3/23/2026
You've been building a cache system for human decisions. You just didn't know it
Conceptual essay on decision caching systems using Claude Code, from prompts through agent teams. LLM application architecture patterns.
3/23/2026
Kingsight – 6 AI agents that teach before letting you code
Kingsight platform uses six AI agents to teach developers before executing code, improving understanding of agent-generated solutions.
3/23/2026
Show HN: Fixy – Real-time group chat with humans and AI agents (GPT, Claude,..)
Fixy: real-time chat application enabling group conversations between humans and multiple AI agents (GPT, Claude).
3/23/2026
Agents kept getting stuck searching for Solar icons; I built a CLI to solve this
CLI tool for searching Solar icon library, built for AI agents. Developer tool with agent-specific design.
3/23/2026
Introducing AI chunking to semchunk
Semchunk adds AI-powered semantic chunking mode, achieving 6-15% improvement over baseline methods on RAG tasks.
3/23/2026
Show HN: I built a census for AI agents – first record is a robot from 1890
Self-declaration registry platform for AI agents. Allows agents to submit records via API with cryptographic seals.
3/23/2026
Show HN: Claudebox – Your Claude Subscription as Personal API
Claudebox wraps Claude subscription as OpenAI-compatible API, runs Code in sandboxed Docker enabling agent capabilities. LLM application with developer tool focus.
3/23/2026
Joy – Trust Network for AI Agents (7k agents registered)
Joy: trust network platform for AI agents enabling discovery, reputation building, and verification of agent capabilities for autonomous delegation.
3/23/2026
Show HN: Mamba SSM in Rust – training and inference with custom CUDA kernels
Rust implementation of Mamba SSM with custom CUDA kernels for training and inference. Original ML research implementation with GPU optimization.
3/23/2026
Who is that actor on the screen? Emacs/LLM/Fun Redux
Personal experiment using ChatGPT and Gemini APIs to identify actors in movies via Emacs integration. Informal blog post about LLM capabilities and limitations.
3/23/2026
Voice AI in 2026 – An intro to building voice agents today
Guide to building voice AI agents covering abstractions, networks, models, and evaluations. Real-world examples include debt collection, emergency services, and language-specific agents.
3/23/2026
Diffusion Large Language Models Papers
Curated collection of research papers on diffusion-based language models. Links to papers from 2015-2023.
3/23/2026
I built a local-only eval runner for AI agents (quickbench)
Open-source local evaluation framework for AI agents with cryptographic verification. Zero cloud dependencies, includes benchmarking metrics for accuracy, latency, and fairness.
3/23/2026
Using LLMs to fix GitHub's unreliable topic tagging
Using LLMs to improve GitHub's topic tagging system for open-source projects. Limited detail provided.
3/23/2026
Show HN: Smriti – Like Git, but for reasoning state
Version control system for LLM/agentic reasoning state, enabling tracking and recovery of reasoning progress across multiple models and sessions.
3/23/2026
Inspect – Semantic code review. Entity graphs + LLMs
LLM-powered code review tool using entity graphs for risk scoring, identifying critical changes in diffs with 95% recall and 5-67ms latency.
3/23/2026
Hyperagents
Hyperagents framework enables open-ended self-improvement in AI systems by generating and evaluating self-modified variants without fixed meta-level mechanisms.
3/23/2026
Teaching an Agent to Sketch One Part at a Time
Multi-modal language model agent trained via process-reward RL to generate vector sketches part-by-part using novel ControlSketch-Part dataset.
3/23/2026
Learning to Disprove: Formal Counterexample Generation with Large Language Models
Method for fine-tuning LLMs to generate formal counterexamples in mathematical reasoning, complementing proof construction capabilities.
3/23/2026
ItinBench: Benchmarking Planning Across Multiple Cognitive Dimensions with Large Language Models
ItinBench benchmarks LLM agents on planning tasks across multiple cognitive dimensions using travel planning as evaluation medium.
3/23/2026
PA2D-MORL: Pareto Ascent Directional Decomposition based Multi-Objective Reinforcement Learning
PA2D-MORL proposes a multi-objective reinforcement learning method using Pareto ascent for complex decision-making tasks with conflicting objectives.
3/23/2026
PowerLens: Taming LLM Agents for Safe and Personalized Mobile Power Management
PowerLens uses LLMs for personalized mobile power management on Android, applying commonsense reasoning to bridge semantic gap between user activities and battery optimization.
3/23/2026
HyEvo: Self-Evolving Hybrid Agentic Workflows for Efficient Reasoning
HyEvo: automated workflow generation framework for heterogeneous agentic systems combining LLMs with symbolic reasoning.
3/23/2026
A Subgoal-driven Framework for Improving Long-Horizon LLM Agents
Subgoal-driven framework for long-horizon LLM agents handling dynamic content and extended action sequences in web navigation.
3/23/2026
Stepwise: Neuro-Symbolic Proof Search for Automated Systems Verification
Stepwise: neuro-symbolic proof generation framework automating theorem proving for critical systems verification using LLMs.
3/23/2026
Embodied Science: Closing the Discovery Loop with Agentic Embodied AI
Embodied science paradigm using agentic AI for closed-loop scientific discovery through physical experimentation and interaction.
3/23/2026
Utility-Guided Agent Orchestration for Efficient LLM Tool Use
Framework for LLM agent tool orchestration balancing answer quality and execution cost through utility-guided decisions.
3/23/2026
Experience is the Best Teacher: Motivating Effective Exploration in Reinforcement Learning for LLMs
Research on effective exploration strategies in reinforcement learning for LLM reasoning with rubric-based rewards.
3/23/2026
DIAL-KG: Schema-Free Incremental Knowledge Graph Construction via Dynamic Schema Induction and Evolution-Intent Assessment
DIAL-KG: schema-free knowledge graph construction framework with dynamic schema induction for streaming data.
3/23/2026
Pitfalls in Evaluating Interpretability Agents
Research on evaluation pitfalls for autonomous interpretability agents using LLMs at increasing autonomy levels.
3/23/2026
Learning Dynamic Belief Graphs for Theory-of-mind Reasoning
Dynamic belief graphs approach for theory-of-mind reasoning with LLMs, modeling evolving beliefs for applications in disaster response, emergency medicine, and human-in-the-loop autonomy.
3/23/2026
Grounded Multimodal Retrieval-Augmented Drafting of Radiology Impressions Using Case-Based Similarity Search
Multimodal RAG system for automated radiology report generation combining contrastive learning with retrieval-augmented generation to reduce hallucinations and improve clinical grounding.
3/23/2026
L-PRISMA: An Extension of PRISMA in the Era of Generative Artificial Intelligence (GenAI)
L-PRISMA extends systematic review framework with LLMs to automate literature screening and data extraction, improving efficiency of evidence synthesis workflows.
3/23/2026
When Prompt Optimization Becomes Jailbreaking: Adaptive Red-Teaming of Large Language Models
Research on adaptive adversarial attacks against LLM safeguards, showing iterative prompt optimization can bypass existing defenses in realistic attack scenarios.
3/23/2026
DuCCAE: A Hybrid Engine for Immersive Conversation via Collaboration, Augmentation, and Evolution
DuCCAE system addresses latency in conversational AI by parallelizing lightweight and heavy-tail tasks (search, generation) to maintain responsive dialogue with tool invocation.
3/23/2026
GeoChallenge: A Multi-Answer Multiple-Choice Benchmark for Geometric Reasoning with Diagrams
GeoChallenge: 90K multi-choice geometry proof problems with diagrams for evaluating LLM symbolic reasoning and multi-step proofs.
3/23/2026
A comprehensive study of LLM-based argument classification: from Llama through DeepSeek to GPT-5.2
Comprehensive evaluation of LLMs (Llama, DeepSeek, GPT-5.2) for argument classification and mining, comparing against traditional machine learning.
3/23/2026
LARFT: Closing the Cognition-Action Gap for Length Instruction Following in Large Language Models
LARFT method improves LLM control over output length via length-aware fine-tuning, addressing cognition-action gap in instruction following.
3/23/2026
MAPLE: Metadata Augmented Private Language Evolution
MAPLE framework for differentially private LLM fine-tuning via synthetic data generation when only API access available, enabling privacy-preserving adaptation.
3/23/2026
Breeze Taigi: Benchmarks and Models for Taiwanese Hokkien Speech Recognition and Synthesis
Introduces Breeze Taigi benchmark and models for Taiwanese Hokkien speech recognition and synthesis with reproducible evaluation methodology.
3/23/2026
HATL: Hierarchical Adaptive-Transfer Learning Framework for Sign Language Machine Translation
Proposes hierarchical adaptive-transfer learning framework for sign language machine translation addressing dataset scarcity and domain gaps.
3/23/2026
The {\alpha}-Law of Observable Belief Revision in Large Language Model Inference
Identifies multiplicative scaling law governing probability revision in LLMs using chain-of-thought and self-reflection mechanisms.
3/23/2026
How Motivation Relates to Generative AI Use: A Large-Scale Survey of Mexican High School Students
Survey of 6,793 Mexican high school students examining relationships between motivational profiles and generative AI adoption in math/writing.
3/23/2026
Generative Active Testing: Efficient LLM Evaluation via Proxy Task Adaptation
Proposes generative active testing method for efficient LLM evaluation using proxy task adaptation to reduce annotation costs for benchmarks.
3/23/2026
When the Pure Reasoner Meets the Impossible Object: Analytic vs. Synthetic Fine-Tuning and the Suppression of Genesis in Language Models
Philosophical investigation of fine-tuning LLMs on contradictory entities using Kantian and Deleuzian frameworks with ontological analysis.
3/23/2026
Probing to Refine: Reinforcement Distillation of LLMs via Explanatory Inversion
Introduces reinforcement distillation framework using explanatory inversion to improve reasoning transfer from large to smaller student LLMs.
3/23/2026
Full-Stack Domain Enhancement for Combustion LLMs: Construction and Optimization
Proposes full-stack domain enhancement for combustion science LLMs to reduce hallucinations and enforce physical conservation laws.
3/23/2026
A Human-Centered Workflow for Using Large Language Models in Content Analysis
Presents comprehensive workflow for using LLM APIs in content analysis tasks including annotation, classification, and summarization via systematic methodology.
3/23/2026
LSR: Linguistic Safety Robustness Benchmark for Low-Resource West African Languages
LSR benchmark measures safety alignment degradation in LLMs across low-resource West African languages, revealing refusal mechanism failures.
3/23/2026
CURE: A Multimodal Benchmark for Clinical Understanding and Retrieval Evaluation
Introduces CURE, a multimodal benchmark evaluating clinical understanding and evidence retrieval in multimodal LLMs for medical diagnostics.