5/8/2026
FedAttr: Towards Privacy-preserving Client-Level Attribution in Federated LLM Fine-tuning
Privacy-preserving attribution methods for detecting data ownership in federated LLM fine-tuning using watermarking techniques.
5/8/2026
UniSD: Towards a Unified Self-Distillation Framework for Large Language Models
Unified self-distillation framework for LLMs without external teachers, addressing free-form trajectories and task-dependent correctness.
5/8/2026
Are We Making Progress in Multimodal Domain Generalization? A Comprehensive Benchmark Study
Benchmark study evaluating multimodal domain generalization methods to determine if performance gains reflect real progress or inconsistent evaluation protocols.
5/8/2026
Superintelligent Retrieval Agent: The Next Frontier of Information Retrieval
Retrieval-augmented agent framework for navigating large knowledge bases with intelligent query reformulation and evidence gathering.
5/8/2026
DeTrigger: A Gradient-Centric Approach to Backdoor Attack Mitigation in Federated Learning
Defense mechanism against backdoor attacks in federated learning using gradient analysis to detect and mitigate model poisoning.
5/8/2026
Teaching Metric Distance to Discrete Autoregressive Language Models
Training method teaching metric distance relationships to discrete autoregressive LLMs for improved performance on numerical and spatial tasks.
5/8/2026
Time series saliency maps: explaining models across multiple domains
Cross-domain Integrated Gradients method for generating saliency maps across multiple domains for time series model interpretability.
5/8/2026
Why and When Deep is Better than Shallow: Implementation-Agnostic State-Transition Model of Deep Learning
Theoretical analysis of depth's role in generalization using state-transition model separating implementation, approximation, and statistical error.
5/8/2026
Position: Adopt Constraints Over Fixed Penalties in Deep Learning
Position paper arguing constraints should replace fixed penalty methods in deep learning for trustworthy AI systems.
5/8/2026
Leveraging Analytic Gradients in Provably Safe Reinforcement Learning
Provably safe reinforcement learning framework integrating analytic gradients with safety safeguards for autonomous robot deployment.
5/8/2026
A Survey of Personalized Federated Foundation Models for Privacy-Preserving Recommendation
Survey on federated foundation models applied to recommendation systems while preserving privacy using decentralized learning.
5/8/2026
Attribution-Guided Pruning for Insight and Control: Circuit Discovery and Targeted Correction in Small-scale LLMs
Mechanistic interpretability method using attribution-guided pruning to discover and correct circuits in small-scale LLMs.
5/8/2026
Variational Kolmogorov-Arnold Network
Variational formulation of Kolmogorov-Arnold Networks addressing hyperparameter selection for basis function count via Bayesian approach.
5/8/2026
Sample-efficient LLM Optimization with Reset Replay
Reset Replay technique for sample-efficient LLM post-training via RL and preference optimization, addressing primacy bias and plasticity degradation.
5/8/2026
Multi-Objective Instruction-Aware Representation Learning in Procedural Content Generation RL
Multi-objective instruction-aware RL for procedural content generation, improving controllability under complex natural language guidance.
5/8/2026
Beyond Memorization: Extending Reasoning Depth with Recurrence, Memory and Test-Time Compute Scaling
Controlled study of multi-step reasoning in LLMs using cellular automata framework, showing models fail extended prediction despite learning local rules.
5/8/2026
Synthetic Counterfactual Labels for Efficient Conformal Counterfactual Inference
Synthetic data augmentation for conformal counterfactual inference, improving prediction intervals under treatment imbalance.
5/8/2026
Robust Filter Attention: Self-Attention as Precision-Weighted State Estimation
Formulates self-attention as robust state estimation with linear SDE model, matching standard attention complexity under isotropic noise assumptions.
5/8/2026
Defending Diffusion Models Against Membership Inference Attacks via Higher-Order Langevin Dynamics
Defense mechanism against membership inference attacks on diffusion models via higher-order Langevin dynamics.
5/8/2026
Frictional Q-Learning
Addresses extrapolation error in off-policy RL using friction analogy, representing replay buffer as action manifold with tangential/normal components.
5/8/2026
A Forensic Analysis of Synthetic Data in RL: Diagnosing and Solving Algorithmic Failures in Model-Based Policy Optimization
Forensic analysis of synthetic data failure modes in model-based RL, diagnosing and solving issues in MBPO policy optimization.
5/8/2026
KaVa: Latent Reasoning via Compressed KV-Cache Distillation
Compressed KV-cache distillation enabling latent reasoning in LLMs, reducing computational costs while internalizing chain-of-thought.
5/8/2026
On the optimization dynamics of RLVR: Gradient gap and step size thresholds
Theoretical analysis of Reinforcement Learning with Verifiable Rewards (RLVR) training dynamics via gradient gap metric, explaining empirical success.
5/8/2026
Exploring Cross-Client Memorization of Training Data in Large Language Models for Federated Learning
Examines cross-sample data memorization risks in federated learning of LLMs, adapting fine-grained detection methods from centralized settings.
5/8/2026
Large Language Model Prompt Datasets: An In-depth Analysis and Insights
Analysis of 129 LLM prompt datasets (>1.22TB) with linguistic patterns and downstream tasks: filtering, classification, quality prediction.
5/8/2026
Efficient Test-Time Adaptation through Latent Subspace Coefficients Search
Gradient-free framework for test-time adaptation under distribution shift via latent subspace search, enabling efficient edge deployment.
5/8/2026
Graph Learning Is Suboptimal in Causal Bandits
Theoretical analysis showing learning causal parent sets is suboptimal for regret minimization in causal bandits with unknown structure.
5/8/2026
A Practitioner's Guide to Kolmogorov-Arnold Networks
Comprehensive review of Kolmogorov-Arnold Networks literature, clarifying relationships with MLP theory, kernel methods, and Kolmogorov superposition theorem.
5/8/2026
NOWS: Neural Operator Warm Starts for Accelerating Iterative Solvers
Hybrid approach combining learned neural operators with iterative solvers to accelerate PDE computations while maintaining reliability outside training distribution.
5/8/2026
High entropy leads to symmetry equivariant policies in Dec-POMDPs
Theoretical analysis proving high entropy regularization in Dec-POMDPs leads to convergent, symmetry-equivariant policies across different initializations.
5/8/2026
Higher usage limits for Claude and a compute deal with SpaceX
Anthropic announces increased compute capacity and usage limits for Claude Code and Claude API through SpaceX partnership.
5/8/2026
Show HN: jj diff review integrated with agents
jj diff review tool integrated with AI agents for code review automation.
5/8/2026
Sley is live: the first native AI programming language
Sley is a new agent-native structural programming language designed for AI agents with deterministic edits and graph-aware tooling. Open source under Apache-2.0.
5/8/2026
Show HN: Loxai.tech and Neutboom – Gen AI's frontier of individuality
Commentary on AI startup failures, LLM wrappers becoming obsolete as foundation models offer direct functionality.
5/8/2026
Claude Code CVE-2026-39861:sandbox escape via symlink
Security vulnerability in Claude Code sandbox allowing symlink escape that bypasses file access restrictions.
5/8/2026
Show HN: AnamDB – An AI-native, differentiable Datalog engine written in Rust
AnamDB: Rust neurosymbolic database engine integrating probabilistic neural perception with deterministic Datalog reasoning. AI-native differentiable architecture.
5/8/2026
Using AI for Just 10 Minutes Might Make You Lazy and Dumb, Study Shows
Carnegie Mellon, MIT, Oxford, UCLA study on cognitive impact of AI chatbot use. 10-minute exposure affects problem-solving.
5/8/2026
Show HN: WordPress WebSocket Relay in Rust with Yjs CRDT Provider for WP 7.0
Rust WebSocket relay for WordPress 7.0 Yjs CRDT collaborative editing, replacing HTTP polling with realtime events.
5/8/2026
AniTroves – An anime database with a custom LLM-based discovery hub
AniTroves anime database with LLM-based conversational discovery interface replacing traditional tag-based search.
5/8/2026
Show HN: Agentctl, a local control plane for coding agents
Go tool providing local control plane for AI coding agents, managing package installs, shell execution, and API calls with policy enforcement.
5/8/2026
AI Contributions to CPAN: The Copyright Question
Discussion of copyright and licensing issues for AI-assisted contributions to CPAN open-source modules, documenting major projects' policies.
5/8/2026
GPT-5.5 Price Increase: What It Costs
Cost analysis of GPT-5.5 pricing and token economics compared to GPT-5.4 with 2x price increase and lower verbosity.
5/8/2026
The IDE Should Become an Operating System for AI
Conceptual argument for IDEs to evolve into programmable operating systems for AI, supporting agents, terminals, logs, tests, and long-running sessions beyond traditional file-based workflows.
5/8/2026
The first repo with 500k+ stars
Open source repository with 500k+ stars containing step-by-step guides for implementing technologies from scratch, maintained by CodeCrafters for learning programming fundamentals.
5/8/2026
Show HN: Selvedge – an MCP server that captures why AI agents change code
Selvedge: MCP server providing long-term memory for AI-coded codebases by logging reasoning behind code changes made by Claude Code, Cursor, and Copilot in SQLite.
5/8/2026
Show HN: Armorer – A secure local control plane for AI agents
Armorer: secure local control plane for AI agents managing lifecycle with Docker isolation, unified UI/CLI for monitoring, and dependency management for agent setup.
5/8/2026
"Surface" a Governed AI-Agentic Surface
Work-in-progress implementation of AI-agentic surface called Surface, based on book 'The Evolution of Software Scale', with GitHub repositories for desktop and agent components.
5/8/2026
Show HN: An OTel exporter that posts the cause to your incident channel
OpenTelemetry exporter for incident alerting that integrates trace data with PagerDuty, OpsGenie, and Slack notifications.
5/7/2026
Spent $130K+ AI token "cloned" Screen Studio: AGI for software feel so close
Developer built ScreenKite product using AI, spending $130K+ on tokens; claims to match/exceed Screen Studio with faster 4K export.
5/7/2026
AI Bots Auditioning for Wall Street Are Mostly Losing
Empirical evaluation showing AI trading systems underperform in contests, with inconsistent decision-making and excess trading.