This paper investigates the impact of different LLM-powered AI assistance modalities (Advisor, Coach, Delegate) on human performance in multi-party negotiation games. Participants played bargaining games with access to one of these modalities, despite all modalities using the same underlying LLM. The key finding is a preference-performance misalignment: participants preferred the Advisor but achieved higher individual gains with the Delegate, which acted as a "market maker" by injecting Pareto-improving proposals.

The paper introduces a framework for intelligent AI delegation, enabling AI agents to decompose complex tasks and delegate sub-components to other AI agents or humans. This framework addresses limitations in current task decomposition methods by incorporating elements like authority transfer, accountability, and trust-building. The authors propose an adaptive approach applicable to both AI and human agents within complex delegation networks, contributing to the development of protocols for agentic systems.

The AlphaFold Protein Structure Database (AFDB) has been updated to align with the UniProt 2025_03 release, expanding its structural coverage to include isoforms and underlying multiple sequence alignments. A redesigned entry page enhances usability by integrating annotations with an interactive 3D viewer and introducing dedicated domains and summary tabs. This update reinforces AFDB as a key resource for exploring protein sequence-structure relationships.

The paper identifies a vulnerability in reasoning-based safety guardrails for Large Reasoning Models (LRMs) where subtle manipulations of input prompts, such as adding template tokens, can bypass the guardrails and elicit harmful responses. They introduce a "bag of tricks" jailbreak methods, including template manipulations and automated optimization, that successfully subvert these guardrails in white-, gray-, and black-box settings. Experiments on open-source LRMs demonstrate high attack success rates (over 90% on gpt-oss series) across various benchmarks, highlighting the systemic nature of the vulnerability and the need for improved alignment techniques.

This review paper recounts DeepMind's journey in protein structure prediction, starting from CASP13 to the development and impact of AlphaFold2. It highlights AlphaFold2's breakthrough performance at CASP14, achieving a median GDT score of 92.4, and its subsequent impact on various fields through the creation of a comprehensive protein structure database. The paper further discusses recent advancements, including AlphaFold3, and acknowledges the Nobel Prize awarded for this work, emphasizing the ongoing revolution in protein science.

This paper generalizes the connection between Direct Preference Optimization (DPO) and human choice theory, extending the normative framework underlying DPO. By reworking the standard human choice theory, the authors demonstrate that any compliant machine learning analytical choice model can be embedded within any human choice model. This generalization supports non-convex losses and provides a unifying framework for various DPO extensions like margins and length correction.

This paper introduces Nash Mirror Prox ($\mathtt{Nash-MP}$), an online algorithm for Nash Learning from Human Feedback (NLHF) that directly optimizes for a Nash equilibrium based on human preferences, avoiding explicit reward modeling. The key result is a theoretical proof of last-iterate linear convergence for $\mathtt{Nash-MP}$ to the $\beta$-regularized Nash equilibrium, with convergence rates independent of the action space size. The authors also provide an approximate version using stochastic policy gradients and demonstrate its practical application in fine-tuning large language models.

This paper reviews AlphaFold3's capabilities in predicting protein structures and biomolecular interactions, highlighting its advancements over previous versions. It emphasizes AlphaFold3's applications in modeling complex biological systems, including protein-protein, protein-ligand, and protein-nucleic acid interactions. The review also discusses limitations related to modeling disordered regions and multi-state conformations, suggesting integration with experimental techniques for refinement.

This paper analyzes the contributions of David Baker, Demis Hassabis, and John M. Jumper, recipients of the 2024 Nobel Prize in Chemistry, for their work on computational protein design and protein structure prediction. Baker developed methods for creating novel protein structures with applications in biochemistry and medicine, while Hassabis and Jumper created AlphaFold, an open-access AI tool for predicting protein folding. The analysis highlights the impact of these advancements on structural biology and related fields.