6/4/2026
Ask HN: Why is it still so hard for LLMs to query NoSQL databases?
Discussion on why LLMs struggle with NoSQL database querying compared to SQL. Explores lack of unified interface across MongoDB, DynamoDB, Cassandra, Redis, Neo4j.
6/4/2026
Show HN: Boxes.dev: ditch localhost; run Claude Code and Codex in the cloud
Boxes.dev: cloud-based agentic dev environment enabling Claude Code and Codex agents to run in isolated cloud containers instead of localhost.
6/4/2026
How Endava is redesigning software delivery around AI agents
Endava redesigns software delivery workflows around AI agents, reshaping enterprise collaboration and accelerating project delivery.
6/4/2026
Dreaming: Better memory for a more helpful ChatGPT
OpenAI improves ChatGPT memory synthesis system to handle staleness, correctness, and scalability across millions of users.
6/4/2026
Mid-Think: Training-Free Intermediate-Budget Reasoning via Token-Level Triggers
Training-free method revealing token-level triggers control reasoning behavior in hybrid LLMs, enabling intermediate-budget reasoning without fine-tuning.
6/4/2026
Optimal Transport under Group Fairness Constraints
Novel fairness constraints for optimal transport matching algorithms ensuring group-level probability targets.
6/4/2026
Outcome-Grounded Advantage Reshaping for Fine-Grained Credit Assignment in Mathematical Reasoning
Outcome-grounded Advantage Reshaping (OAR) improves credit assignment in GRPO for mathematical reasoning tasks by fine-grained token-level rewards.
6/4/2026
Learning collision operators from plasma phase space data using differentiable simulators
Method for inferring collision operators from plasma phase space data using differentiable simulators and gradient-based optimization.
6/4/2026
Success Conditioning as Policy Improvement: The Optimization Problem Solved by Imitating Success
Theoretical analysis proving success conditioning (rejection sampling, goal-conditioned RL, Decision Transformers) solves a specific optimization problem for policy improvement.
6/4/2026
On Forgetting and Stability of Score-based Generative models
Stability analysis and sampling error bounds for score-based generative models using Markov chain properties.
6/4/2026
Can Large Language Models Generalize Procedures Across Representations?
LLMs show limited generalization of procedures across code, graph, and natural language representations.
6/4/2026
The Cost of Learning Under Multiple Change Points
Online learning analysis for multiple change point environments with detection and regret bounds.
6/4/2026
Drift-Diffusion Matching: Embedding dynamics in latent manifolds of asymmetric neural networks
Framework for embedding asymmetric RNN dynamics in latent manifolds extending classical symmetric models.
6/4/2026
Path-conditioned training: a principled way to rescale ReLU neural networks
Path-conditioned training exploits rescaling symmetries in ReLU networks improving training dynamics.
6/4/2026
Plug-and-Play Diffusion Meets ADMM: Dual-Variable Coupling for Robust Medical Image Reconstruction
Plug-and-play diffusion with ADMM solver improves medical image reconstruction using dual-variable coupling.
6/4/2026
Evaluating Zero-Shot and One-Shot Adaptation of Small Language Models in Leader-Follower Interaction
Small language models adapted via zero/one-shot learning for role classification in human-robot interaction.
6/4/2026
Beyond Pixel Histories: World Models with Persistent 3D State
Interactive world models with persistent 3D state representation enable spatially consistent video generation.
6/4/2026
Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning
Constraining safety-critical tokens during fine-tuning preserves LLM alignment on benign downstream tasks.
6/4/2026
Analysis-Driven Procedural Generation of an Engine Sound Dataset with Embedded Control Annotations
Dataset generation for engine sound synthesis with procedural modeling and control annotations.
6/4/2026
Safety Under Scaffolding: How Evaluation Conditions Shape Measured Safety
Safety evaluation scores differ across deployment configurations (ReAct, multi-agent, map-reduce) versus direct API.
6/4/2026
Quantum entanglement provides a competitive advantage in adversarial games
Quantum entanglement provides competitive advantage in zero-sum adversarial reinforcement learning games.
6/4/2026
Revisiting Model Stitching In the Foundation Model Era
Vision foundation models remain stitchable across different objectives and data despite architectural differences.
6/4/2026
On-the-fly Repulsion in the Contextual Space for Rich Diversity in Diffusion Transformers
On-the-fly repulsion mechanism increases diversity in text-to-image diffusion transformers reducing typicality bias.
6/4/2026
Gradient estimators for parameter inference in discrete stochastic kinetic models
Gradient estimators enable parameter inference for stochastic kinetic models using discrete simulation algorithms.
6/4/2026
Attention-Based Sampler for Diffusion Language Models
Attention-based sampler enables parallel decoding in diffusion language models improving inference efficiency.
6/4/2026
Policy Split: Incentivizing Dual-Mode Exploration in LLM Reinforcement with Dual-Mode Entropy Regularization
Policy Split enables dual-mode exploration in LLM reinforcement learning balancing diversity and accuracy.
6/4/2026
Do LLMs Hold Their Values? MANTA: A Multi-Turn Adversarial Benchmark for Animal Welfare Reasoning
MANTA benchmark evaluates LLM value alignment on animal welfare across multi-turn adversarial queries.
6/4/2026
Markov Chain Decoders Overcome the Heavy-Tail Limitations of Lipschitz Generative Models
Markov chain decoders enable generative models to handle heavy-tailed distributions beyond Lipschitz constraints.
6/4/2026
Neural Autoregressive Control Variates for the Quantum Monte Carlo Sign Problem
Autoregressive models reduce sign problem in quantum Monte Carlo simulations using control variates.
6/4/2026
Structured Prompt Optimization Meets Reinforcement Learning for Global and Local Interpretability over Complex Text
eXTC combines structured prompt optimization with reinforcement learning for interpretable text classification, balancing performance, reasoning transparency, and scalability.
6/4/2026
No More K-means: Single-Stage Sparse Coding for Efficient Multi-Vector Retrieval
Single-stage sparse coding approach for efficient multi-vector retrieval replacing K-means clustering to reduce storage and computational overhead in billion-scale token retrieval.
6/4/2026
Equivariant Latent Alignment via Flow Matching under Group Symmetries
Equivariant latent alignment via flow matching for geometry-aware generative models incorporating group symmetries for improved novel view synthesis.
6/4/2026
LastAct: Trajectory-Guided Latest-Activity Localization for Real-Time Smart-Home Activity Recognition
LastAct localizes human activities in continuous smart-home sensor streams by addressing boundary contamination in sliding-window inference.
6/4/2026
SHARP: Sleep-based Hierarchical Accelerated Replay for Long Range Non-Stationary Temporal Pattern Recognition
SHARP framework for learning long-range non-stationary temporal patterns in streaming data through hierarchical accelerated replay, addressing backpropagation limitations.
6/4/2026
PaCX-MAE: Physiology-Augmented Chest X-Ray Masked Autoencoder
PaCX-MAE cross-modal distillation framework injecting physiological priors into chest X-ray encoders through masked autoencoding and contrastive-predictive objectives.
6/4/2026
ProtoAda: Prototype-Guided Adaptive Adapter Expansion and Geometric Consolidation for Multimodal Continual Instruction Tuning
ProtoAda framework for multimodal continual instruction tuning of LLMs with prototype-guided adapter expansion to reduce task interference in sequential learning.
6/4/2026
P$^2$-DPO: Grounding Hallucination in Perceptual Processing via Calibration Direct Preference Optimization
P²-DPO applies direct preference optimization to reduce hallucinations in vision-language models by targeting perceptual processing and robustness against image degradation.
6/4/2026
PerchRL: Vision-Based Agile Perching on Inclined Platforms under Rapid and Irregular Motion
PerchRL uses reinforcement learning for autonomous quadrotor perching on moving inclined platforms with limited field of view, combining state-based pretraining and vision-based control.
6/4/2026
A 3D Isovist World Model -- Revealing a City's Unseen Geometry and Its Emergent Cross-City Signature
3D isovist world models for embodied agent navigation that predict traversable geometry rather than visual appearance, revealing city structure across multiple cities.
6/4/2026
Qwen-Image-Flash: Beyond Objective Design
Qwen-Image-Flash optimizes text-to-image diffusion model distillation by analyzing training recipes beyond distillation objectives for few-step acceleration.
6/4/2026
Synthesize and Reward -- Reinforcement Learning for Multi-Step Tool Use in Live Environments
PROVE framework for training LLMs to orchestrate multi-step tool calls using reinforcement learning. Addresses stateful execution environments, synthetic query drift, and verbose tool-calling patterns.
6/3/2026
Show HN: Code3DBench – single-image to executable Three.js benchmark
Code3DBench tool generates executable Three.js 3D code from single images using AI.
6/3/2026
Inference Optimization for MiniMax Sparse Attention
Research on inference optimization for MiniMax sparse attention mechanism in transformer models.
6/3/2026
The American AI Sovereign Wealth Fund Act [video]
Video about proposed AI sovereign wealth fund legislation.
6/3/2026
A New Path Forward for Session
Session is a decentralized end-to-end encrypted messaging app with no phone number requirement and onion routing.
6/3/2026
Scaling Past Informal AI – Carina Hong, Axiom Math
Axiom startup solved 12/12 Putnam math competition problems, outperforming prior AI systems on difficult undergraduate math exam.
6/3/2026
WebKit position stated as opposed to WebMCP
WebKit position statement on WebMCP specification for web machine learning, requesting standards body feedback.
6/3/2026
Show HN: TurboPrefill – Multi-GPU prefill acceleration for llama.cpp
TurboPrefill optimization for multi-GPU prefill acceleration in llama.cpp to improve inference throughput.
6/3/2026
Only 18% of AI engineering spend reaches shipped product
Analysis of 2.4K organizations showing only 18% of AI engineering spend reaches production; $0.82 per dollar lost to inefficiency.
6/3/2026
SpaceX Seeks $75B in Record IPO Plan to Fund AI, Launch
SpaceX IPO announcement mentioning AI funding alongside space launch capabilities.