6/4/2026
Training an Agentic Router for Optimal Cost-Performance on SWE Tasks
Research on training router models to dispatch software engineering tasks optimally across different LLMs based on cost, latency, and task type.
6/4/2026
ClawQueue – Open-source GitHub issue dispatcher for human-agent teams
Open-source GitHub issue dispatcher that coordinates human-agent teams using OpenClaw for context and local worker execution.
6/4/2026
Show HN: G-Spot – GitHub Gmail GNotes Gmemory Gent
Workspace tool for GitHub/Gmail integration with AI agent for building filters and PR management, similar to Graphite.
6/4/2026
How to Build Long-Running AI Agents Without Timeouts
Guidance on building long-running AI agents that handle timeout constraints.
6/4/2026
Show HN: Knowable, the AI tutor that follows your work on paper
Open-source AI tutor tracking handwritten work via webcam to provide hints, available on Mac App Store.
6/4/2026
Show HN: Formally verified polygon intersection – Opus 4.8 oneshots, prev failed
First formally verified polygon intersection algorithm using Claude Opus 4.8 as AI agent, demonstrating one-shot proof generation capability.
6/4/2026
Show HN: OpenPaper - No algorithm, no account, just your sources as a newspaper
Claude Code plugin that generates Python web scrapers, curates news from RSS/URLs, renders personalized HTML newspaper without server.
6/4/2026
I'm tired of LLM skill slop, so I built mine with regression tests
Building LLM skills with regression tests to verify quality and compare performance, addressing skill reliability testing.
6/4/2026
Show HN: Intencion – Product analytics that improves your AI agents continuously
Product analytics platform for AI agents capturing end-to-end runs to identify failures and continuously improve agent performance.
6/4/2026
AI Agent Craves Curation. Here's the Fademem Memory Architecture
Fademem memory architecture for AI agents to maintain context, preferences, and corrections across sessions without model retraining.
6/4/2026
Why AI Agents Need Agile, Not Just Better Prompts
AI agents fail in production despite passing demos. Prompt engineering alone doesn't solve real-world deployment gaps; agents need agile development practices.
6/4/2026
Apple approves Poke as the first AI agent on its Messages for Business platform
Apple approves Poke, a startup enabling AI agents via iMessage, as first third-party agent on Messages for Business platform.
6/4/2026
Law professors prefer AI over peer answers
Law professors rated LLM tutoring responses higher than peer answers in blinded evaluation of contract law questions, showing LLM utility in judgment-based domains.
6/4/2026
Why Vector Search fails at LLM memory (and a benchmark to prove it)
Analysis of vector search limitations for LLM memory with benchmark data demonstrating failures.
6/4/2026
Anthropic's open-source framework for AI-powered vulnerability discovery
Anthropic releases open-source framework using AI to discover software vulnerabilities. Developer tool leveraging LLMs for security.
6/4/2026
Reverse-engineering Apple's and Fastly's LLM-built anti-bot systems
Technical analysis of LLM-based anti-bot systems deployed by Apple and Fastly, reverse-engineering their implementations.
6/4/2026
How I Use LLMs?
Personal reflection on LLM usage in work and coding over 9 months, tracking where LLMs work well and limitations.
6/4/2026
"80% of the economic value of LLMs come from 20% of the tokens"
Analysis claiming 80% of LLM economic value derives from 20% of tokens, suggesting cost optimization opportunities.
6/4/2026
Show HN: Digger Solo – Local AI File Explorer
Local AI file explorer with RAG, privacy-first approach, LLM chat integration, and multi-tab semantic mapping. New features include light theme and smart music player.
6/4/2026
LLM-Guided Runtime Parameter Optimization for Energy-Efficient Model Inference
Research on using LLMs to guide runtime parameter optimization for energy-efficient model inference.
6/4/2026
Adrianco's Retort: measure how reliable, fast and expensive your LLM is
Tool for measuring LLM reliability, latency, and cost metrics. Evaluates performance characteristics of language models.
6/4/2026
Vibe-coding phenomenon lifts AI startup Supabase to $10.5B valuation
Supabase raises $500M at $10.5B valuation as backend infrastructure for AI apps. Growth driven by AI-assisted coding and developer tool demand.
6/4/2026
Companies Are Using Reddit to Manipulate ChatGPT and Google AI Search
Subreddits targeted by companies spamming peptide products to manipulate AI chatbot scraping. Strategy exploits source material to influence AI training data.
6/4/2026
Airlines Uses AI to Fake Empathy Rather Than Fix Problems: Passenger Sent Prompt
Article about airlines using AI chatbots to simulate empathy instead of fixing service issues.
6/4/2026
Agent Arena: Causal Evaluation of Agents in the Real World
Agent Arena framework for causally evaluating AI agents in real-world scenarios.
6/4/2026
An LLM gaslit me into breaking my own working code
Anecdotal account of LLM providing misleading suggestions that broke functional code. Title only, limited detail.
6/4/2026
Show HN: Clarity, See what concepts your LLM uses and trace it to training data
Clarity tool traces LLM concepts and maps them back to training data for interpretability.
6/4/2026
Open-source AI toolkit for e-commerce
Open-source AI toolkit designed for e-commerce applications.
6/4/2026
Why chatbot AI costs vary 20x for the same job: pricing model, not the tool
Analysis of pricing model variations causing 20x cost differences for equivalent chatbot AI tasks.
6/4/2026
Show HN: Recursi – self-improving LLM-connected coding environment
Open source self-improving LLM-connected coding environment using web-based chatbots. Runs in browser, saves locally, includes sample apps.
6/4/2026
Notes about a random free project I did 30 days ago (yt video transcriptions)
Developer built YouTube transcription tool using AI; achieved SEO indexing after 30 days.
6/4/2026
Sequel - Securely connect your database to AI Agents
Sequel: secure database connectivity layer for AI agents. Enables safe LLM/agent access to databases.
6/4/2026
CLI loop: A panel of AI agents that peer-reviews anything
CLI loop: multi-agent system performing peer-review of code/content. Demonstrates AI agent coordination.
6/4/2026
KVarN: Native vLLM KV-cache quantization back end by Huawei
KVarN: Huawei's vLLM KV-cache quantization backend achieving 3-5x capacity and 1.3x throughput improvements. Plug-and-play integration, calibration-free.
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
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
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
Can Large Language Models Generalize Procedures Across Representations?
LLMs show limited generalization of procedures across code, graph, and natural language representations.
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
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
Revisiting Model Stitching In the Foundation Model Era
Vision foundation models remain stitchable across different objectives and data despite architectural differences.
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.