The Archive

Meta settlement highlights structural challenges in content regulation frameworks for large tech platforms.

Progress-augmented curriculum learning method for multi-task RL post-training of LLMs balancing task update magnitude with reward gains.

GNN-based beamforming design for mmWave cell-free massive MIMO systems using sub-6 GHz channel state information.

Learning-based congestion-aware route scheduling for semiconductor fab material handling using transport-network-aware dynamic representation.

ScienceArena: olympiad-style benchmark from IPhO, IChO, IBO, USAPhO, USNCO 2025–2026 with process-credit rubrics to address contamination and saturation in frontier LLM evaluation.

TSExplorer: cross-platform annotation tool for time-series data with interactive 2D visualizations and exploratory workflows.

Trajectory-Initialized Neural Double Q-Routing for overhead hoist transport systems in semiconductor fabs using adaptive routing under resource contention.

Lot Machine: multimodal pipeline to extract structured metadata from historical German auction catalogs for provenance research.

UtilMem: benchmark with 1,717 instances measuring evidence integration across long-term conversational memory, not pointwise recall.

Survey on tensor decomposition methods for LLMs covering tokenization, embeddings, pre-training, adaptation, inference, compression, and interpretability.

Anytime-valid statistical monitoring using conformal martingales for deployed ML systems; examines exchangeability assumptions in real forecast streams.

CM2: multi-agent MLLM framework integrating perception, RAG, networked reasoning, and gated fusion for multimodal cultural reasoning tasks.

Topo² framework: persistent-homology measurement separating memorization from generalization as causally independent geometric channels in noisy-label learning.

Computational analysis of sexism in Hansard 1803–2005: LLM classification of gender perspectives and ambivalent sexism patterns across 6,531 parliamentary speeches.

VisER detects object hallucination in vision-language models by distinguishing visual evidence from text-prefix bias using training-free internal signals.

Perception-centered architecture framework for persistent language agents maintaining utility across long-lived, evolving task environments with memory and tools.

ImageEval 2026 shared task benchmarks Arabic multimodal capabilities: spoken VQA, hallucination detection, and culturally-grounded text-to-image generation.

ToxLens applies graph learning to molecular toxicity prediction with leakage-aware splitting and calibrated uncertainty across 11 toxicity endpoints.

Hi-Q hierarchically refines queries for multi-hop QA by detecting when evidence supports a claim versus when further query refinement is needed.

CHASE simulation studies ecosystem homogenization when content creators optimize for LLM ranking signals, showing rank-citation correlation in generated responses.

Multilingual study of solvability detection in LLMs extends ReliableMath to French and Greek, analyzing whether failures stem from belief or language-specific expression.

Study shows high-resource languages more reliably elicit mathematical reasoning in LLMs; proposes feature transfer from HRLs to activate latent LRL computations.

RetroGen enables self-improving process supervision for open-ended long-form generation by retrospectively analyzing abundant final artifacts instead of rare trajectories.

Code-CoT introduces credit-addressable reasoning for multimodal geometry: assigns learning credit at semantic unit level rather than trajectory-level to improve VLM deduction.

OpenAI endorses California SB 1119, legislation for age-appropriate AI safeguards targeting teen users.

Polimill deploys OpenAI GPT models and Codex to enable Japanese municipalities to search administrative knowledge and accelerate development.

ChatGPT Ads reaches $1B annualized revenue run rate with global expansion, emphasizing free/affordable access.

The U.S. is shutting out more foreign-made drones and robots. China’s scale means the global competition may simply move elsewhere.

Simon Willison breaks down OpenAI's ChatGPT Work product architecture: cloud-based version and desktop app with local file/program access.

Elon Musk says a secretive new SpaceX foundry will let him cast his own turbine blades and get gas power online 18 months faster than anyone else — but it's a bet on a fuel source that's already triggering lawsuits and health studies everywhere his (and others') turbines have gone in.