Knowledge catalogue
arXiv:2605.04700v1 Announce Type: cross Abstract: Jailbreak attacks on audio language models (ALMs) optimize audio perturbations to elicit unsafe generations, and they typically update the entire wave
arXiv:2605.04345v1 Announce Type: new Abstract: We formally establish the equivalence between Observation Delay (OD) and Action Delay (AD) in cooperative partially observable multi-agent systems using
arXiv:2605.04617v1 Announce Type: new Abstract: Wearable human activity recognition (WHAR) models often suffer from performance degradation under real-world cross-user distribution shifts. Test-time a
Terence Tao recognized that plausibility and veracity are not the same, and that current tools are better at the former than the latter. [edit: the video is from 2024, and affirms what i said in 2019
Dozens of US states are considering legislation to allow people to install plug-in solar systems, often called balcony solar. These small arrays require little to no setup and could help cut emissions
The illustration on this story is quite funny, but the study itself has quite big implications I think. The point is that there might be better ways to design AI systems so that they don’t simply do e
Theories are proven by accurate predictions. I first read @GaryMarcus in the late ’90s, and his ideas helped shape my deterministic ML frameworks, GAIuS & KATO. He’s been proven right repeatedly, yet
“they hadn’t figured out how OpenAI would pay for it” may turn out to be the epitaph for an entire era. 🪦 scoop from @anissagardizy8 @theinformation, and credit her with the great line “OpenAI has mad
arXiv:2605.04653v1 Announce Type: new Abstract: Aligning large visual generative models with human feedback is often performed through pairwise preference optimization. While such approaches are conce
arXiv:2605.04081v1 Announce Type: new Abstract: Causal Bayesian Networks (CBNs) are a powerful tool for reasoning under uncertainty about complex real-world problems. Such problems evolve over time, r
arXiv:2605.04877v1 Announce Type: cross Abstract: Multimodal emotion recognition (MER) benefits from combining text, audio, and vision, yet standard fusion often fails when modalities conflict. Crucia
arXiv:2605.04494v1 Announce Type: cross Abstract: Reinforcement learning from human feedback (RLHF) has been popular for aligning text-to-image (T2I) diffusion models with human preferences. As a main
arXiv:2605.04946v1 Announce Type: new Abstract: Batch normalization (BN) is central to modern deep networks, but its effect on the realized function during training remains less understood than its op
Gary Marcus reflects on predictions he made two years prior and reaffirms his confidence in their accuracy. The post suggests Marcus is reviewing his track record on forecasts, likely related to artif
arXiv:2508.11196v2 Announce Type: replace Abstract: Recent advances in vision-language models (VLMs) have demonstrated strong generalization in natural image tasks. However, their performance often de
arXiv:2511.08195v3 Announce Type: replace Abstract: UI-to-code aims to translate UI screenshots into executable front-end code. Despite progress with vision-language models (VLMs), most existing metho
arXiv:2605.04730v1 Announce Type: new Abstract: Visual localization is a core technology for augmented reality and autonomous navigation. Recent methods combine the efficient rendering of 3D Gaussian
arXiv:2605.04874v1 Announce Type: cross Abstract: Direct Preference Optimization (DPO) has proven to be an effective solution for mitigating hallucination in Multimodal Large Language Models (MLLMs) b
arXiv:2605.03598v2 Announce Type: cross Abstract: Understanding how biological and artificial neural networks implement computation from connectivity is a central problem in neuroscience and machine l
arXiv:2505.11815v2 Announce Type: replace Abstract: Current vision-language models have been explored for multi-modal embedding tasks like information retrieval. However, they face significant challen
arXiv:2605.04732v1 Announce Type: new Abstract: Simulation-based planning with rollouts is a widely-deployed technique for decision making in stochastic environments. The primary instrument of simulat
arXiv:2512.05226v2 Announce Type: replace Abstract: Domain shift remains a key challenge in deploying machine learning models to the real world. Unsupervised domain adaptation (UDA) aims to address th
What happens when AIs become smarter than us? Why would they keep humans around if given the choice? Our new paper argues that only trying to control AIs is a limited strategy, and that a stable, mutu
arXiv:2605.05172v1 Announce Type: new Abstract: Behavior Cloning (BC) has emerged as a highly effective paradigm for robot learning. However, BC lacks a self-guided mechanism for online improvement af
arXiv:2605.04972v1 Announce Type: new Abstract: Aligning large language models with expert judgment is especially difficult in subjective evaluation tasks, where experts may disagree, rely on tacit cr
Jan Leike, OpenAI's safety and alignment lead, resigned in May 2024, citing concerns about the company's commitment to safety practices and the deprioritization of safety work relative to product deve
Wow this paper has been “just published” many times for nearly a year. I have called out at least two other “influencers” for the same thing on the same paper, including another one earlier this week.
yep, really embarassing One of the richest VCs, Marc Andreessen, has shown again that he thinks generative AI works perfectly well if you use the right prompts, thus ignoring the statistical nature of
100% stood the test of time: “What Ilya saw” was Sam’s bad behavior, not AGI. When Altman got fired “What did Ilya see?” became a wildly popular conspiracy meme. I think we can safely say now that wha
2nd episode of The Roman Forum is an interview with AI Safety/Governance expert Connor Leahy @NPCollapse. Connor is a great speaker and is lobbying to get government to ban Superintelligence. My first
arXiv:2605.01100v1 Announce Type: new Abstract: This work presents a knowledge-driven decision-support system that integrates structured defect knowledge with LLM-based reasoning to provide explainabl
arXiv:2605.03787v1 Announce Type: new Abstract: Labeling medical images is a major bottleneck in the field of medical imaging, as it requires domain-specific expertise, and it gets further complicated
arXiv:2506.04680v2 Announce Type: replace Abstract: During the development of wearable exoskeletons, evaluations involving human subjects pose inherent safety risks. Therefore, systematic testing is o
arXiv:2605.03262v1 Announce Type: new Abstract: We introduce the Yat kernel $k_{b,arepsilon}(mathbf{w},mathbf{x})=frac{(mathbf{w}^opmathbf{x}+b)^2}{|mathbf{x}-mathbf{w}|^2+arepsilon},qquad bge 0, arep
Jon Brodkin / Ars Technica: A US appeals court strikes down a 2023 FCC rule banning broadband access discrimination based on income, race, and more; Chair Brendan Carr welcomes the ruling — An appeals
arXiv:2508.14994v3 Announce Type: replace-cross Abstract: In hazardous and remote environments, robotic systems perform critical tasks demanding improved safety and efficiency. Among these, quadruped
arXiv:2605.00968v1 Announce Type: cross Abstract: Positional encoding plays a pivotal role in determin?ing the extrapolation and generalization performance of wireless foundation models for channel st
arXiv:2605.03212v1 Announce Type: cross Abstract: Modeling latent clinical constructs from unconstrained clinical interactions is a unique challenge in affective computing. We present ADAPTS (Agentic
arXiv:2605.02911v1 Announce Type: new Abstract: Future sixth-generation (6G) mobile networks are envisioned to be equipped with a diverse set of powerful, yet highly specialized, optimization experts.
arXiv:2605.01032v2 Announce Type: new Abstract: We present an algebraic semantics for governed execution in which governance is axiomatized, compositional, and coterminous with expressibility. The fra
arXiv:2509.20789v4 Announce Type: replace Abstract: The remarkable success of modern AI has been closely tied to scaling laws, yet the finite supply of high-quality data makes data efficiency--learnin
🚨Alphabet GOOGL trades at 133x free cash flow. For context: its pre-COVID multiple was ~20x. And free cash flow hasn't grown since 2021. GQG Partners — one of the world's top institutional investors —
Am I right that hyperscaling compute is the biggest bet in history? Any counter examples? It’s way more expensive than the Manhattan Project, the Apollo project, and railways across the US. If it does
Amazing the shit X gave me in November 2023 for saying Sam wasn’t always candid. He wasn’t. Period. Murati testified that, by fall 2023, Sam Altman was “not always” candid with her, and undermined her
And how many times have I told you that Sam is no longer the right CEO for OpenAI? Murati: My issues with Sam were very much around management and providing direction to the organization and decision
Anthropic: Anthropic researchers detail “model spec midtraining”, which adds a stage between pretraining and fine-tuning to improve generalization from alignment training — Sara Price2, Samuel Marks2,
arXiv:2605.01740v1 Announce Type: cross Abstract: An agentic-AI runtime issues tool calls, sends messages, and actuates devices on behalf of an LLM. Catching the four ways an action can diverge from i
arXiv:2605.04045v1 Announce Type: new Abstract: Audio-Visual Intelligence (AVI) has emerged as a central frontier in artificial intelligence, bridging auditory and visual modalities to enable machines
arXiv:2605.03222v1 Announce Type: new Abstract: Activation-alignment measures such as Representational Similarity Analysis (RSA), Canonical Correlation Analysis (CCA), and Centered Kernel Alignment (C
arXiv:2605.03452v1 Announce Type: new Abstract: High-quality data collection is a fundamental cornerstone for training humanoid whole-body visuomotor policies. Current data acquisition paradigms predo
arXiv:2512.04277v3 Announce Type: replace Abstract: Post-training with reinforcement learning (RL) typically optimizes a single scalar objective and ignores structure in how solutions are produced. We
arXiv:2601.00421v2 Announce Type: replace Abstract: This paper explores how semantic-space reasoning, traditionally used in computational linguistics, can be extended to tactical decision-making in te
arXiv:2605.01758v1 Announce Type: new Abstract: Large multimodal model-based Multi-Agent Systems (MASs) enable collaborative complex problem solving through specialized agents. However, MASs are vulne
arXiv:2604.06132v2 Announce Type: replace Abstract: Large language models are increasingly deployed as autonomous agents for multi-step workflows in real-world software environments. However, existing
C’mon BBC. Zilis was sharp as a tack on the stand, on her role in the OpenAI *nonprofit* board, and how she managed conflicts as they began to develop, and this (which is not really even news since it
arXiv:2605.02734v1 Announce Type: new Abstract: Learning to Defer (L2D) enables a model to predict autonomously or defer to an expert, but prior work largely assumes flat label spaces. We study the fi
Gary Marcus clarifies that in the legal proceeding he's discussing, the jury serves an advisory role while the judge retains decision-making authority on the case outcome. Marcus expresses confidence
arXiv:2605.02658v2 Announce Type: new Abstract: Shortcut learning causes deep learning models to rely on non-essential features within the data. However, its formation in deep neural network training
arXiv:2602.20078v3 Announce Type: replace-cross Abstract: Scaling cooperative multi-agent reinforcement learning (MARL) is fundamentally limited by cross-agent noise. When agents share a common reward
arXiv:2605.03327v1 Announce Type: new Abstract: Reinforcement learning is crucial for aligning large language models to perform complex reasoning tasks. However, current algorithms such as Group Relat