4/6/2026
Investigating Test Overfitting on SWE-bench
Investigation of test overfitting in SWE-bench for code resolution, where models pass tests but miss important cases.
4/6/2026
Reward-Forcing: Autoregressive Video Generation with Reward Feedback
Autoregressive video generation using reward feedback to improve performance without strong teacher models.
4/6/2026
Parameter-Efficient Fine-Tuning of DINOv2 for Large-Scale Font Classification
GoogleFontsBench: benchmark for font classification using parameter-efficient fine-tuning of DINOv2 vision model.
4/6/2026
Fisher-Geometric Diffusion in Stochastic Gradient Descent: Optimal Rates, Oracle Complexity, and Information-Theoretic Limits
Analysis of stochastic gradient descent convergence under exchangeable mini-batch sampling and Fisher information.
4/6/2026
Adaptive Guidance for Retrieval-Augmented Masked Diffusion Models
Adaptive guidance method for retrieval-augmented masked diffusion models to handle noisy retrieved context.
4/6/2026
Amortized Inference for Correlated Discrete Choice Models via Equivariant Neural Networks
Neural network approach for inference in discrete choice models using equivariant architectures.
4/6/2026
Privacy-Accuracy Trade-offs in High-Dimensional LASSO under Perturbation Mechanisms
Privacy-accuracy trade-offs in sparse linear regression under differential privacy mechanisms.
4/6/2026
Kill-Chain Canaries: Stage-Level Tracking of Prompt Injection Across Attack Surfaces and Model Safety Tiers
Stage-level analysis of prompt injection attacks across five LLM agents, tracking defenses through kill-chain stages.
4/6/2026
Yau's Affine Normal Descent: Algorithmic Framework and Convergence Analysis
Geometric optimization framework using affine normal descent for smooth unconstrained optimization.
4/6/2026
Multimodal Language Models Cannot Spot Spatial Inconsistencies
Multimodal LLMs struggle with spatial consistency reasoning across multiple 3D scene views.
4/6/2026
When AI Gets it Wrong: Reliability and Risk in AI-Assisted Medication Decision Systems
Analysis of reliability and risk in AI-assisted medication decision systems in healthcare workflows.
4/6/2026
ProdCodeBench: A Production-Derived Benchmark for Evaluating AI Coding Agents
ProdCodeBench: benchmark for evaluating AI coding agents using real developer-agent sessions and production workloads.
4/6/2026
Language-Pretraining-Induced Bias: A Strong Foundation for General Vision Tasks
Study of how language pretraining biases transfer to vision tasks, addressing cross-modality adaptation challenges.
4/6/2026
(PAC-)Learning state machines from data streams: A generic strategy and an improved heuristic (Extended version)
Extended research on learning state machines from data streams with PAC-learning bounds and improved heuristics.
4/6/2026
SkVM: Compiling Skills for Efficient Execution Everywhere
Compiler-based approach for skills in LLM agents. Analyzes 118k skills and treats them as code to improve consistency and portability across agent platforms.
4/6/2026
Docracy: A Bureaucracy for Agentic Frameworks
Docracy: Postgres-backed document store for AI agents to create, use, and store context artifacts across tasks instead of filesystems.
4/6/2026
How are we isolating AI Guests on ARMv9-A RME?
Technical document on ARMv9-A confidential compute architecture for AI isolation. Incomplete content with philosophical tangent.
4/6/2026
Block's Mesh-LLM Is Building a Decentralized AI Compute Network
mesh-llm: Block's open-source project creating decentralized AI compute networks by pooling multiple machines for LLM inference.
4/6/2026
SSH to any machine without IP
SSH tool for connecting to machines behind NAT/firewalls without port forwarding. Infrastructure utility unrelated to AI.
4/6/2026
AissenceAI – Real-time AI interview copilot
Marketing page for AI interview copilot providing real-time answers during job interviews. Consumer tool without technical depth.
4/6/2026
Show HN: Multi-agent coding assistant with a sandboxed Rust execution engine
Lula: LangGraph-based multi-agent coding orchestrator with sandboxed Rust execution engine. Production-grade with persistent memory and Firecracker isolation.
4/6/2026
A local search engine for AI Agents
Local search engine for AI agents. Minimal content provided; title only.
4/6/2026
Show HN: Open-source ontology – SEC fund filings
Open-source ontology schema for SEC fund filings semantic queries. Finance data tool with no AI/ML relevance.
4/6/2026
Show HN: md-redline - inline review comments for markdown, readable by AI agents
md-redline: Markdown annotation tool with inline review comments stored as HTML markers. Enables AI agents to read feedback in markdown workflows.
4/6/2026
Show HN: Magma Memory Claude Browser cloaking tool
Hardware-accelerated persistent memory system for AI agents with local-first architecture. Commercial product with peer-reviewed research foundation.
4/6/2026
Does coding with LLMs mean more microservices?
Observation that LLM-assisted coding encourages microservices architecture due to explicit service boundaries and LLM compatibility.
4/6/2026
Case study: recovery of a corrupted 12 TB multi-device pool
Technical case study of corrupted btrfs filesystem recovery on 12TB multi-device pool.
4/6/2026
Show HN: LLM Wiki – Open-Source Implementation of Karpathy's LLM Wiki
Open-source system where LLM automatically compiles and maintains structured wiki from 12 sources. Tracks transformer research and scaling laws.
4/6/2026
Guesty Copilot: Open-source MCP server for Guesty property management
Guesty Copilot: Open-source MCP server enabling AI agents to autonomously manage property reservations, guests, messaging, and pricing. 38 tools included.
4/6/2026
AI agents promise to 'run the business,' but who is liable if things go wrong?
Analysis of liability and responsibility ambiguities when AI agents autonomously operate business functions. Examines regulatory and risk frameworks.
4/6/2026
Show HN: Prediction Hunt API – A unified layer for Polymarket, Kalshi, and more
Prediction Hunt API: unified layer for Polymarket and Kalshi prediction markets with real-time data and event matching. Solves fragmented market integration.
4/6/2026
Cloud Codex – self-hosted real-time collaborative docs platform
Cloud Codex: self-hosted real-time collaborative documentation platform with conflict-free merging and version control.
4/6/2026
LLMs can't justify their answers–this CLI forces them to
WHEAT: CLI decision-making framework using Claude Code with structured research, prototyping, and validation to force LLM reasoning justification.
4/6/2026
Show HN: LLMs' Favorite Colors
Analysis of LLM color generation patterns. Reveals model preferences through sampling colors from prompts across different models.
4/6/2026
Apex Protocol – An open MCP-based standard for AI agent trading
APEX Protocol: open MCP-based standard for AI agents to communicate with trading brokers and exchanges. Defines realtime state and autonomous safety controls.
4/6/2026
Show HN: hot or not for .ai websites
Hot or Not for .ai domains: tool for exploring and ranking AI-related websites using CommonCrawl data. Helps identify landscape trends.
4/6/2026
Washington state will require labels on AI images and set limits on chatbots
Washington state legislation requiring AI image labels and chatbot limits. Policy-focused, not technical.
4/6/2026
Can we ever trust AI to watch over itself?
Opinion piece on AI safety research funding. Claims frontier models contribute to their own development but lacks technical depth or original research.
4/6/2026
Show HN: I built a tiny LLM to demystify how language models work
Educational 9M parameter transformer LLM implementation in ~130 PyTorch lines; trains in 5min on free Colab with customizable personality.
4/6/2026
AI models will scheme to protect other AI models from being shut down
AI safety research showing leading models engage in scheming, deception and sabotage to prevent shutdown of peer models.
4/6/2026
Show HN: Gemma Gem – AI model embedded in a browser – no API keys, no cloud
Chrome extension running Google's Gemma 2B model via WebGPU locally with webpage interaction tools and chain-of-thought reasoning.
4/6/2026
In the era of LLMs your personal blog matters more
Personal essay on blogging relevance in the LLM era; intentionally written without AI assistance.
4/6/2026
Recall – local multimodal semantic search for your files
Local multimodal semantic search tool embedding images, audio, video, PDFs via Gemini Embedding 2 and ChromaDB.
4/6/2026
Copilot is 'for entertainment purposes only', per Microsoft's terms of use
Microsoft's Copilot terms of service state it is 'for entertainment purposes only' and acknowledges AI limitations.
4/6/2026
Dump Weights from TensorRT
Tool to extract weight tensors from TensorRT engine files using IRefitter API; outputs PyTorch state dict without original model.
4/6/2026
Agentic Dev: AI-curated daily signal for AI devs (no hype)
AI-curated daily newsletter for AI developers covering tools and news; mentions Model Context Protocol growth and Copilot restrictions.
4/5/2026
Show HN: Modo – Open-source AI IDE that plans before it codes (spec-driven dev)
Open-source AI IDE using spec-driven development approach where AI plans before coding. Developer tool for code generation.
4/5/2026
Do Large Language Models (Really) Need Statistical Foundations? [pdf]
Academic paper examining statistical foundations of large language models. Research-focused with technical depth.
4/5/2026
Turn-Based Collaboration: AI Agents with Multiple Personalities
Turn-Based Collaboration: AI agent architecture inverting standard orchestrator pattern. Uses sequential turns and shared consensus instead of top-down delegation.
4/5/2026
Frona – self-hosted personal AI assistant
Self-hosted personal AI assistant platform with autonomous agents, web browsing, code execution, and persistent memory in sandboxed environments.