3/19/2026
InPhyRe Discovers: Large Multimodal Models Struggle in Inductive Physical Reasoning
Benchmark study showing large multimodal models fail at inductive physical reasoning beyond training distribution.
3/19/2026
EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing
EdiVal-Agent framework for automated, fine-grained evaluation of multi-turn image editing using object-centric assessment.
3/19/2026
Detecting Data Contamination from Reinforcement Learning Post-training for Large Language Models
Detection methods for data contamination in RL post-training phase of LLMs, addressing evaluation validity gap.
3/19/2026
CBF-RL: Safety Filtering Reinforcement Learning in Training with Control Barrier Functions
CBF-RL integrates Control Barrier Functions into RL training to enforce safety constraints during policy learning.
3/19/2026
Neighbor GRPO: Contrastive ODE Policy Optimization Aligns Flow Models
Neighbor GRPO extends Group Relative Policy Optimization to flow matching models with contrastive ODE-based approach for generative model alignment.
3/19/2026
Representation-Aware Unlearning via Activation Signatures: From Suppression to Knowledge-Signature Erasure
Knowledge Immunization Framework for selective knowledge erasure from LLMs via representation-aware activation signatures, addressing GDPR and safety.
3/19/2026
Beware Untrusted Simulators -- Reward-Free Backdoor Attacks in Reinforcement Learning
Study demonstrating reward-free backdoor attacks on RL agents through compromised simulators.
3/19/2026
PAND: Prompt-Aware Neighborhood Distillation for Lightweight Fine-Grained Visual Classification
Knowledge distillation framework for fine-grained visual classification using vision-language models with prompt-aware calibration.
3/19/2026
GTS: Inference-Time Scaling of Latent Reasoning with a Learnable Gaussian Thought Sampler
Learnable Gaussian sampling method for inference-time scaling in latent reasoning models to improve reasoning path generation.
3/19/2026
Ethical Fairness without Demographics in Human-Centered AI
Method for achieving fairness in AI systems without demographic attributes for human-centered applications.
3/19/2026
REFINE-DP: Diffusion Policy Fine-tuning for Humanoid Loco-manipulation via Reinforcement Learning
Method for fine-tuning diffusion policies with reinforcement learning for humanoid robot loco-manipulation tasks.
3/19/2026
ASAP: Attention-Shift-Aware Pruning for Efficient LVLM Inference
Pruning method for efficient large vision-language model inference by exploiting attention patterns and addressing token redundancy.
3/19/2026
I Know What I Don't Know: Latent Posterior Factor Models for Multi-Evidence Probabilistic Reasoning
Framework for aggregating noisy heterogeneous evidence in probabilistic reasoning tasks with explicit uncertainty quantification.
3/19/2026
Theoretical Foundations of Latent Posterior Factors: Formal Guarantees for Multi-Evidence Reasoning
Theoretical framework for aggregating multiple evidence sources in probabilistic prediction with formal guarantees for multi-evidence reasoning.
3/19/2026
k8s multi tenancy: its a cluster f*ck
Kubernetes multi-tenancy challenges with AI agents requiring ephemeral environments. Infrastructure scaling issues.
3/19/2026
The Flawed Ephemeral Software Hypothesis
Analysis arguing software won't become disposable despite AI coding agents, critiquing concepts like 'vibe coding' and ephemeral apps.
3/19/2026
South Korea's SDT opens Quantum-AI data center
South Korea's SDT opens first commercial quantum-AI hybrid data center in Seoul with 20-qubit Kreo quantum computer and Nvidia DGX B200 integration.
3/19/2026
Scheduled: Open-Source Calendly, but AI
Scheduled: Open-source AI agent integrated with Gmail that autonomously reads meeting request emails, checks calendar availability, and drafts proposed times.
3/19/2026
Show HN: ATO – a GUI to see and fix what your LLM agents configured
ATO: GUI control panel managing multiple LLM agents (Claude Code, Codex, OpenClaw, Hermes) with workflow orchestration and MCP integration.
3/19/2026
AI pilot program in LA County courts will help judges craft rulings
LA County courts pilot AI tool (Learned Hand) to summarize legal motions and draft rulings based on judge writing styles.
3/19/2026
The Agentic Organization
Conceptual framework on how AI agents transform organizational structure and decision-making beyond efficiency gains.
3/19/2026
I Handed Off Maintenance of My Chrome Extension to AI
Case study: developer maintaining open-source Chrome extension with AI assistance for bug fixes and feature development.
3/19/2026
OpenAI to acquire Astral
OpenAI acquires Astral, integrating open source Python developer tools (uv, Ruff) into Codex ecosystem to enhance Python development tooling.
3/18/2026
Chainguard Is Now Protecting You from AI Agent Skills Gone Rogue
Chainguard Agent Skills: security solution protecting against malicious AI agent skills with verification and sandboxing for YAML-based agent plugins.
3/18/2026
Fighting Context Drift
Trepan: local-first architectural linter enforcing code intent and preventing architecture drift without cloud data transmission.
3/18/2026
PondDB – Self-hosted agent memory database built on DuckDB
PondDB: open-source DuckDB-based memory database for multi-agent systems enabling SQL querying of agent state and decision history.
3/18/2026
Show HN: A Genetic algorithm that red-teams your copy with 100 LLM personas
Genetic algorithm that uses 100 LLM personas to red-team and improve landing page copy generation, addressing generic AI writing outputs.
3/18/2026
Show HN: Blobsearch – Object storage and DuckDB based Elasticsearch alternative
Blobsearch: DuckDB-based log storage and querying alternative using S3 and Parquet for cost-effective log management.
3/18/2026
Claude Code's poor time awareness
Bug report on Claude Code's poor time-awareness limiting task optimization and efficiency in code completion.
3/18/2026
Show HN: Jira-MCP – Full Jira for AI agents in 3 tools, not 72
MCP tool providing simplified Jira integration for AI agents via 3 composable tools instead of 72 API endpoints.
3/18/2026
Skillfile, the declarative skill manager, now auto-discovers skills in repos
Skillfile: declarative manifest system for managing AI agent skills across Claude, Cursor, Gemini and other platforms with versioning and deployment.
3/18/2026
Business-as-Code with LittleHorse 1.0
LittleHorse 1.0: microservice orchestration engine enabling Business-as-Code approach for distributed process definition.
3/18/2026
Pervaziv AI Code Review GitHub Action
GitHub Action providing AI code review via Pervaziv.
3/18/2026
FastAPI-compatible Python framework with Zig HTTP core; 7x faster
TurboAPI: FastAPI-compatible Python framework with Zig HTTP core, 7x faster with zero-copy responses.
3/18/2026
A BEAM-native personal autonomous AI agent built on Elixir/OTP
Open-source personal autonomous AI agent built on Elixir/OTP that monitors feeds, executes workflows, routes tasks to cheapest suitable LLM. Single-user, auditable codebase.
3/18/2026
iCloud Photo Sharing Sucks – Can AI Help Me Replace It?
Personal project using Claude to build photo sharing app replacing iCloud. Practical LLM application with implementation discussion.
3/18/2026
We Made AI Gamble. What Poker Revealed About Frontier LLMs
Headline about poker experiments with frontier LLMs. Appears duplicate of article [3] with less content.
3/18/2026
We Made LLMs Gamble: Heres What Poker Revealed About Frontier AI Models
Research using Claude Sonnet and Gemini Flash agents to play poker, revealing reasoning capabilities and strategic decision-making in frontier LLMs through game theory.
3/18/2026
Show HN: Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training
Experiment replicating RYS method on consumer AMD GPUs, discovering discrete reasoning circuits in 24B LLM by duplicating layers improves logical deduction from 0.22 to 0.76.
3/18/2026
Show HN: Stop GPU pods placement getting bottlenecked by reserved VRAM
GPU runtime for Nvidia GPUs enabling safe VRAM overcommit, fractional core allocation, and weight deduplication.
3/18/2026
VibePod adds Ollama/vLLM back end support for Claude Code and Codex
VibePod adds Ollama/vLLM backend support for Claude Code and Codex.
3/18/2026
Why Smart Engineers Still Miss What Makes Enterprise AI Work
Enterprise AI adoption gap: models and agents scale but organizational context understanding lags. Governance and activation challenges remain.
3/18/2026
Show HN: Real-time local TTS (31M params, 5.6x CPU, voice cloning, ONNX)
Local TTS model with 31M params, voice cloning, voice blending. 5.6x realtime on CPU, ONNX export, Apache 2.0 license.
3/18/2026
Warranty Void If Regenerated
Using Claude to generate fiction stories with world-building documents for creative writing projects.
3/18/2026
Phantom – A memory system for local LLMs that enriches itself while you sleep
Phantom: persistent memory system for local LLMs with continuous enrichment loop and knowledge organization.
3/18/2026
GFS – Git for databases, built for AI coding agents (commit, branch, checkout)
GFS: Git-like version control for databases, compatible with Claude Code and MCP agents. Docker-based isolation for safe DB management.
3/18/2026
Anthropic's code execution pattern for MCP cuts agent token usage from 150K – 2K
Anthropic's MCP code execution pattern reduces agent token usage from 150K to 2K.
3/18/2026
Claude Is Getting Better. What About Us?
Essay on skill development and debugging abilities in context of improved Claude capabilities.
3/18/2026
The more I work with AI (LLMs) the more disillusioned I become
Opinion piece skeptical of LLM capabilities, questioning replacement of white-collar work.
3/18/2026
Autoresearching Apple's "LLM in a Flash" to run Qwen 397B locally
Research applying Apple's LLM-in-Flash technique to run Qwen 397B model locally.