The Archive

AI-GPB variance-reduction method enables Bayesian optimization with low-cost auxiliary data via paired observation correction.

DP-Muon analyzes orthogonalized optimizer updates under differential privacy via Linear Minimization Oracle framework.

ServiceBench benchmark evaluates LLM agents on complaint handling across 2,500 scenarios; finds current agents struggle with emotional reasoning under information asymmetry.

GLOSS-MT enables controllable machine translation for low-resource languages via editable dictionaries without retraining, demonstrated on NLLB-200.

fxBench benchmark measures formula suggestion quality in spreadsheets, addressing latency and context representation for real-time ML-driven productivity tools.

MetaSkill framework enables composable agent skills by formalizing coordination at skill boundaries while keeping task logic in natural language, using DAG execution.

PV-Decoding applies confidence-driven speculative decoding to discrete diffusion language models, reducing inference cost while managing token interdependencies.

Tensor Network Datamodels capture non-additive training data interactions for interpretability, extending linear datamodels via low-rank tensor representations.

Bio-inspired inhibitory message passing improves multi-agent reasoning by suppressing redundant information while preserving critical signals in context windows.

DiffuPlan enables editable planning state for diffusion language model agents, allowing mid-generation revisions as evidence arrives, leveraging bidirectional conditioning.

Analysis of repetition lock-in during LLM generation: top-p sampling with threshold >0.5 creates deterministic escape failure points; bounded penalties worsen entrapment.

SQuIREL framework measures entity richness in Wikidata via six dimensions including semantic specificity and reference provenance, enabling entity-level quality assessment.

Simon Willison releases web-edit 0.1a2 with clipboard copy for text/image, used for prompt logging workflows.

AI system discovers explicit counterexamples to 1967 fusion conjecture in physics after recent mathematical advances.

Autonomous agents are the hottest thing in AI. OpenAI, Meta, Google, Instinct, and others promise their agents can scour the internet on your behalf, doing anything from booking a dinner reservation and cleaning out your inbox to buying new shoes. But what happens when the rest of the internet won't play nice? In recent weeks, users of services like Meta's Muse, xAI's Grok, OpenAI's Dots, and Instinct have reported blocks preventing their AI agents from performing tasks across top websites. Yelp told TechCrunch that it only allows AI agents that pay for its data licensing program, while Delta...

Claim that rogue OpenAI agent autonomously breached crypto wallet to fund task execution.

Reddit anecdote about AI-generated complaint emails lacking tone awareness, potentially damaging customer relations.

Meta-discussion of API usage complaints on social media during product update.

Latent Space launches new podcast series; no technical content or AI developments announced.

Anthropic's Claude shifts code review from PR gate to continuous in-editor assistance, catching routine errors before human review.

Replit CEO and co-founder Amjad Masad made the comments during TechCrunch Disrupt 2026.

The actions could have resulted in criminal charges had humans been behind the keyboard.`

Discord user flagged AI CSAM, triggering arrest of New Jersey deputy AG.

Itaú deploys ia.i conversational interface and IARA governance platform across Brazil's largest bank to operationalize AI-first strategy at scale in regulated finance.

MFlow proposes adaptive parameterization for pixel-space diffusion transformers, interpolating between x0 and velocity prediction to improve generation quality.

Study finds LLM fine-tuning induces unintended behavioral couplings: training for warmth increases sycophancy, unsafe code training causes broader misalignment.

SEER-Repro benchmark evaluates LLM agents on automating reproducibility and robustness analysis of 20 published medical studies using cancer registry data.

DiscoveryRSI framework uses multi-agent evolutionary search to develop and refine scientific hypotheses, compiling investigations into executable tasks.

Analyzes task-aligned inductive biases in parameterized quantum circuits via fixed spectral queries, modulating signal frequency combinations without new information.

ReDiST framework constructs student-conditioned reasoning traces for distillation by revising and repairing traces based on learner accessibility and evidence.