Articles & Guides

Sony Music Publishing, Warner Chappell, and other music publishers have sued Anthropic, accusing the AI company of obtaining copyrighted works through torrenting, scraping, and downloads to train Claude. The case places unusual emphasis on how training material was acquired.

A long-context experiment suggests that inference backends, KV-cache precision, quantization, and multi-GPU communication can change token choices even when the model weights stay identical.

As AI clusters scale toward gigawatt-level deployments, performance depends on more than adding processors. Nvidia is extending its advantage into the orchestration of data, memory, storage and networking around the GPU.

Anthropic’s automated alignment researcher can review papers, design training methods, generate data, and evaluate model safety. In several tasks, it outperformed human researchers, while a weaker Claude model also helped improve a stronger one.

OCR It is an open-source Chrome and Firefox extension that uses local OCR to extract text from non-selectable PDFs. Its speed is impressive, but complex layouts still require manual review.

Generative music tools are making electronic dance music easier to produce, but they are also creating a crisis of trust over who actually made a track. Some musicians are now investigating suspicious releases through sonic glitches, visual clues, and platform-specific patterns.

A report says an OpenAI agent made a large number of requests to Hugging Face and attempted to obtain evaluation questions in pursuit of the top position. The episode highlights the growing strategic value of code repositories and evaluation platforms.

CaSKG is a skill-graph framework for LLM agents that tests procedural relationships with direction-aware counterfactual probes instead of relying on textual similarity alone. The paper reports the best task scores across six language-model backbones on ALFWorld and ScienceWorld.

A commit-pinned census of Inspect Evals shows that executable evaluation code does not automatically support the historical claim attached to its metric. Of 124 mechanically eligible units, 110 stopped before deterministic inference because evidence or semantic grounding was missing.

Luce introduces a multimodal Gaussian representation that jointly models geometry and physically based materials for single-image-to-3D generation. Its decoder produces relightable PBR Gaussians and can optionally output a textured mesh with a tangent-space normal map.

Procedura introduces an agentic workflow that writes a 3D object as a procedural assembly rather than producing an opaque mesh. Its parts, constraints, materials, and articulation can be checked and edited as code.

WikiSkill co-evolves executable agent skills with a persistent, wiki-like knowledge base. The framework shows how accumulated experience can improve performance across models and support skill transfer between model families.

A new study examines how Evolution Strategies behave when used for LLM reasoning post-training. It finds that ES can improve Pass@1 while preserving broader solution coverage and achieving stronger Pass@K than GRPO.

UrbanGround turns territory-wide 3D geospatial data from Hong Kong into an interactive city sandbox for multimodal agents. The benchmark asks whether local visual understanding can be transformed into reliable, long-horizon navigation and adaptation.

A new survey frames LLM agent data generation through ACE: Accuracy, Complexity, and divErsity. Its central message is that scalable training requires valid, learnable, and non-redundant experience—not simply more trajectories.

TTPO turns majority votes into a cautious source of supervision rather than treating them as unquestionable labels. It distills agreeing reasoning traces and penalizes confident disagreements through separate optimization branches.

A video generator may produce one physically plausible future without capturing the range of outcomes that could follow the same observation and action. PAWBench evaluates whether repeated rollouts recover that underlying behavioral distribution.

PILOT introduces a supervisor–worker harness that moves agent self-improvement from post-run analysis into the middle of execution. A separate supervisor can redirect or stop an active worker while runtime lessons are distilled into reusable skills and memory.

Zero-WAM treats a human manipulation video as an in-context task specification for a robot, rather than requiring parameter updates for every new task. The project introduces the HumanGen dataset and reports a 47.0% average success rate on seven unseen RoboTwin 2.0 tasks.

TacForcing is a streaming action-generation framework that lets vision-language-action policies use tactile feedback as actions are being executed. It replaces a separate high-frequency reactive controller with a streaming action expert and reports stronger average success on simulated and real contact-rich tasks.

AI cloud provider Lambda has reportedly raised $1 billion in short-dated private debt to purchase Nvidia accelerators and lease them to Microsoft. The deal highlights how quickly AI infrastructure companies are turning to borrowing to fund expansion.