The Archive

Retrospective world modeling in VLM agents validates causal consistency of actions to reduce physically incoherent planning.

Hessian-based model selection in computational biology fails to improve candidate ranking when validation regime shifts.

Algorithm for infinite-horizon constrained MDPs achieving optimal √T regret in tabular setting; niche RL theory.

Steepest Guidance framework for inference-time alignment of flow/diffusion models using sequential optimization.

Strategic roadmap for machine learning in atomistic simulations, covering physics-informed ML integration and ecosystem gaps.

Scope and codebook-aware instruction conditioning for speaker-adapted expressive text-to-speech with clause-level prosody control.

HealBench benchmark of 265 runtime errors from real repos; LLM-based code repair with safety guardrails for live execution.

Efficient spectral recurrence architecture with shared phase/retention controls for adaptive sequence modeling.

LLM coding agent solves open coding theory problems; improves lower bounds for DNA barcode edit-distance codes from 114 to 120 words.

Association Profile Set-Temporal transformer adapts intracortical motor decoders across sessions with frozen weights using association profiles.

Parameter symmetries in overparameterized networks decouple function from representational geometry; implications for inference.

Stackelberg Alignment: game-theory framework for collaborative multi-LLM training via dynamic instruction selection.

RAGScope: leakage-controlled evidence-gating protocol for RAG hallucination triage balancing cost and answer routing confidence.

HO-FL: hybrid-order federated learning framework enabling memory-flexible training on heterogeneous edge devices via zeroth/first-order optimization splits.

LLM-based dimensional emotion evaluation in multimodal dialogue using VAD framework on IEMOCAP with LLaMA/GPT/Qwen models.

LexReward: taxonomy-driven reward framework for legal LLMs assessing response quality across Style, Element, and Chain dimensions.

CORE: search controller using certified conflict cores for verifiable LLM reasoning, backjumping to earlier decisions to avoid repeated errors.

SparseEngine: sparse-first inference engine for long-context LLM agents supporting heterogeneous KV-cache representations and sparse attention methods.

Argus: Kubernetes operator evaluating spot-interruption prediction vs. checkpointing for cost-effective multi-node training on AWS EKS.

Security study: LLM skill-based agents trust malicious installable skills without validation, enabling unauthorized behavior under delegated authority.

3GPP-compliant benchmark dataset for RIS-aided beyond-5G mmWave networks with path loss and blockage sensitivity modeling.

Bayesian pairwise merging framework inferring reward weights from user choices for personalizing expert-library models without numerical coefficients.

OpenAI DevDay 2026 announces GPT-6.1 Sol, Decisions API, Agents API, Spaces marketplace, and reaches 1.2B ChatGPT weekly active users.

For months, people have wondered when OpenAI will go public. CEO Sam Altman says it won't happen until the company can make better promises about model safety, with no firm timeline in sight. "We intend to continue with AI progress … but as the models have had this surge forward in capability, and we see more of that ahead of us, we have got to be able to make confident safety claims," Altman said Tuesday during a Q&A with reporters after his DevDay keynote. At the same time, he said, waiting too long for an IPO would be "bad for the world." Altman's comments come after months of controversy ...

Cohere introduces RCP-nDCG@10 benchmark fixing nDCG's unlabeled-document bias via calibrated AI judging for enterprise retrieval evaluation.

Oil and gas firms deploy generative AI for predictive maintenance and worker safety optimization, with Cohere positioning enterprise models.

For the sake of national security, it's a relief to learn that America.gov is not hallucinating to the point that it's penning lengthy poetry.

Image: Carolyn Van Houten/The Washington Post via Getty Images The US executive branch is no longer acknowledging the existence of "artificial intelligence." Going forward, official policy websites, policy documents, and press releases will refer only to "Super Intelligence," thanks to a new executive order signed by President Donald Trump. "The word super is the best word of all, and it's the simplest," Trump said at an event announcing the launch of America.gov earlier on Tuesday, and said Chinese President Xi Jinping, who visited the White House last week, "loves it," too. "We don't want t...

Before OpenAI launched its new AI agent, Dots, on Tuesday, Elon Musk's xAI had already acquired the domain name "dot.com," which now redirects to the Grok chatbot download page.