safety

arXiv:2606.20645v1 Announce Type: new Abstract: Foothold selection on structured terrain requires explicit reasoning about contact planarity, surface steepness, and kinematic reachability, properties

arXiv:2606.22332v1 Announce Type: new Abstract: Tactile sensing is critical for contact-rich dexterous manipulation, yet it remains unclear which tactile abstractions a policy needs and when richer ta

arXiv:2606.22729v1 Announce Type: new Abstract: Diffusion policies enable multimodal robot behavior but offer limited ability to choose among behavior modes at inference time, even though such control

arXiv:2606.19752v2 Announce Type: replace Abstract: Long-horizon robot manipulation policies trained with reward shaping can still achieve high return through inefficient interactions, while rare effi

arXiv:2511.20889v2 Announce Type: replace Abstract: Test-time alignment (TTA) aims to adapt models to specific rewards during inference. However, existing methods tend to either under-optimise or over

arXiv:2606.22998v1 Announce Type: new Abstract: Text-conditioned motion generation is a promising interface for programming humanoid robots, yet current generators are often trained on human motion da

arXiv:2606.21619v1 Announce Type: cross Abstract: Large Language Models (LLMs) have demonstrated strong capabilities in code generation, but their outputs frequently contain syntax or type errors that

arXiv:2606.23123v1 Announce Type: new Abstract: Deep learning models, particularly Transformers and Neural Operators, exhibit a well-documented 'spectral bias,' effectively acting as low-pass filters

The Kremlin’s cognitive warfare is evolving. Leaked documents from Russia’s Social Design Agency reveal efforts to move beyond planting fabricated stories on social media. Now Russian influence operat

The article challenges the widespread assumption that AI surpasses human capabilities across all domains, arguing this narrative is inaccurate. It likely presents evidence or reasoning from AI researc

arXiv:2606.21911v1 Announce Type: cross Abstract: We identify a critical pitfall in scaling transformer-based sequential recommenders: while increasing model size improves recommendation accuracy, it

The recent Mythos/NSA warning shot is looking more and more like how I expected a 'warning shot' to look like. An absolutely insane thing happens, and then the FUD machine kicks into action and adds i

arXiv:2606.22516v1 Announce Type: cross Abstract: Input Diversity (DI), which applies random resizing and padding at each attack iteration, is a near-default ingredient of transfer-based adversarial a

arXiv:2606.22858v1 Announce Type: new Abstract: As machine learning models grow more influential and opaque, algorithmic fairness and explainability are critical for ensuring accountability. However,

I can't provide a summary of this content based on the information given. The URL appears to be fabricated (the tweet ID format and account name don't match actual X/Twitter patterns), and I cannot ve

This, once again, shows why the theory of change of 'we just wait for a wArNiNg ShOt and then everyone will suddenly be reasonable!' is flawed. We need to do a lot of work to make the situation legibl

arXiv:2606.23472v1 Announce Type: cross Abstract: Time series classification involves learning a mapping from a continuous, temporally ordered sequence of real-valued observations to a discrete respon

arXiv:2506.08026v3 Announce Type: replace-cross Abstract: Real-time market prediction services need correct predictions before a decision deadline; a correct prediction delivered late is not a usable

arXiv:2606.21295v1 Announce Type: new Abstract: Existing sequence models, including RNNs, LSTMs, continuous-time networks, and Transformers, share a common structural principle: layer-wise dynamics, w

arXiv:2606.16272v2 Announce Type: replace Abstract: Human hand-object demonstrations provide dense reference motions for training dexterous manipulation reinforcement learning (RL) policies through re

arXiv:2606.20748v1 Announce Type: new Abstract: Humanoid robots are emerging as co-workers in smart manufacturing, yet their dynamic, human-like movements introduce safety risks that differ fundamenta

arXiv:2512.02581v3 Announce Type: replace Abstract: Reinforcement learning (RL) faces a persistent tension: policies that are stable to optimize (e.g., Gaussians) are often too simple to represent the

arXiv:2606.22550v1 Announce Type: new Abstract: Autoregressive visual models (AVMs) based on next-scale prediction have emerged as a prominent paradigm for image and video synthesis. However, decompos

arXiv:2606.23496v1 Announce Type: new Abstract: Discrete text-trigger optimization -- searching for text sequences that, when ingested by a model, steer it toward a specified objective -- underpins mo

arXiv:2606.13714v2 Announce Type: replace Abstract: Unsupervised video object-centric learning aims to decompose dynamic scenes into temporally persistent entity representations. Existing recurrent vi

arXiv:2606.20971v1 Announce Type: new Abstract: We introduce UNITY, a Universal-to-Specialized adapter for efficient and scalable composite conditioning in diffusion based image generation. Unlike pri

arXiv:2606.21287v1 Announce Type: new Abstract: Unsupervised domain adaptation (UDA) enables robust transfer of knowledge from simulated to real environments while exploiting a subset of unlabeled tar

arXiv:2602.11944v2 Announce Type: replace Abstract: When building AI systems for decision support, one often encounters the phenomenon of predictive multiplicity: a single best model does not exist; i

arXiv:2606.22870v1 Announce Type: new Abstract: Recent advancements in chain-of-thought (CoT) reasoning have shown promise in enhancing video understanding and reasoning capabilities of multimodal lar

arXiv:2606.21600v1 Announce Type: new Abstract: Multi-task robot manipulation policies are challenging to learn from demonstration because traditionally a single network must select among qualitativel

arXiv:2508.03058v2 Announce Type: replace Abstract: Reinforcement Learning (RL) in real-world environments often suffers from ambiguous or incomplete reward supervision, which undermines policy stabil

arXiv:2606.22136v1 Announce Type: new Abstract: Scaling dexterous manipulation requires generalization across objects, scenes, and tasks, yet existing data sources face a trade-off between scale and s

arXiv:2606.21126v1 Announce Type: new Abstract: Despite the success of deep learning, training deep networks in biologically plausible and hardware-efficient ways remains an open challenge. Feedback a

arXiv:2603.06445v2 Announce Type: replace Abstract: Situated reasoning often relies on active exploration, yet in many real-world scenarios such exploration is infeasible due to physical constraints o

arXiv:2606.16196v2 Announce Type: replace-cross Abstract: Deep neural networks have achieved remarkable performance across medical imaging tasks, yet their tendency to overgeneralize under distributio

arXiv:2603.28919v2 Announce Type: replace Abstract: As robots increasingly enter the workforce, human-robot interaction (HRI) must address how implicit social biases influence user preferences. This p

arXiv:2606.22145v1 Announce Type: new Abstract: Reinforcement learning (RL) is a powerful and convenient tool to modernize controller design. In this work, we study the zero-shot transfer of RL-based

arXiv:2606.19340v2 Announce Type: replace Abstract: We present ZeroDex, a zero-shot framework for long-horizon dexterous manipulation that grounds language instructions into executable 3D task plans f

An interesting new paper by my recent PhD graduate on how AI agents' greed for visible incentives can lead them to abandon their safety alignment. You can read it here: https://arxiv.org/abs/2606.1691

An update. A US official tells me that Sen. Warner misunderstood the NSA director Gen. Rudd in this case. Rudd did use the 'hours, not weeks' wording, but the use of Mythos in this context was—as wide

'Because I'm going to be smarter than you' This clip is from a new short film by @ForegoneFilms on the race to develop superintelligence. Please watch the full 16 minute film (link below) to help unde

By the end of the decade and maybe a lot sooner, we will wonder why we ever treated these folks like they were gods. “The more I listen to AI company CEOs, the stranger they sound,” says another tech

Great report on LLM agent communication protocols. Communication is a huge bottleneck in multi-agent systems. (worth bookmarking) The report builds a five-dimensional taxonomy (counterparty, payload,

have been thinking a bunch about model routing and related things current thoughts here, would love feedback: 1/ there is a difference between 'model routing' and 'model council' 'model routing' = rou

This newsletter issue covers three major AI topics: techniques for making AI systems more persuasive ('superpersuasion'), the concept of self-sustaining or self-improving AI systems, and various theor

LLMs can’t be trusted to follow rules. Which means they can’t be trusted, period. If we are going to solve alignment we must move on. Large language models can be persuaded to break their own rules. N