully determined by which alternative is presented first