Knowledge catalogue
arXiv:2601.19963v2 Announce Type: replace-cross Abstract: Training a high-performing neural decoder can be difficult when only limited data are available from a recording session. To address this chal
arXiv:2604.22562v1 Announce Type: cross Abstract: Client contribution estimation in Federated Learning is necessary for identifying clients' importance and for providing fair rewards. Current methods
Don't try to build a self-improving AI agent without evals. You are just wasting time and compute. An agent can't improve from traces it can't evaluate. This is why it's exciting to see @FutureAGI_ go
arXiv:2604.22152v1 Announce Type: new Abstract: Evaluating robotics policies across thousands of environments and thousands of tasks is infeasible with existing approaches. This motivates the need for
arXiv:2604.22119v1 Announce Type: new Abstract: As reasoning capacity and deployment scope grow in tandem, large language models (LLMs) gain the capacity to engage in behaviors that serve their own ob
arXiv:2604.22167v1 Announce Type: cross Abstract: Language models are increasingly capable and are being rapidly deployed on a population-level scale. As a result, the safety of these models is increa
arXiv:2604.22089v1 Announce Type: cross Abstract: Generative Artificial Intelligence (GAI) systems that can automatically generate content in the form of source code or other contents (e.g., images) h
arXiv:2604.22586v1 Announce Type: new Abstract: We propose FlowAnchor, a training-free framework for stable and efficient inversion-free, flow-based video editing. Inversion-free editing methods have
Gary Marcus is a prominent AI researcher, cognitive scientist, and entrepreneur who frequently discusses artificial intelligence, machine learning, and cognitive science on X (formerly Twitter). His p
@GaryMarcus Never ever use generative AI for anything critical. The technology is probablistic all the way down and therefore inherently unreliable. Why is this so hard to understand? It’s wild how ma
@GaryMarcus These coding tools are intellectual chain saws. Powerful in the hands of a caring professional, extremely dangerous to the operator when used improperly. Implying AGI-adjacency has created
Generative AI as the Hindenberg 🦔 Michael Wooldridge, professor of AI at Oxford, is warning that the race to market AI has raised the risk of a 'Hindenburg moment' that could shatter global confidence
arXiv:2601.10386v2 Announce Type: replace-cross Abstract: Accurate survival prediction in Non-Small Cell Lung Cancer (NSCLC) requires integrating clinical, radiological, and histopathological data. Mu
arXiv:2604.22193v1 Announce Type: new Abstract: Large language models (LLMs) often need to balance their internal parametric knowledge with external information, such as user beliefs and content from
arXiv:2604.22103v1 Announce Type: cross Abstract: Street-view perception models predict subjective attributes such as safety at scale, but remain correlational: they do not identify which localized vi
arXiv:2502.03698v4 Announce Type: replace Abstract: Learning from demonstrations is a popular approach to train AI models; however, their vulnerability to adversarial attacks remains underexplored. We
I don’t always agree with @ElonMusk. I often disagree. Sometimes loudly. But he’s basically right here, as far I can see. Scam Altman and Greg Stockman stole a charity. Full stop. Greg got tens of bil
arXiv:2604.22631v1 Announce Type: new Abstract: Modern automatic speech recognition (ASR) systems have been observed to function better for certain speaker groups (SGs) than others, despite recent gai
If extremely violent criminals are not imprisoned, eventually they will murder innocent people This bodega owner told ABC a year ago that he fears for his safety in NY Last night, he was kiIIed by a s
arXiv:2604.22479v1 Announce Type: new Abstract: Driver drowsiness is a major cause of traffic accidents worldwide, posing a serious threat to public safety. Vision-based driver monitoring systems ofte
Gary Marcus discusses technical and security issues affecting the X platform, including malfunctioning polls, compromised accounts, and increased phishing scams in direct messages. The post appears to
arXiv:2604.22235v1 Announce Type: cross Abstract: Industrial robots are widely used in manufacturing, yet most manipulation still depends on fixed waypoint scripts that are brittle to environmental ch
arXiv:2604.22244v1 Announce Type: new Abstract: Ensuring safety for black-box hybrid dynamical systems presents significant challenges due to their instantaneous state jumps and unknown explicit nonli
arXiv:2604.22565v1 Announce Type: cross Abstract: Large Language Models (LLMs) can reason well, yet often miss decisive evidence when it is buried in long, noisy contexts. We introduce HiLight, an Evi
arXiv:2408.16286v5 Announce Type: replace Abstract: Designing a safe policy for uncertain environments is crucial in real-world control systems. However, this challenge remains inadequately addressed
arXiv:2604.22158v1 Announce Type: cross Abstract: We study the problem of adaptive control of the stochastic linear quadratic regulator (LQR) with constraints that must be satisfied at every time step
On Star Price's new @hitpausepod podcast, ControlAI's US Director Connor Leahy (@NPCollapse) explains how AIs can already tell when they're being tested. Currently, we can still catch them cheating, b
arXiv:2604.22540v1 Announce Type: cross Abstract: Most Neural Networks (NNs) for classification are trained using Cross-Entropy as a loss function. This approach requires the model to have an explicit
People keep blaming the users for the growing number of vibe-coded reasoners. Doing so is *half* right. Users *are*screwing up - by letting vibe coded stuff access their files, without proper backups,
arXiv:2604.22117v1 Announce Type: cross Abstract: Aligned large language models(LLMs) remain vulnerable to adversarial manipulation, and their dependence on web-scale pretraining creates a subtle but
arXiv:2604.22169v1 Announce Type: cross Abstract: Generic group-based RL assumes that sampled rollout groups are already usable learning signals. We show that this assumption breaks down in sparse-hit
arXiv:2604.22143v1 Announce Type: cross Abstract: Large language models are increasingly used to mediate everyday interpersonal dilemmas, yet how their advisory defaults interact with the concentrated
arXiv:2604.22591v1 Announce Type: new Abstract: The real-world deployment of Vision-Language-Action (VLA) models remains limited by the risk of unpredictable and irreversible physical harm. However, w
arXiv:2604.22154v1 Announce Type: cross Abstract: Emerging AI systems in behavioral health and psychiatry use multi-step or multi-agent LLM pipelines for tasks like assessing self-harm risk and screen
arXiv:2604.22662v1 Announce Type: cross Abstract: Shapley values are a cornerstone of explainable AI, yet their proliferation into competing formulations has created a fragmented landscape with little
@robertskmiles @GaryMarcus “fails less often than a human does” is a terrible metric. We build deterministic software and test it to destruction to ensure it’s 99.99999 reliable . That’s been the basi
Safety and innovation are not mutually exclusive: for many companies, especially in high-trust industries, AI’s risks are also a hindrance to adoption. In this op-ed in the @FT, I underscore that Euro
Sam Altman cannot be trusted. • The OpenAI board fired him because he was not always honest with them. They said he should not control powerful AI. • A major report talked to over 100 people and saw s
Scam Altman didn’t tell the OpenAI board that he OWNED the OpenAI Startup Fund. Altman lied in congressional testimony that he didn’t have financial gain from OpenAI. Ex-board member of OpenAI calls S
arXiv:2604.22374v1 Announce Type: new Abstract: Sign language translation (SLT) converts continuous sign videos into spoken-language text, yet it remains challenging due to the intrinsic modality mism
arXiv:2511.14427v3 Announce Type: replace-cross Abstract: Effective contact-rich manipulation requires robots to synergistically leverage vision, force, and proprioception. However, Reinforcement Lear
Stop blaming the users. It’s not that simple: People keep blaming the users for the growing number of vibe-coded reasoners. Doing so is *half* right. Users *are*screwing up - by letting vibe coded stu
arXiv:2510.11586v2 Announce Type: replace Abstract: Many in-silico simulations of human survey responses with large language models (LLMs) focus on generating closed-ended survey responses, whereas LL
arXiv:2601.12430v2 Announce Type: replace Abstract: Vision-language model (VLM) hallucination is commonly linked to imbalanced allocation of attention across input modalities: system, image and text.
arXiv:2604.22337v1 Announce Type: new Abstract: Most tabular-data generators match marginal statistics yet ignore causal structure, leading downstream models to learn spurious or unfair patterns. We p
Gary Marcus critiques common misconceptions about how large language models (LLMs) function, suggesting that most users have an incorrect mental model of what LLMs actually do when generating response
This statement critiques the asymmetrical attribution of outcomes in AI development, where users are blamed for failures while AI models receive credit for successes. Gary Marcus uses the metaphor of
arXiv:2604.21938v1 Announce Type: cross Abstract: Embodied AI is widely discussed as a job-displacement problem. The deeper risk, however, is governance lag: the inability of public institutions to ke
The divorce between Microsoft and OpenAI that I projected three years ago is now underway. Time for everyone to start seeing other people. OpenAI is moving away from its exclusive Microsoft arrangemen
The EU AI Act is the first comprehensive regulation for AI systems, and it’s right around the corner. With the clock running down, we took a deeper look at what these requirements mean for your agents
Gary Marcus made a critical commentary on Sam Altman's priorities, using a pun on 'AGI' (Artificial General Intelligence) to suggest that Altman's actual focus is on 'Adjusted Gross Income' rather tha
The people building the most powerful technology in history cannot tell you what is happening inside their own systems. This is insane. We started the Torchbearer Community for exactly this reason. Re
The question of whether Altman and Brockman did what they promised ≠ the question of whether Elon is a good person. Musk argued that Altman and Brockman broke promises and abused the notion of a nonpr
there’s seem to be a ton of confusion about this post. for clarity: The question of whether Altman and Brockman did what they promised ≠ the question of whether Elon is a good person. Musk argued that
arXiv:2604.22351v1 Announce Type: cross Abstract: Mid-infrared astronomy from the ground faces critical challenges in accurately detecting and quantifying sources due to the dominant spatially and tim
arXiv:2604.22709v1 Announce Type: new Abstract: While long, explicit chains-of-thought (CoT) have proven effective on complex reasoning tasks, they are costly to generate during inference. Non-verbal
this is exactly what tools like @denieddotdev was built for (behavioral auth) some agent behavior should be deterministically blocked by a separate policy layer, not via prompt instructions reach out
This is the 7th consecutive week with new reporting that ChatGPT was used in connection with murder or suicide. OpenAI doesn’t “benefit all of humanity.” Safety researchers keep quitting OpenAI becaus
This is totally wrong. Blaming the user is missing the point that (a) coding agents have been overhyped and (b) can’t reliably obey the rules given to them in system prompts and other guardrails. Sorr
arXiv:2506.17299v2 Announce Type: replace-cross Abstract: As large language models (LLMs) become increasingly deployed in safety-critical applications, the lack of systematic methods to assess their v