New Model Releases A

RIKEN, Japan's national research institute, has decided to name its supercomputer dedicated to AI for Science 'Rikyu' (理究). The announcement also explains the origin and reasoning behind the chosen name.

GMO AI & Robotics, a GMO Internet Group subsidiary, has signed an official distributor agreement with China's Unitree Robotics for the Japanese market. It will offer end-to-end support, from deployment to maintenance, aiming to accelerate humanoid robot adoption across Japan.

OpenAI has added a new 'Record & Replay' feature to its Codex coding agent. Users record on-screen operations, and the AI then reproduces those steps to carry out the task automatically, according to ITmedia.

Gartner reports that US state authorities imposed about $3.425 billion in privacy-law violation fines in 2025, exceeding the combined total of the previous five years. It expects enforcement to keep accelerating through 2028, urging CISOs to reconsider their privacy and compliance posture.

OpenAI's Japan arm announced it has started testing ad displays within ChatGPT in Japan. The article explains how the ads appear and how users can hide them.

Simon Willison introduced Datasette Apps, letting developers host custom HTML/JS applications inside a Datasette instance. The apps can read Datasette's databases, enabling lightweight, data-backed web apps served directly from the data exploration tool itself.

UNIEGO is a unified egocentric video encoder trained via a hierarchical multi-teacher distillation framework. Representation-specific proxy models translate knowledge from teachers spanning multiple viewpoints, modalities, and foundation models into a single egocentric space, while remaining deployable from egocentric video alone.

The paper introduces 'privacy via predictability,' a fine-grained privacy framework that explicitly incorporates an attacker's core prior knowledge. It aims to ease the costly privacy-accuracy tradeoff imposed by the worst-case guarantees of differential privacy.

The paper studies multi-task Bayesian in-context learning, using in-context learning to perform Bayesian predictive inference across tasks. It targets the intractability of exact inference and the cost or restrictiveness of scalable approximations, aiming for uncertainty quantification and data efficiency.

The paper introduces Execution-State Capsules, a graph-bound mechanism to checkpoint and restore execution state for low-latency, small-batch, on-device physical-AI serving. It targets scenarios beyond the high-throughput, high-concurrency regime that paged or radix KV caches mainly serve.

Policy-adherent tool-calling agents in customer-service domains must track task state across turns while following rules. LedgerAgent introduces structured state to help such agents stay consistent and policy-compliant.

StylisticBias investigates the visual cues that shape how multimodal large language models judge people. The study finds that a small set of human visual cues drives most of the social biases exhibited by MLLMs, which are increasingly deployed in consequential settings.

Neurosymbolic systems such as DeepProbLog combine neural perception with probabilistic logic, but standard inference has limits. DeepSWIP introduces quotient-WMC counterfactuals to enable counterfactual reasoning in neural probabilistic logic programs.

Autonomous agents are increasingly wired into cloud, deployment, and data-control workflows, straining production security. This work proposes sovereign execution brokers that enforce certificate-bound authority within agentic control planes.

Flow-matching text-to-speech achieves strong zero-shot quality but stays static after deployment. FlowEdit uses associative memory to enable lifelong pronunciation adaptation without full retraining.

LiveCodeBench has become a widely adopted benchmark for evaluating large language models on code. Multi-LCB extends it to multiple programming languages to assess multilingual code generation.

The paper presents Probe-and-Refine, a method for tuning the repository guidance (such as AGENTS.md files) that LLM-based coding agents rely on. It targets the higher-level operational knowledge—file layout, test workflows, and error-prone patterns—that is not contained in the code itself.

Securing AI agents that operate in complex digital environments has become critical, motivating runtime verification. This paper presents an efficient and sound probabilistic verification approach for AI agents.

Style-content dual-reference generation aims to synthesize an image that preserves structure while adopting a reference style. FreeStyle leverages community LoRA mining to give free control over style and content.

It is unclear whether LLMs that score well on vulnerability benchmarks truly reason about security or merely pattern-match. This work diagnoses the limits of fine-tuning LLMs for vulnerability detection in systems software.

When large language models act as evaluators in multi-agent systems, their systematic evaluation biases can spread through the system. This work analyzes how such evaluator bias propagates across agents.

The paper proposes a hierarchical recovery mechanism for cross-device agent systems, moving beyond coarse-grained global replanning. It targets real-world computer-use tasks that span multiple applications and devices and must coordinate heterogeneous environments under dynamic runtime failures.

This paper develops a general framework for analyzing multi-agent systems with feedback loops between agents, as well as general many-body systems, and characterizes their optimal order.

OpenAI introduced new usage analytics and updated spend controls for ChatGPT Enterprise, helping organizations track and manage AI costs while scaling with confidence. Admins gain visibility into per-team consumption and can set limits to optimize spend.

The paper studies how to train spatially grounded vision-language models for radiology without manual spatial annotations. It introduces RefRad2D, a large-scale bilingual (German/English) dataset of 1.2M CT and MR image-text pairs derived from clinical practice, with VQA and spatial grounding subsets.

The paper presents a mechanistic study of representation retention in continual learning, using a controlled toy-world framework to make the drivers of forgetting observable and testable. It examines how sparsity and superposition relate to forgetting, isolating mechanisms that real datasets usually entangle.

The paper develops neural network surrogates with uncertainty quantification for inverse problems in partial differential equations. It targets the inference of unknown model parameters from noisy or incomplete observations, where traditional numerical methods are costly, particularly in Bayesian settings.

The paper proposes Pseudo-Feature Padding, a lightweight defense against false data injection attacks in power grids. It targets the vulnerability of deep neural network detectors in cyber-physical systems, where attackers can craft inputs to evade detection during critical operations.

The paper improves Direct Advantage Estimation (DAE) for scalable and sample-efficient deep reinforcement learning. It addresses DAE's reliance on full environment observability and the computational overhead of modeling transition probabilities, which limit its use in realistic settings.

Data videos combine dynamic charts, voice narration, and synchronized animation to convey insights. DataMagic automatically transforms tabular data into such data-insight videos.