Hackers made death threats against this security researcher. In April 2024, a mysterious someone using the online handles "Waifu" and "Judische" began posting death threats on Telegram and Discord channels aimed at a cybersecurity researcher named Allison Nixon. These anonymous personas targeted Nixon because she had become a formidable threat: As chief research officer at the cyber investigations firm Unit 221B, named after Sherlock Holmes's apartment, she had built a career tracking cybercriminals and helping get them arrested. Though she'd done this work for more than a decade, Nixon couldn't understand why the person behind the accounts was suddenly threatening her. And although she had taken an interest in the Waifu persona in years past for crimes he boasted about committing, he hadn't been on her radar for a while when the threats began, because she was tracking other targets. Now Nixon resolved to unmask Waifu/Judische and others responsible for the death threats--and take them down for crimes they admitted to committing.

ALS stole this musician's voice. AI let him sing again. Patrick Darling used a music tool from ElevenLabs to perform a song with his former bandmates. There are tears in the audience as Patrick Darling's song begins to play. It's a heartfelt song written for his great-grandfather, whom he never got the chance to meet. But this performance is emotional for another reason: It's Darling's first time on stage with his bandmates since he lost the ability to sing two years ago.

Post-training of Large Language Models (LMs) often prioritizes accuracy and helpfulness at the expense of diversity. This creates a tension: while post-training improves response quality, it also sharpens output distributions and reduces the range of ideas, limiting the usefulness of LMs in creative and exploratory tasks such as brainstorming, storytelling, or problem solving. We address this challenge with Diversity-Aware Reinforcement Learning (DARLING), a framework that jointly optimizes for response quality and semantic diversity. At its core, DARLING introduces a learned partition function to measure diversity beyond surface-level lexical variations. This diversity signal is then combined with a quality reward during online reinforcement learning, encouraging models to generate outputs that are both high-quality and distinct. Experiments across multiple model families and sizes show that DARLING generalizes to two regimes: non-verifiable tasks (instruction following and creative writing) and verifiable tasks (competition math). On five benchmarks in the first setting, DARLING consistently outperforms quality-only RL baselines, producing outputs that are simultaneously of higher quality and novelty. In the second setting, DARLING achieves higher pass@1 (solution quality) and pass@k (solution variety). Most strikingly, explicitly optimizing for diversity catalyzes exploration in online RL, which manifests itself as higher-quality responses.

We test the robustness of debate as a method of scalable oversight by training models to debate with data generated via self-play. In a long-context reading comprehension task, we find that language model based evaluators answer questions more accurately when judging models optimized to win debates. By contrast, we find no such relationship for consultancy models trained to persuade a judge without an opposing debater present. In quantitative and qualitative comparisons between our debate models and novel consultancy baselines, we find evidence that debate training encourages stronger and more informative arguments, showing promise that it can help provide high-quality supervision for tasks that are difficult to directly evaluate.

Pigs, rats, and locusts have it easy these days--they can bother whoever they want. But back in the Middle Ages, such behavior could have landed them in court. If a pig bit a child, town officials would hold a trial like they would for a person, even providing the offender with a lawyer. Getting insects to show up in court en masse was a bit more difficult, but the authorities tried anyway: They'd send someone out to yell the summons into the countryside. That's hilarious, yes, but also a hint at how humans might navigate a new, even more complicated relationship.

Medical knowledge graphs (KGs) constructed from Electronic Medical Records (EMR) contain abundant information about patients and medical entities. The utilization of KG embedding models on these data has proven to be efficient for different medical tasks. However, existing models do not properly incorporate patient demographics and most of them ignore the probabilistic features of the medical KG. In this paper, we propose DARLING (Demographic Aware pRobabiListic medIcal kNowledge embeddinG), a demographic-aware medical KG embedding framework that explicitly incorporates demographics in the medical entities space by associating patient demographics with a corresponding hyperplane. Our framework leverages the probabilistic features within the medical entities for learning their representations through demographic guidance. We evaluate DARLING through link prediction for treatments and medicines, on a medical KG constructed from EMR data, and illustrate its superior performance compared to existing KG embedding models.

The W3C Web Ontology Language (OWL) is a powerful knowledge representation formalism at the basis of many semantic-centric applications. Since its unrestricted usage makes reasoning undecidable already in case of very simple tasks, expressive yet decidable fragments have been identified. Among them, we focus on OWL 2 RL, which offers a rich variety of semantic constructors, apart from supporting all RDFS datatypes. Although popular Web resources - such as DBpedia - fall in OWL 2 RL, only a few systems have been designed and implemented for this fragment. None of them, however, fully satisfy all the following desiderata: (i) being freely available and regularly maintained; (ii) supporting query answering and SPARQL queries; (iii) properly applying the sameAs property without adopting the unique name assumption; (iv) dealing with concrete datatypes. To fill the gap, we present DaRLing, a freely available Datalog rewriter for OWL 2 RL ontological reasoning under SPARQL queries. In particular, we describe its architecture, the rewriting strategies it implements, and the result of an experimental evaluation that demonstrates its practical applicability. This paper is under consideration in Theory and Practice of Logic Programming (TPLP).

I've been a technology activist for decades now, and I've read innumerable profound and enduring critiques of technology. In recent years, though, artificial intelligence has come under more fire than most developing trends. The pronouncements, hype, and foolishness surrounding it have risen to heights that stand out even by the outlandish standards of tech absurdity. Like me, you've probably encountered some of the better, smarter critiques along with all the silliness and insanity. Some of the greats are Cathy O'Neil's outstanding 2016 book Weapons of Math Destruction, and the excellent research reports from the nonprofit AI Now institute, and also Patrick Ball's spectacular papers published through the essential and dreadfully under-resourced Human Rights Data Analysis Group.

Despite the growing use of artificial-intelligence tools on a global basis, there is no universal code of ethics to govern its use. This is a key question the technology industry is beginning to wrestle with as the use of AI generates results both positive and negative. The technology has already been used for positive outcomes in a number of areas, including improving Australia's beaches, delivering reliable weather forecasts, and detecting human disease more accurately. There is also the other side of the coin, where AI has come under fire for injecting racial bias into criminal sentencing decisions and reinforcing gender discrimination. AI-powered facial recognition tools have been subjected to especially harsh criticism by privacy and human rights organizations.

Back in 2015, a hitchhiker was murdered on the streets of Philadelphia. It was no ordinary crime. The hitchhiker in question was a little robot called Hitchbot. The "death" raised an interesting question about human-robot relationship - not so much whether we can trust robots but whether the robots can trust us. The answer, it seems, was no.