Goto

Collaborating Authors

 Personal


RLAC: Reinforcement Learning with Adversarial Critic for Free-Form Generation Tasks

arXiv.org Artificial Intelligence

Open-ended generation tasks require outputs to satisfy diverse and often implicit task-specific evaluation rubrics. The sheer number of relevant rubrics leads to prohibitively high verification costs and incomplete assessments of a response, making reinforcement learning (RL) post-training with rubric-based rewards difficult to scale. This problem is exacerbated by the fact that often the best way to combine these rubrics into one single reward is also highly prompt-specific. We propose Reinforcement Learning with Adversarial Critic (RLAC), a post-training approach that addresses these challenges via dynamic rubric verification. Our approach employs a large language model (LLM) as a critic that dynamically identifies only the most likely failure modes (e.g., a factual error or unhandled edge case), which are then verified by an external validator to optimize both generator and critic jointly. By training both the generator and the critic, this game enhances the critic's error detection and the generator's output quality while reducing required verifications. Our experiments demonstrate that RLAC improves factual accuracy in text generation and correctness in code generation, while also outperforming exhaustive verification and reward model methods. We show that dynamic critics are more effective than fixed critics, showcasing the potential of RLAC for scaling RL post-training to free-form generation tasks.


Open Character Training: Shaping the Persona of AI Assistants through Constitutional AI

arXiv.org Artificial Intelligence

The character of the "AI assistant" persona generated by modern chatbot large language models influences both surface-level behavior and apparent values, beliefs, and ethics. These all affect interaction quality, perceived intelligence, and alignment with both developer and user intentions. The shaping of this persona, known as character training, is a critical component of industry post-training, yet remains effectively unstudied in the academic literature. We introduce the first open implementation of character training, leveraging Constitutional AI and a new data pipeline using synthetic introspective data to shape the assistant persona in a more effective and controlled manner than alternatives such as constraining system prompts or activation steering. Specifically, we fine-tune three popular open-weights models using 11 example personas, such as humorous, deeply caring, or even malevolent. To track the effects of our approach, we introduce a method which analyzes revealed preferences, uncovering clear and holistic changes in character. We find these changes are more robust to adversarial prompting than the above two alternatives, while also leading to more coherent and realistic generations. Finally, we demonstrate this fine-tuning has little to no effect on general capabilities as measured by common benchmarks. We describe and open-source our full post-training method, the implementation of which can be found at https://github.com/maiush/OpenCharacterTraining.


DEEPAMBIGQA: Ambiguous Multi-hop Questions for Benchmarking LLM Answer Completeness

arXiv.org Artificial Intelligence

Large language models (LLMs) with integrated search tools show strong promise in open-domain question answering (QA), yet they often struggle to produce complete answer set to complex questions such as Which actor from the film Heat won at least one Academy Award?, which requires (1) distinguishing between multiple films sharing the same title and (2) reasoning across a large set of actors to gather and integrate evidence. Existing QA benchmarks rarely evaluate both challenges jointly. To address this, we introduce DeepAmbigQAGen, an automatic data generation pipeline that constructs QA tasks grounded in text corpora and linked knowledge graph, generating natural and verifiable questions that systematically embed name ambiguity and multi-step reasoning. Based on this, we build DeepAmbigQA, a dataset of 3,600 questions requiring multi-hop reasoning and half of them explicit name ambiguity resolving. Experiments reveal that, even state-of-the-art GPT-5 show incomplete answers, achieving only 0.13 exact match on ambiguous questions and 0.21 on non-ambiguous questions. These findings highlight the need for more robust QA systems aimed at information gathering and answer completeness.


Balancing Caregiving and Self-Care: Exploring Mental Health Needs of Alzheimer's and Dementia Caregivers

arXiv.org Artificial Intelligence

Alzheimer's Disease and Related Dementias (AD/ADRD) are progressive neurodegenerative conditions that impair memory, thought processes, and functioning. Family caregivers of individuals with AD/ADRD face significant mental health challenges due to long-term caregiving responsibilities. Yet, current support systems often overlook the evolving nature of their mental wellbeing needs. Our study examines caregivers' mental wellbeing concerns, focusing on the practices they adopt to manage the burden of caregiving and the technologies they use for support. Through semi-structured interviews with 25 family caregivers of individuals with AD/ADRD, we identified the key causes and effects of mental health challenges, and developed a temporal mapping of how caregivers' mental wellbeing evolves across three distinct stages of the caregiving journey. Additionally, our participants shared insights into improvements for existing mental health technologies, emphasizing the need for accessible, scalable, and personalized solutions that adapt to caregivers' changing needs over time. These findings offer a foundation for designing dynamic, stage-sensitive interventions that holistically support caregivers' mental wellbeing, benefiting both caregivers and care recipients.


Mysterious drones spotted over military base storing US nuclear weapons

Daily Mail - Science & tech

China's president Xi caught knifing Trump in brutal attack just hours after historic summit World's'most trusted' broadcaster the BBC doctored Trump speech a week before the election, whistleblower reveals I won't ever forget what I saw at Andy Cohen's party. He may admit he's hooking up with guys on every dating app but this is the truth about men like him: KENNEDY'Venomous' Republican split over Israel hits new low as fiery feud reaches White House America's most dangerous cities revealed: Crime, natural disaster risks and financial safety top the list of growing concerns Drivers mock new design for world's best-selling car: 'Did it already get into a wreck?' I learned the horrifying risks of'miracle' ADHD drugs and stopped taking them... but it was too late Roller coaster camera caught utter terror on people's faces after seat belt failed on 208ft ride that travels at 75mph The leafy suburb under an hour from Manhattan where wealthy New Yorkers are fleeing to escape'woke' Mamdani's socialist dystopia The five cities with America's most pleasant climate revealed - and they're all in the same state A girl, 15, bludgeoned to death in a gated enclave, a Kennedy cousin released and the brother who'knows the truth' about the death that haunts Camelot Sex aids and poppers... the sordid discoveries made by royal aides after party Andrew threw for Epstein and Ghislaine Maxwell - and the truth about those massages: ROBERT JOBSON READ MORE: New Jersey UFO mystery solved! Mysterious drones were spotted near Belgium's Kleine Brogel air base, where US nuclear weapons are stored, prompting fears of a potential espionage operation. Belgium's Defense Minister Theo Francken confirmed that drones entered the base's airspace in two waves on Saturday and Sunday night.


Chefs, your jobs are safe for now! Humanoid robot attempts to cook a stir-fry - but ends up flinging the food on the floor and slipping over in the mess

Daily Mail - Science & tech

Trump threatens to walk out on Norah O'Donnell as 60 Minutes EDITS OUT astonishing meltdown White House makes'venomous' split with Israel: Fiery feud engulfs Trump insiders with alliance on the brink I won't ever forget what I saw at Andy Cohen's party. He may admit he's hooking up with guys on every dating app but this is the truth about men like him: KENNEDY Sad secrets of privileged son, 20, accused of murdering his self-made single mother near their $1.9m home, then screaming'Mama' Three Americans among seven killed when avalanche obliterates Himalayan climbers' base camp Thomas Massie remarries 16 months after losing wife of 31 years... as Trump ally launches sick attack Trump stuns 60 Minutes' Norah O'Donnell as he breaks terrifying news about China and Russia nukes Ex-CIA spy shares an easy way to tell if someone is lying... and the tactic he uses to strengthen his love life Justin Baldoni's bombshell $400M case against Blake Lively and Ryan Reynolds is'formally ended by a judge' JD Vance declares himself'UFO' lunatic as he vows to pull back the curtain on government secrets Sex aids and poppers... the sordid discoveries made by royal aides after party Andrew threw for Epstein and Ghislaine Maxwell - and the truth about those massages: ROBERT JOBSON Top Democrat lawmaker becomes international fugitive after she was freed on bail'for stealing thousands from vulnerable man, 83' George Clooney gives rare insight into life with wife Amal and their twins - as he details his relationship with his kids, lauds his'beautiful' family and brands himself'very lucky' Shohei Ohtani's wife makes rare appearance to celebrate Dodgers star's World Series win I learned the horrifying risks of'miracle' ADHD drugs and stopped taking them... but it was too late A girl, 15, bludgeoned to death in a gated enclave, a Kennedy cousin released and the brother who'knows the truth' about the death that haunts Camelot Justin Trudeau's rapper son sounds worse than ever in latest music video despite father's burgeoning romance with Katy Perry Moment'knifeman who hurt 11 people in Huntingdon train rampage storms barber shop moments after stabbing 14-year-old boy' Meghan is mocked for her new Christmas recipe... boiled water! Chefs, your jobs are safe for now! Robots might be poised to replace humans in factories and warehouses, but chefs don't need to worry about losing their jobs anytime soon. In a viral video, which has amassed over 6.3 million views, a humanoid robot attempts to make a stir-fry for its owner - with disastrous results.


JD Vance vows to 'get to the bottom' of UFO phenomena as disclosure edges closure

Daily Mail - Science & tech

Trump stuns 60 Minutes' Norah O'Donnell as he breaks terrifying news about China and Russia nukes Justin Baldoni's bombshell $400M case against Blake Lively and Ryan Reynolds is'formally ended by a judge' Thomas Massie remarries 16 months after losing wife of 31 years... as Trump ally launches sick attack Sex aids and poppers... the sordid discoveries made by royal aides after party Andrew threw for Epstein and Ghislaine Maxwell - and the truth about those massages: ROBERT JOBSON I learned the horrifying risks of'miracle' ADHD drugs and stopped taking them... but it was too late George Clooney gives rare insight into life with wife Amal and their twins - as he details his relationship with his kids, lauds his'beautiful' family and brands himself'very lucky' Astonishing new evidence of Atlantis reveals advanced civilization preserved by Ancient Egypt's priests... before disaster hit Three Americans among seven killed when avalanche obliterates Himalayan climbers' base camp Trump's secret plan to deploy US troops to Mexico revealed with drone strikes in the works Jayden Daniels' X-ray results revealed in huge moment for the Commanders' season after QB's horror elbow injury Top Democrat lawmaker becomes international fugitive after she was freed on bail'for stealing thousands from vulnerable man, 83' I won't ever forget what I saw at Andy Cohen's party. He may admit he's hooking up with guys on every dating app but this is the truth about men like him: KENNEDY Ex-CIA spy shares an easy way to tell if someone is lying... and the tactic he uses to strengthen his love life Shohei Ohtani's wife makes rare appearance to celebrate Dodgers star's World Series win Deborra-Lee Furness' bold move after split from Hugh Jackman - and why the actor is not happy about it So many single men are taking this new drug cocktail before dates. The results in the bedroom are startling... as I discovered during one marathon session: JANA HOCKING Devastating impact of Mamdani's election will be FAR WORSE than first thought: Exclusive poll finds America's greatest city facing'historic' population wipeout Moment'knifeman who hurt 11 people in Huntingdon train rampage storms barber shop moments after stabbing 14-year-old boy' Meghan is mocked for her new Christmas recipe... boiled water! JD Vance vows to'get to the bottom' of UFO phenomena as disclosure edges closure MORE: Fierce debate erupts over'non-human' technology in space after video captures UFO surviving Hellfire strike Vice President JD Vance has joined the list of high-ranking government officials wanting answers about UFOs and extraterrestrials. In an interview released Wednesday, the vice president doubled down on his promise to'get to the bottom of' the existence of alien life .


Forthcoming machine learning and AI seminars: November 2025 edition

AIHub

This post contains a list of the AI-related seminars that are scheduled to take place between 3 November and 31 December 2025. All events detailed here are free and open for anyone to attend virtually. Agni Orfanoudaki (University of Oxford) Association of European Operational Research Societies To receive the seminar link, sign up to the mailing list . Nicholas Barbara (University of Sydney) EPFL Zoom link is here . Jose Carrillo (University of Oxford) University of Minnesota Zoom registration is here .


The Case That A.I. Is Thinking

The New Yorker

The Case That A.I. Is Thinking ChatGPT does not have an inner life. Yet it seems to know what it's talking about. How convincing does the illusion of understanding have to be before you stop calling it an illusion? Dario Amodei, the C.E.O. of the artificial-intelligence company Anthropic, has been predicting that an A.I. "smarter than a Nobel Prize winner" in such fields as biology, math, engineering, and writing might come online by 2027. He envisions millions of copies of a model whirring away, each conducting its own research: a "country of geniuses in a datacenter." In June, Sam Altman, of OpenAI, wrote that the industry was on the cusp of building "digital superintelligence." "The 2030s are likely going to be wildly different from any time that has come before," he asserted. Meanwhile, the A.I. tools that most people currently interact with on a day-to-day basis are reminiscent of Clippy, the onetime Microsoft Office "assistant" that was actually more of a gadfly. A Zoom A.I. tool suggests that you ask it "What are some meeting icebreakers?" or instruct it to "Write a short message to share gratitude." Siri is good at setting reminders but not much else. A friend of mine saw a button in Gmail that said "Thank and tell anecdote." When he clicked it, Google's A.I. invented a funny story about a trip to Turkey that he never took. The rushed and uneven rollout of A.I. has created a fog in which it is tempting to conclude that there is nothing to see here--that it's all hype. There is, to be sure, plenty of hype: Amodei's timeline is science-fictional.


Waymo killed KitKat. California neighborhood mourns a corner-store cat

Los Angeles Times

Things to Do in L.A. Tap to enable a layout that focuses on the article. KitKat was friendly with many customers of Randa's Market in San Francisco's Mission District. This is read by an automated voice. Please report any issues or inconsistencies here . San Francisco has been mourning the death of KitKat, a beloved corner-store cat who died after being struck by a Waymo robotaxi last week.