The Archive

Human-written hallucination samples improve VLM benchmark stability across 4 languages vs. model-generated negatives.

LLMs exhibit medical sycophancy—abandoning correct answers under user pushback—driven by conversational factors, not fixed model properties.

KING: Graph neural network for cross-embodiment kinematic models in legged and wheeled robots via proprioceptive geometry learning.

Cloud-ScPO derives preference pairs from LLM hidden-state geometry to improve mathematical reasoning without human annotation.

MedUPS: benchmark of 21,874 mid-stream clinical decision points from 5,535 cases for LLM diagnostic assistance on uncommon medical scenarios.

Attribute-level unlearning for MLLMs enables fine-grained removal of sensitive identity information while preserving model utility.

FusedBFN: Bayesian flow network for dual-target 3D molecular generation via joint feature fusion from two protein targets.

Hierarchical Solomonoff Induction extends ideal sequence prediction via hyperpriors over Solomonoff priors for dataset conditioning.

Capability-taxonomy-driven pipeline for curating regression eval sets across multi-customer agent-extensibility platforms under query budgets.

Agentic Technical Debt (AgTD): framework mapping root causes of technical debt in autonomous multi-agent systems with persistent memory.

LLMs as examiners silently omit valid answers when authoring test sets; greedy one-shot generation fails to enumerate complete solution spaces.

Method to repair synthetic training data for zero-shot image captioning by fixing entity misalignment at fine-grained level.

SCHEDBench: benchmark with 1,132 scheduling instances to evaluate LLM constraint faithfulness under natural-language variation.

FDR-controlled feature selection framework for grouped features in sequential and neural models using block-level mirror statistics.

Contrastive pretraining framework for single-cell transcriptomic foundation models to learn cell representations beyond gene reconstruction.

Transfer learning approaches for Named Entity Recognition on small, unlabelled datasets across multiple domains.

Microsoft-led open letter signed by 235 AI companies including NVIDIA and OpenAI argues against US government restrictions on open-weight models on safety grounds.

Two-sided audit framework for self-improving AI-for-science systems to distinguish real gains from search artifacts and oracle drift.

Simon Willison's June 2026 newsletter roundup covering model releases (GPT-5.6, Claude Opus 5, DeepSeek-V4), open letters, and accidental cyberattacks by OpenAI and Anthropic test models.

Study of LLM persona panels (GPT-4.1) as synthetic research participants, showing marginal-check validity depends on prompt and uncertainty assumptions.

Quantile Coupling Flow Matching: lightweight coupling method for flow-matching generative models with subquadratic cost in batch size.

Search-GRT: RL method to train LLM search agents for multi-hop QA with guided retrieval to reduce sparse-reward training issues.

Zero-query jailbreaks for text-to-image systems via filter-generator discrepancy; transfer-based attack requiring no target queries.

PROGRESS trains search-augmented LLM agents using coverage-guided RL rewards to improve query decomposition over outcome-only supervision.

TrajWiki proposes trajectory-based external memory for long-horizon dialogue agents with traceable, updatable, diagnostically transparent storage.

CryptoProver synthesizes proofs for cryptographic libraries by inferring internal specifications from API contracts, verified on curve25519-dalek and chacha20.

PMMC compiles multimodal memory at consolidation time for LVLM agents to preserve image-text binding and temporal updates without query-time overhead.

Essay critiquing institutional dismissal of AI anthropomorphism as user error, arguing for epistemic pluralism in human-AI interaction interpretation.

SHAP-based interpretable ML framework for predicting asphalt concrete splitting strength using TabPFN, XGBoost, and classical ML baselines.

Data-driven elastic-net SVM with learnable simplex-constrained weights over candidate pinball losses, with empirical oracle inequality bounds.