The Archive

Stratechery announces Stripe sponsorship, explaining implications for editorial independence and content direction.

Orbitol generates crystal structures via Crystalite diffusion Transformer with Wyckoff-orbit tokens, enabling direct control over space-group symmetry constraints.

Circuits of LLM Classifiers organize multiple LLM predictors into adaptive query sequences to maximize accuracy within budget, treating models as noisy measurement instruments.

HDC-FL clusters federated learning clients via classifier heads to personalize shared encoder training without explicit communication overhead.

Models multi-agent coordination as graph-structured block optimization, proving optimal fractional coloring preserves agent edits' combined gains.

Integrates charts as epistemic actions inside ReAct-loop data agents to improve reasoning over structured data beyond text-only analysis.

LAW uses pure self-conditioning in diffusion language models, outperforming discrete/continuous baselines on Sudoku and maze reasoning tasks.

GoiBERT distills compact Vietnamese encoders (26M–49M params) via two-stage knowledge transfer from larger multilingual teachers.

Looped Uniform Diffusion Language Models improve reasoning via selective layer recycling during text denoising, reducing training cost vs. unlooped UDLMs.

Blacksmith, a platform that lets companies build and test AI code more effectively, just announced a $66 million Series C round led by Sequoia that values the company at $1 billion.

Evaluates frontier AI models (Claude, GPT-4, others) on LIBERO-10 robot manipulation via closed-loop visual control without task-specific training.

SWE-DIVE benchmarks coding agents via multi-turn interaction with dynamic user simulation beyond single-prompt SWE-bench evaluation.

Controlled arithmetic-rule experiments separate rule transfer, joint reliability, and information use in decision model accuracy analysis.

SP-HLM couples sparse-pose interface with learned motion priors and vision-language-action models for humanoid loco-manipulation control.

HGM-Dual improves agent harness optimization for long-horizon planning by incorporating lineage-aware guidance, extending self-improvement beyond coding tasks.

Behavioral Localization identifies trajectory positions where soft skills have concentrated impact, reducing computational costs for tool-using agents.

AT&T deploys OpenAI models in LegalEdge to automate legal intake and scale internal legal operations.

Study reveals silent information loss in document ingestion layers for enterprise agents, causing confident errors from formatting data loss in PDF/Word conversion.

Stationarity constraints improve machine-learned exchange-correlation functionals for quantum chemistry, achieving better energies and densities than SCAN.

ExGrad improves adversarial training by explicitly subtracting clean gradients, achieving adaptive margin control on CIFAR-10.

GuideEvolve framework adds expert guidance and 3Ts mechanisms (think fast/slow, throwback, termination) to self-evolving agents.

Code Comprehensor uses dependency callgraphs and hierarchical documentation for large repository code understanding by LLMs.

Agent Behavior Maps profile LLM-harness combinations via action log-likelihoods on reference trajectories, enabling performance prediction across 100 LLMs.

Dreaming Learning mitigates model collapse by training on high-temperature synthetic samples that preserve subdominant eigenvalue spectrum diversity.

Sparse autoencoders convert opaque neural AI-text detectors into interpretable classifiers operating on human-readable concepts for high-stakes deployment.

Open-source telecom alarm root-cause generator using real Internet topologies and RFC 8632 standards; dataset benchmarking contribution for infrastructure diagnostics.

EPIC reformulates PPO-style RL for LLM policy training via exact-KL prefix importance-ratio clipping, addressing policy-context mismatch in long-response generation.

HiddenLens, a six-layer BERT-LLM pipeline for detecting implicit assumptions in software requirements to reduce costly late-stage resolution.

MHG-Mem, a multimodal hypergraph memory for long-term agent recall that preserves visual evidence while tracking entity state evolution across sessions.