safety

This is the most interesting paper I have read this week. The authors test a wide range of LLMs on a massive dataset of behavioural experiments, with more than 200,000 participants and nearly 26 milli

To folks who claim I am always wrong, I have a few questions: Was I wrong that these systems would continue to hallucinate and be untrustworthy? That Sam was a liar? That these companies would struggl

arXiv:2605.22709v1 Announce Type: cross Abstract: Electromagnetic (EM) side-channel analysis traditionally assumes a stationary, close-proximity probe - a threat model that underestimates aerial adver

arXiv:2602.02112v2 Announce Type: replace-cross Abstract: Masked diffusion models (MDMs) are a potential alternative to autoregressive models (ARMs) for language generation, but generation quality dep

arXiv:2605.06597v2 Announce Type: replace Abstract: Self-distillation (SD) offers a promising path for adapting large language models (LLMs) without relying on stronger external teachers. However, SD

arXiv:2605.21906v1 Announce Type: new Abstract: Computed tomography (CT) is a central to three-dimensional medical imaging, yet CT-based artificial intelligence remains fragmented across task-specific

Gary Marcus responds to Ethan Mollick's post about AI capabilities, noting that while he objects to claims made in the top-level post, Mollick does provide important caveats and nuance in subsequent r

Gary Marcus notes that researcher @thestalwart has found more recent AI models to be less vulnerable to certain attacks or exploits, and suggests that a comprehensive study across multiple models woul

arXiv:2605.21654v1 Announce Type: cross Abstract: Reinforcement learning substantially improves pretrained language models, but it remains understudied why critic-free methods such as PPO and GRPO wor

arXiv:2605.22817v1 Announce Type: cross Abstract: Language models must now generalize out of the box to novel environments and work inside inference-scaling search procedures, such as AlphaEvolve, tha

arXiv:2605.22651v1 Announce Type: new Abstract: CLIP-style contrastive pretraining typically curates web-scale image-text pairs using sample-level filtering signals, often based on pair-level alignmen

💯: Why OpenAI keeps taking childish shots at me, is exactly what @FrankRundatz says below: “you have the added audacity of being influential enough to move the needle on the timing of OpenAI’s record-

arXiv:2605.21801v1 Announce Type: cross Abstract: Post-training has become central to improving reasoning and alignment in large language models, where critic-free models enable scalable learning from

arXiv:2503.02885v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have gained traction in educational settings, often framed as virtual tutors or teaching assistants. Following ea

arXiv:2605.05765v2 Announce Type: replace Abstract: Inspired by the development of OpenClaw, there is a growing demand for mobile-based personal agents capable of handling complex and intuitive intera

Yeah, I don't think I've ever met someone who has directly worked for an extended period on safety or security at an AI company who thinks things are fine readiness wise or incentive wise etc. https:/

arXiv:2605.20940v1 Announce Type: new Abstract: Accurate estimation of wheat spike volume is important for yield component analysis and stress resilience assessment, yet field-based measurement remain

arXiv:2605.20494v1 Announce Type: new Abstract: Reliable assessment of tropical cyclone (TC) risk is limited by the brevity and spatial sparsity of the historical record, particularly for the rare, hi

arXiv:2410.03296v4 Announce Type: replace Abstract: Instruction-tuned LLMs are able to provide extit{an} explanation about their output to users by generating self-explanations, without requiring the

arXiv:2605.21125v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO), a prominent algorithm within the Reinforcement Learning from Verifiable Rewards (RLVR) framework, has achieve

arXiv:2602.04916v3 Announce Type: replace-cross Abstract: Large language models (LLMs) have significantly advanced protein representation learning. However, their capacity to interpret and design anti

arXiv:2605.20233v1 Announce Type: new Abstract: Assessing learner competency in clinical simulation requires expert observation that is time-intensive, difficult to scale, and subject to inter-rater v

arXiv:1908.05972v3 Announce Type: replace Abstract: This paper significantly improves on, and finishes to validate, an approach proposed in previous research in which safety outcomes were predicted fr

arXiv:2605.21083v1 Announce Type: cross Abstract: Materials discovery and biomedical translation increasingly require models that can reason across composition, processing, structure, biological respo

all these dudes posting about anthropic’s profits without checking the fine print. (see my earlier tweet) there’s some fine print i believe that n the openai result too, and a lot that has not been di

Always read the fine print: Anthropic is projecting its first (slightly) profitable quarter ever, which is amazing—assuming it actually happens —but if it does it will be in no small part because they

… and (if am not mistaken) the only one in the top 14 to have never turned a profit. Welcome to the world of vibe investing! OpenAI is likely to raise 60 billion in its IPO, more than double Saudi Ara

arXiv:2605.21109v1 Announce Type: new Abstract: Reliable confidence estimates are important for safely deploying vision-based controllers in autonomous racing, where safety predictions must be derived

approximately equals “I am already rich and fuck you if you lose your job to AI” Marc Andreessen to Joe Rogan: Why AI Workers Beat Human Workers⁣ ⁣ 'Never gets drunk. Never gets sick. Never gets depre

arXiv:2605.21115v1 Announce Type: cross Abstract: Federated learning (FL) has emerged as a promising paradigm for managing electric vehicle (EV) battery data in intelligent transportation systems (ITS

arXiv:1907.11769v4 Announce Type: replace Abstract: In light of the increasing availability of digitally recorded safety reports in the construction industry, it is important to develop methods to exp

arXiv:2605.20643v1 Announce Type: cross Abstract: Self-distillation enables language models to learn on-policy from their own trajectories by using the same model as both student and teacher, with the

arXiv:2602.08819v2 Announce Type: replace-cross Abstract: Reward models are central to aligning language models with human preferences via reinforcement learning (RL). As RL is increasingly applied to

arXiv:2605.21214v1 Announce Type: new Abstract: Reinforcement learning (RL) often exhibits high variance across training runs, leading to unreliable performance and posing a major challenge to deploym

arXiv:2605.21027v1 Announce Type: new Abstract: Enterprise analytics aims to make organizational data accessible for decision-making, yet non-technical users still face barriers when using traditional

arXiv:2605.20996v1 Announce Type: new Abstract: Most value-based and actor--critic reinforcement learning methods rely on Bellman-style recursions, yet these recursions collapse under non-exponential

Biomedical AI may be headed for a replication crisis. (This work below is not about AI-generated reports; it’s about studies of biomedicine that use ML in their methods, and how they are evaluted.) In

arXiv:2602.06500v2 Announce Type: replace Abstract: Scaling inference methods such as Markov chain Monte Carlo to high-dimensional models remains a central challenge in Bayesian deep learning. A promi

arXiv:2605.20282v1 Announce Type: new Abstract: Machine unlearning in Vertical Federated Learning (VFL) has attracted growing interest, yet existing methods certify forgetting solely using output-leve

can’t believe people assume that success on highly verifiable problems in math (where we don’t even know how many tests were performed and how many might have failed) iautomatically generalize to ever

Gary Marcus examines and fact-checks recent claims made by OpenAI and Anthropic in their public announcements and media coverage. The post links to detailed analysis questioning the mathematical valid

arXiv:2512.09806v2 Announce Type: replace Abstract: Deep learning-based methods have recently achieved significant success in image reconstruction problems. However, challenges have emerged, as these

arXiv:2605.20975v1 Announce Type: new Abstract: Federated Learning enables collaborative model training across decentralized data sources without data transfer. Averaging-based FL is limited by the pr

arXiv:2605.20971v1 Announce Type: new Abstract: The growing sophistication of GAN-based image manipulation presents significant challenges for digital forensics. This study compares the performance of

arXiv:2602.02304v2 Announce Type: replace-cross Abstract: Large-scale foundation models exhibit behavioral shifts when subjected to interventions such as scaling, fine-tuning, reinforcement learning w

arXiv:2602.08686v2 Announce Type: replace Abstract: Prefill-only KV compression freezes a token subset at the end of prefill and decodes from it without further eviction. The retention decision is the