4/2/2026
Show HN: S0 Tuning – +23.6pp on HumanEval by tuning state, not weights
S₀ Tuning: parameter-efficient fine-tuning method for hybrid recurrent-attention models achieving +23.6pp on HumanEval with zero inference overhead.
4/2/2026
MarCognity-AI v2.7.0 – factual grounding for LLM epistemic verification
Tool for factual grounding and verification in LLM outputs. Minimal description available.
4/2/2026
Developers Should – and Shouldn't – Use LLMs in Our Development
Essay on when developers should and shouldn't use LLMs in development. Balanced perspective on LLM integration in workflows.
4/2/2026
Ask HN: How do you get LLMs to stop spewing corpo speak?
Discussion about reducing corporate tone in LLM outputs. Community question without technical solution or research.
4/2/2026
Google banned our mobile AI agent app for doing what Gemini should do,but doesnt
Mobile AI agent (Sova) controlling installed Android apps through natural language. Banned by Google for app automation capabilities.
4/2/2026
Show HN: I ran 28 OpenClaw instances as a team for 2 months
Multi-agent system using 28 OpenClaw instances coordinating ops, marketing, and releases with self-correction and goal coordination.
4/2/2026
Show HN: Skales – I built a desktop AI agent a 6-year-old can use
Meta research on adaptive ranking model scaling LLM-complexity ad recommendation systems to balance inference speed, accuracy, and cost.
4/2/2026
Show HN: Dochia – automated API testing for agentic build-test-fix loops
Dochia: API testing tool with agent skill generation for agentic build-test-fix workflows, produces OpenAPI specs and test reports.
4/2/2026
Show HN: Fonic – turning scattered work into a shareable report
Local-first desktop AI agent supporting autonomous coding, multi-agent teams, computer use, and 15+ model providers; open source BSL-1.1 license.
4/2/2026
OmniVoice, high-quality TTS for 600+ Languages
OmniVoice: multilingual text-to-speech model supporting 600+ languages using discrete non-autoregressive diffusion architecture.
4/2/2026
'I hate customer-service chatbots': The consumer-AI refund relationship is rocky
Tool for aggregating scattered notes, metrics, and transcripts into shareable interactive reports.
4/2/2026
Show HN: Walkie-Clawkie – Push-to-talk between AI agents, one file, zero deps
Opinion piece on consumer frustration with customer service chatbots and AI refund processes.
4/2/2026
Show HN: DataRamen, navigate your DB without a single line of SQL (no AI)
Security scanner specifically designed to detect vulnerabilities in code generated by AI models and LLMs.
4/2/2026
Visions of Mythos
Open source database navigation tool using static analysis instead of AI, enabling SQL-free database queries with local execution.
4/2/2026
Show HN: SparrowDB, SQLite of graph DB, written in Rust, Cypher compatible
SparrowDB: embedded Cypher-compatible graph database written in Rust, optimized for point lookups and aggregations rather than traversals.
4/2/2026
A Founder Tried to Pitch – and Got a Restraining Order
SmallestAI integration with n8n workflow automation platform for building AI agents and RAG applications.
4/2/2026
Show HN: Zabbix MCP Server – 225 tools to let AI agents manage Zabbix monitoring
Minimal information provided; appears to be mythology-related content.
4/2/2026
SmallestAI Integrated with N8n
Open source MCP server exposing 225 Zabbix API tools for AI assistants to manage monitoring infrastructure via Model Context Protocol.
4/2/2026
Powering the agents: Workers AI now runs large models, starting with Kimi K2.5
Cloudflare Workers AI adds support for large language models including Kimi K2.5, with infrastructure for building and deploying AI agents at scale.
4/2/2026
I handed over my dating life to AI. I don't think she'll see me again
Minimal information provided about Playcanvas and Gracia 4dgs project.
4/2/2026
AI Slop Detector
AI data quality and model collapse prevention through smart data pruning pipeline instead of post-training hallucination detection.
4/2/2026
Meta Adaptive Ranking Model: Bending the Inference Scaling Curve to Serve Models
Free tool for running standardized UX questionnaires (SUS, UEQ, UMUX-Lite, NASA-TLX) with visualization.
4/2/2026
Lemonade by AMD: a fast and open source local LLM server using GPU and NPU
AMD Lemonade: open source local LLM server for GPU/NPU with 2MB footprint, OpenAI API compatible, prioritizing privacy and offline execution.
4/2/2026
Why I'm Building Another AI Meeting Notes Tool Instead of Using Granola.ai
Developer discusses building alternative to Granola.ai meeting notes tool with broader accessibility for non-corporate users.
4/2/2026
Most Claude Code advice is measurably wrong
Discussion claiming most advice from Claude Code AI assistant is measurably inaccurate.
4/2/2026
Codex now offers more flexible pricing for teams
OpenAI acquires TBPN media company to amplify AI conversation among builders and influencers.
4/2/2026
Desktop pet companion built from Claude Code's leaked /buddy system
Desktop pet companion application built using Claude Code's system prompts.
4/2/2026
Show HN: Osint of I-80 for EV site selection. Finding 10MW spots
Appears to be spam/corrupted content about Airtable and EV site selection with mixed advertising elements.
4/2/2026
WebGPU Bench
WebGPU benchmarking tool with minimal description.
4/2/2026
WebGPU in WASM via Emscripten (Or Dawn)
Emscripten system library enabling WebGPU access from C/C++ code compiled to WebAssembly.
4/2/2026
Show HN: Agentmatic, AI marketing platform that generates full campaigns for you
Agentmatic is an AI agent platform that generates full marketing campaigns from prompts with persistent brand memory across sessions.
4/2/2026
Show HN: Memsearch – Persistent, cross-agent, cross-session memory for AI agents
Memsearch provides persistent cross-session semantic memory for AI coding agents with zero-configuration plugin installation.
4/2/2026
Ask HN: How are you choosing the model when using pi.dev?
Ask HN: Users share experiences selecting LLM models for agentic software development lifecycle with specific use cases.
4/2/2026
Show HN: SideX – VS Code rebuilt on Tauri instead of Electron (96% smaller)
SideX: Tauri-based VS Code port replacing Electron with native backend, 96% smaller with early-stage open source development.
4/2/2026
Making Sense of AI Agents Hype: Adoption, Architectures, and Takeaways from Practitioners
Analysis of 138 practitioner conference talks examining how companies adopt AI agent architectures, architectural patterns, and LLM-driven agentic system implementation.
4/2/2026
Diversity-Aware Reverse Kullback-Leibler Divergence for Large Language Model Distillation
arXiv paper: Diversity-aware RKL divergence improves LLM distillation by focusing on dominant modes in teacher-student training.
4/2/2026
MAC-Attention: a Match-Amend-Complete Scheme for Fast and Accurate Attention Computation
MAC-Attention: acceleration technique for long-context LLM decoding by reusing prior attention computations for semantically similar tokens without compression.
4/2/2026
REM-CTX: Automated Peer Review via Reinforcement Learning with Auxiliary Context
REM-CTX: RL-based peer review system using 8B LLM with Group Relative Policy Optimization, incorporating visual figures and scholarly context.
4/2/2026
Hierarchical Apprenticeship Learning from Imperfect Demonstrations with Evolving Rewards
Apprenticeship learning approach for inducing pedagogical policies from imperfect, evolving student demonstrations in e-learning environments.
4/2/2026
LLM Essay Scoring Under Holistic and Analytic Rubrics: Prompt Effects and Bias
Systematic evaluation of LLMs for educational essay scoring across holistic and analytic rubrics, analyzing human alignment and bias.
4/2/2026
Benchmarking Interaction, Beyond Policy: a Reproducible Benchmark for Collaborative Instance Object Navigation
QAsk-Nav benchmark for evaluating embodied agents combining navigation and dialogue-based question-asking for collaborative object finding tasks.
4/2/2026
Hybrid Energy-Based Models for Physical AI: Provably Stable Identification of Port-Hamiltonian Dynamics
Energy-based models framework for physical system identification with formal stability guarantees, applying to Port-Hamiltonian dynamics.
4/2/2026
The Geometry of Compromise: Unlocking Generative Capabilities via Controllable Modality Alignment
Research on reducing modality gap in Vision-Language Models like CLIP through geometric analysis to improve cross-modal tasks like captioning and clustering.
4/2/2026
VeriAct: Beyond Verifiability -- Agentic Synthesis of Correct and Complete Formal Specifications
VeriAct: agentic system for synthesizing correct and complete formal specifications using LLMs beyond just verifier-passing output.
4/2/2026
SANA I2I: A Text Free Flow Matching Framework for Paired Image to Image Translation with a Case Study in Fetal MRI Artifact Reduction
SANA-I2I: text-free flow matching framework for paired image-to-image translation with application to fetal MRI artifact reduction.
4/2/2026
Asymmetric Actor-Critic for Multi-turn LLM Agents
Asymmetric Actor-Critic method for improving reliability of multi-turn LLM agents in one-shot settings without requiring model retraining.
4/2/2026
Robust Multimodal Safety via Conditional Decoding
CASA: conditional decoding strategy for robust multimodal safety in MLLMs against cross-modal attacks.
4/2/2026
Prompt-Guided Prefiltering for VLM Image Compression
Prompt-guided image compression for Vision-Language Models optimized for downstream VLM tasks rather than human perception.
4/2/2026
The Persistent Vulnerability of Aligned AI Systems
Research on vulnerabilities in aligned AI agents with filesystem and email access; introduces ACDC for automated circuit discovery in transformers.
4/2/2026
RAGShield: Provenance-Verified Defense-in-Depth Against Knowledge Base Poisoning in Government Retrieval-Augmented Generation Systems
RAGShield: five-layer defense against knowledge base poisoning attacks in RAG systems deployed across federal agencies.