6/26/2026
Modern GPU Programming for MLSys Book
Educational resource on GPU programming for ML systems, covering attention kernels, LLM kernels, and optimization techniques.
6/26/2026
AI coding will be more expensive than human developers
Gartner analysis predicting AI coding costs will exceed developer salaries by 2028 due to token consumption and billing models.
6/26/2026
Show HN: Appaca – AI Workspace for Operators
HN show: Appaca, an AI workspace for operators using LLMs to generate code. Third iteration of no-code/LLM idea.
6/26/2026
Ludwig Spec Driven Development MCP
Astryx is an open-source design system built on React and StyleX, originally from Meta's internal tools. Explicitly agent-ready and customizable.
6/26/2026
Researcher got a death threat computing Rosetta Stone for Indus script [video]
Ludwig is a specification-driven development framework where prose-first markdown specs drive LLM code generation and verification. Targets Rust with cargo-based deterministic checks.
6/26/2026
Staying VIGILant: Mitigating Visual Laziness via Counterfactual Visual Alignment in MLLMs
Method addressing visual hallucinations in MLLMs by mitigating visual laziness through counterfactual alignment.
6/26/2026
Geometry-Aware MCTS for Extremal Problems in Combinatorial Geometry
MCTS approach with geometry awareness for solving extremal combinatorial geometry problems efficiently.
6/26/2026
Adaptive Evaluation of Out-of-Band Defenses Against Prompt Injection in LLM Agents
Defense mechanisms for tool-using LLM agents against indirect prompt injection via out-of-band policies.
6/26/2026
Evaluation-Strategy Gap in Fault Diagnosis of Deep Learning Programs
Study of performance gaps in deep learning fault diagnosis techniques across seen vs. unseen programs.
6/26/2026
Temporal Validity in Retrieval Memory: Eliminating Stale-Fact Errors for AI Agents over Evolving Knowledge
RAG system addressing temporal validity problem where stale facts cause AI agents to serve outdated information.
6/26/2026
Scientific discovery as meta-optimization: a combinatorial optimization case study
Formalizes scientific discovery as meta-optimization where LLMs simultaneously modify evaluation criteria and optimize within expanded theory spaces.
6/26/2026
MIRROR: Novelty-Constrained Memory-Guided MCTS Red-Teaming for Agentic RAG
Presents MIRROR, a memory-guided MCTS framework for systematic red-teaming of multimodal agentic RAG systems across multiple attack surfaces.
6/26/2026
Memory Depth, Not Memory Access: Selective Parametric Consolidation for Long-Running Language Agents
Studies memory depth and parametric consolidation in long-running language agents, introducing loop-drift protocol to test durable behavior adaptation.
6/26/2026
Jailbreaking for the Average Jane: Choosing Optimal Jailbreaks via Bandit Algorithms for Automatically Enhanced Queries
Examines automated jailbreak selection using bandit algorithms for non-expert malicious actors to craft effective LLM attacks.
6/26/2026
RolloutPipe: Overlapping Pipelined Rollout and Training in Disaggregated On-Policy LLM Reinforcement Learning
Describes RolloutPipe, a system for overlapping pipelined rollout and training in disaggregated RL architectures for LLM post-training with verifiable rewards.
6/26/2026
Semantic Early-Stopping for Iterative LLM Agent Loops
Proposes semantic early-stopping for multi-agent LLM loops using embedding similarity to halt when output meaning stops improving, reducing token waste.
6/26/2026
Parametric Open Source Games
Introduces parametric open-source games, a continuous model where players choose parameters converted to actions, with equilibrium existence results.
6/26/2026
DMuon: Efficient Distributed Muon Training with Near-Adam Overhead
Proposes DMuon, distributed training method for matrix-orthogonalization optimizers reducing communication overhead compared to element-wise optimization.
6/26/2026
Learning to Fold: prizewinning solution at LeHome Challenge 2026 (1st place online, 2nd offline)
Describes prizewinning bimanual garment folding solution combining vision-language-action policy with reinforcement learning loop for robotic manipulation.
6/26/2026
Forecasting With LLMs: Improved Generalization Through Feature Steering
Uses sparse autoencoders to inspect LLM internal states for forecasting tasks, identifying time-specific knowledge versus generalizable patterns.
6/26/2026
Hierarchical Muon: Tiled Newton-Schulz Updates for Efficient Muon Optimization
Introduces Hierarchical Muon (HiMuon), tiled Newton-Schulz optimization for efficient dense neural network training with reduced computational overhead.
6/26/2026
CARVE: Content-Aware Recurrent with Value Efficiency for Chunk-Parallel Linear Attention
Proposes CARVE, memory-aware recurrent architecture with content-aware gating for efficient chunk-parallel linear attention in sequence models.
6/26/2026
Ribbon: Scalable Approximation and Robust Uncertainty Quantification
Introduces Ribbon, scalable approximation to Dirichlet-reweighted bootstrap for efficient uncertainty quantification in high-dimensional models.
6/26/2026
When Does Combining Language Models Help? A Co-Failure Ceiling on Routing, Voting, and Mixture-of-Agents Across 67 Frontier Models
Analyzes fundamental ceiling on multi-model LLM systems (routing, voting, mixture-of-agents), showing accuracy limited by co-failure rate across 67 frontier models.
6/26/2026
Generative Models on Analog Hardware with Dynamics
Proposes methods for implementing generative models on analog hardware with physics-determined dynamics for low-power computation.
6/26/2026
Empowering GUI Agents via Autonomous Experience Exploration and Hindsight Experience Utilization for Task Planning
Introduces planning experience exploration for GUI agents using small open-source MLLMs, improving task planning and cross-website generalization.
6/26/2026
When are likely answers right? On Sequence Probability and Correctness in LLMs
Investigates alignment between sequence probability and correctness in LLMs, quantifying when higher likelihood corresponds to correct outputs.
6/26/2026
Theory of the Frequency Principle for General Deep Neural Networks
Theoretical analysis of frequency principle phenomenon showing DNNs learn target functions from low to high frequencies during training.
6/26/2026
Consistent Distributed Ranking of Generative Models via Kernel Distances
Proposes kernel distance method for ranking generative models in distributed settings based on output fidelity and diversity.
6/26/2026
Over-parameterization and Adversarial Robustness in Neural Networks: An Overview and Empirical Analysis
Studies relationship between over-parameterization in neural networks and adversarial robustness, analyzing vulnerability to adversarial examples.
6/26/2026
Byzantine-Robust Aggregation for Securing Decentralized Federated Learning
Byzantine-robust aggregation algorithms for secure decentralized federated learning without central servers.
6/26/2026
Chisme: Heterogeneity-Aware Gossip Learning
Chisme: gossip learning framework addressing heterogeneity in resource-constrained edge devices for privacy-preserving distributed learning.
6/26/2026
Mitigating Hallucinations via Inter-Layer Consistency Aggregation in Large Vision-Language Models
Training-free hallucination mitigation in vision-language models via inter-layer consistency aggregation during decoding.
6/26/2026
DMSC: Dynamic Multi-Scale Coordination Framework for Time Series Forecasting
DMSC: dynamic multi-scale coordination framework addressing static decomposition and inflexible fusion in time series forecasting.
6/26/2026
Learning to Select Maximum Clique Algorithms: From Traditional Machine Learning to a Dual-Channel Hybrid Neural Architecture
Hybrid neural architecture for instance-aware algorithm selection on maximum clique problems, combining ML and neural networks.
6/26/2026
Limited Reference, Reliable Generation: A Two-Component Framework for Tabular Data Generation in Low-Data Regimes
Two-component framework for tabular data generation in low-data regimes combining GANs and fine-tuned LLMs.
6/26/2026
Learning Robust Penetration Testing Policies under Partial Observability: A systematic evaluation
Reinforcement learning framework for penetration testing under partial observability, addressing belief state aggregation challenges.
6/26/2026
Rotary Position Encodings for Graphs
Rotary position encodings applied to graph-structured data using graph Laplacian spectrum for improved attention mechanisms.
6/26/2026
Eyes-on-Me: Scalable RAG Poisoning through Transferable Attention-Steering Attractors
Eyes-on-Me: scalable poisoning attack on RAG systems using reusable attention attractors to manipulate retrieval and generation.
6/26/2026
Computationally-efficient Graph Modeling with Refined Graph Random Features
GRFs++: refined graph random features with walk-stitching for efficient kernel computations on graph-structured data.
6/26/2026
Reinforcement Fine-Tuning of Flow-Matching Policies for Vision-Language-Action Models
Reinforcement fine-tuning of flow-matching Vision-Language-Action models through online interaction, improving performance beyond supervised data.
6/26/2026
Improved Bounds for Private and Robust Alignment
Theoretical bounds on private and robust alignment of language models under privacy constraints and adversarial corruption.
6/26/2026
Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs
Investigation of spurious rewards paradox in RLVR for LLMs: models bypass reasoning when trained with incorrect rewards, identified via perplexity divergence.
6/26/2026
Dual-Prototype Disentanglement: A Context-Aware Enhancement Framework for Time Series Forecasting
Dual-Prototype Disentanglement framework for context-aware time series forecasting by dynamically separating temporal patterns.
6/26/2026
DASH: Faster Shampoo via Batched Block Preconditioning and Efficient Inverse-Root Solvers
DASH optimizer: faster Shampoo implementation via batched block preconditioning and efficient inverse-root solvers for second-order optimization.
6/26/2026
Learning State-Tracking from Code Using Linear RNNs
Linear RNNs trained on code for state-tracking tasks, bridging sequence-to-sequence learning with next-token prediction in language models.
6/26/2026
SEMixer: Semantics Enhanced MLP-Mixer for Multiscale Mixing and Long-term Time Series Forecasting
SEMixer: MLP-Mixer architecture with random attention for multiscale time series forecasting, addressing redundancy and noise in temporal data alignment.
6/26/2026
Learning Long-Range Dependencies with Temporal Predictive Coding
Temporal Predictive Coding improved for learning long-range dependencies in recurrent systems on neuromorphic hardware through better credit assignment mechanisms.
6/26/2026
BrepCoder: A Unified Multimodal Large Language Model for Multi-task B-rep Reasoning
BrepCoder is a multimodal LLM for CAD tasks using B-rep format instead of point clouds/images, enabling unified multi-task reasoning without task-specific modifications.
6/26/2026
Training-Free Generation of Protein Sequences from Small Family Alignments via Stochastic Attention
Training-free protein sequence generation using stochastic attention on small sequence alignments.