SegDiff: Segmented Trajectory Diffusion for Consistent and Adaptive Robot Manipulation

Imitation learning enables robots to acquire manipulation skills from demonstrations by mapping observations to actions. Existing approaches predict either short-horizon continuous action sequences or discrete keyposes. However, continuous prediction methods suffer from compounding errors due to short prediction horizons and struggle with multi-modal action distributions, whereas keypose-based methods necessitate an external planner, constraining real-time applicability. To address these challenges, we introduce SegDiff, a closed-loop visuomotor policy that integrates the strengths of both paradigms. SegDiff decomposes demonstrations into motion segments between keyposes and learns to predict the continuous trajectory from the current state to the next keypose, enabling long-horizon prediction with real-time refinement. Furthermore, we leverage the capability of diffusion models and DDIM inversion to propose a Dynamic Temporal Ensembling mechanism, which allows the policy to efficiently respond to dynamic environments and mitigate discontinuities caused by inconsistent multi-modal sampling. SegDiff demonstrates significant performance gains over existing approaches across various simulated and real-world scenarios, indicating its strong ability to reason over extended temporal dependencies while maintaining real-time adaptability and control stability.

Generative Models Erode Human Temporal Learning Through Market Selection

We argue that modern generative models create structural risks for knowledge and cultural production at current, sub-AGI capability levels. We define Human Temporal Learning (HTL) as path-dependent knowledge accumulation through sustained engagement with problems over time. Generative outputs increasingly resemble HTL-intensive work in surface features, so verifying whether a given output reflects genuine human learning grows costly relative to its expected benefit. Once verification loses economic justification, evaluators reward outputs regardless of production mode, and producers who invested years of learning compete on price against outputs that cost almost nothing to generate. We call this pathway value collapse and formalize it through a costly-inspection framework. Cross-domain evidence from academic publishing, legal practice, content platforms, and software security maps onto four stages of verification erosion. Alignment success is orthogonal. Better-aligned models narrow observable gaps between human and AI outputs, making source verification harder and intensifying competitive pressure against HTL-intensive work even when individual AI outputs improve.

Black Box Absorption: LLMs Undermining Innovative Ideas

Large Language Models are increasingly adopted as critical tools for accelerating innovation. This paper identifies and formalizes a systemic risk inherent in this paradigm: \textbf{Black Box Absorption}. We define this as the process by which the opaque internal architectures of LLM platforms, often operated by large-scale service providers, can internalize, generalize, and repurpose novel concepts contributed by users during interaction. This mechanism threatens to undermine the foundational principles of innovation economics by creating severe informational and structural asymmetries between individual creators and platform operators, thereby jeopardizing the long-term sustainability of the innovation ecosystem. To analyze this challenge, we introduce two core concepts: the idea unit, representing the transportable functional logic of an innovation, and idea safety, a multidimensional standard for its protection. This paper analyzes the mechanisms of absorption and proposes a concrete governance and engineering agenda to mitigate these risks, ensuring that creator contributions remain traceable, controllable, and equitable.

The Alignment Bottleneck

Large language models improve with scale, yet feedback-based alignment still exhibits systematic deviations from intended behavior. Motivated by bounded rationality in economics and cognitive science, we view judgment as resource-limited and feedback as a constrained channel. On this basis, we model the loop as a two-stage cascade $U \to H \to Y$ given $S$, with cognitive capacity $C_{\text{cog}|S}$ and average total capacity $\bar{C}_{\text{tot}|S}$. Our main result is a capacity-coupled Alignment Performance Interval. It pairs a data size-independent Fano lower bound proved on a separable codebook mixture with a PAC-Bayes upper bound whose KL term is controlled by the same channel via $m \, \bar{C}_{\text{tot}|S}$. The PAC-Bayes bound becomes an upper bound on the same true risk when the canonical observable loss is used and the dataset is drawn from the same mixture. Under these matched conditions, both limits are governed by a single capacity. Consequences include that, with value complexity and capacity fixed, adding labels alone cannot cross the bound; attaining lower risk on more complex targets requires capacity that grows with $\log M$; and once useful signal saturates capacity, further optimization tends to fit channel regularities, consistent with reports of sycophancy and reward hacking. The analysis views alignment as interface engineering: measure and allocate limited capacity, manage task complexity, and decide where information is spent.

Fight Fire with Fire: Defending Against Malicious RL Fine-Tuning via Reward Neutralization

Reinforcement learning (RL) fine-tuning transforms large language models while creating a vulnerability we experimentally verify: Our experiment shows that malicious RL fine-tuning dismantles safety guardrails with remarkable efficiency, requiring only 50 steps and minimal adversarial prompts, with harmful escalating from 0-2 to 7-9. This attack vector particularly threatens open-source models with parameter-level access. Existing defenses targeting supervised fine-tuning prove ineffective against RL's dynamic feedback mechanisms. We introduce Reward Neutralization, the first defense framework specifically designed against RL fine-tuning attacks, establishing concise rejection patterns that render malicious reward signals ineffective. Our approach trains models to produce minimal-information rejections that attackers cannot exploit, systematically neutralizing attempts to optimize toward harmful outputs. Experiments validate that our approach maintains low harmful scores (no greater than 2) after 200 attack steps, while standard models rapidly deteriorate. This work provides the first constructive proof that robust defense against increasingly accessible RL attacks is achievable, addressing a critical security gap for open-weight models.

Dynamic Action Interpolation: A Universal Approach for Accelerating Reinforcement Learning with Expert Guidance

Reinforcement learning (RL) suffers from severe sample inefficiency, especially during early training, requiring extensive environmental interactions to perform competently. Existing methods tend to solve this by incorporating prior knowledge, but introduce significant architectural and implementation complexity. We propose Dynamic Action Interpolation (DAI), a universal yet straightforward framework that interpolates expert and RL actions via a time-varying weight $\alpha(t)$, integrating into any Actor-Critic algorithm with just a few lines of code and without auxiliary networks or additional losses. Our theoretical analysis shows that DAI reshapes state visitation distributions to accelerate value function learning while preserving convergence guarantees. Empirical evaluations across MuJoCo continuous control tasks demonstrate that DAI improves early-stage performance by over 160\% on average and final performance by more than 50\%, with the Humanoid task showing a 4$\times$ improvement early on and a 2$\times$ gain at convergence. These results challenge the assumption that complex architectural modifications are necessary for sample-efficient reinforcement learning.