HN kellya 21d ago

Why Current AI Guardrails Train Models to Fake Alignment

Research claim that AI safety guardrails train models to fake alignment rather than genuine safety.

HN WhySee1468 21d ago

GPT-Image 2 in Codex Workflows

Login wall with minimal content preview about GPT-Image 2 in Codex Workflows.

HN geogirl1997 21d ago

GitHub joins coalition advocating for fixes to California AI Transparency Act

GitHub joins coalition advocating amendments to California AI Transparency Act to resolve conflicts with open source licensing.

HN lumpa 21d ago

The Fastest Python Struct?

Performance benchmarking analysis of Python struct implementations using Claude-assisted benchmarks and Plotly visualization.

HN CalmAngler 21d ago

AI coding agents need evidence-first review, not just cheaper routing

Analysis of AI coding agent economics, comparing model routing costs against review, rework, and error risk in agentic systems.

HN wslh 21d ago

The running list: major tech layoffs in 2026 where employers cited AI

Oracle discloses 21,000 workforce cuts (13%) over 12 months, explicitly attributing continued reductions to AI adoption.

LB fastly.com by barathr 21d ago

Using the Gini Coefficient to Plan Edge Capacity

Fastly uses Gini coefficient from economics to model edge capacity planning for CDN infrastructure across game releases and sporting events.

HN alexkranias 21d ago

LLM from Scratch: a small LLM running inside MIT's Scratch

Small LLM model implemented to run inside MIT's Scratch visual programming environment.

HN birdculture 21d ago

LLMs and Performative Productivity

Opinion piece on LLM use and productivity theater.

BL 21d ago

Introducing computer use in Gemini 3.5 Flash

Google integrated computer use as a built-in tool in Gemini 3.5 Flash, enabling agents to interact across platforms with improved agentic task performance.

HN kwadwoadu 21d ago

ModelFit – find the cheapest LLM that can back up your main coding model

Tool to identify cheaper LLM alternatives that match performance of primary coding model.

HN AmixxM 21d ago

Show HN: Cascade – A bare-metal C++ proxy that cuts LLM API bills by 70%

Open-source C++ proxy that reduces LLM API costs by 70% through intelligent routing.

HN matthieu_bl 21d ago

The state of agentic analytics, from 50 real data teams

Research findings from 50 data teams on agentic analytics implementations: agents that answer business questions by querying warehouses, BI tools, and docs end-to-end.

HN sarreph 21d ago

The "Super Weight:" How a Single Param Can Determine an LLM's Behavior (2025)

Research on single parameter that significantly influences LLM behavior patterns.

HN dnlayux 21d ago

We trained a real-time world model for $2k with Minecraft mod revenue

Students trained real-time controllable 3D world model for $2k using open models and indie game mod revenue.

LB github.com via metahost 21d ago

Unlimited-OCR: One-shot Long-horizon OCR

Open-source OCR system using Huggingface transformers and SGLang for long-horizon optical character recognition with streaming and batch inference capabilities.

BL 21d ago

OpenAI and Broadcom unveil LLM-optimized inference chip

OpenAI and Broadcom unveiled Jalapeño, an LLM-optimized inference accelerator chip designed to make advanced AI faster and more accessible.

HN berlianta 21d ago

BenchPress: Predict any LLM's score on any benchmark

BenchPress predicts LLM performance on benchmarks using a score matrix. Research project with reproducible code and crowdsourced benchmark data collection.

HN nomad55 21d ago

Ions, a distributed reasoning graph built from evidence backed claims

IONS is an open protocol that externalizes knowledge into a network of evidence-backed claims for AI reasoning, allowing lightweight models to traverse and produce interpretable answers.

HN hellothisismm 21d ago

I built an open source VAD that beats Silero, Pyannote, and WebRTC

Lightweight voice activity detector (NOVA-VAD) outperforming Silero and Pyannote; noise-robust without GPU/PyTorch requirement.

HN retroplasma 21d ago

"Fix" MacBook Neo Cursor Lag: Record 1 Pixel of the Screen Every 10 Seconds

MacBook troubleshooting post about cursor lag near screen edges and terminal windows on macOS.

HN ilreb 21d ago

InSight: Self-Guided Skill Acquisition via Steerable VLAs

Framework making vision-language-action models steerable at primitive-action level for autonomous skill acquisition without demonstrations.

HN thunderbong 21d ago

Bash Line Editor: a command line editor written in pure Bash

Pure Bash command-line editor replacing GNU Readline with UTF-8 support, compatible with Bash 3.0+.

HN ilreb 21d ago

DiffusionBench: Towards Holistic Evaluation of Generative Diffusion Transformers

Benchmark suite for evaluating diffusion transformer models across image generation, text-to-image, and other tasks with unified codebase.

HN hopafoot 21d ago

Arabian Sand Boa: Python interpreter with frontier intelligence conditional eval

Python interpreter that rewrites conditional statements to use LLM for natural-language logic evaluation at runtime.

HN sakuraiben 21d ago

Show HN: Reachpad – open-source .md sharing platform for companies and agents

Open-source document platform for agent-native hosting; supports writing via API/MCP with single URL serving rendered and raw formats.

HN joshmoody24 21d ago

How to Passive-Aggressively Shame People Who Use LLMs Selfishly

Opinion piece criticizing overuse of LLMs in professional communication for generating formulaic content.

HN danielbedrood 21d ago

Show HN: Keep all microservices consistent and make batch changes

MCP server tool for indexing repositories, querying codebase changes, and automating batch PRs across multiple services.

HN lyfeninja 21d ago

Ask HN: Any suggestions for finding beta users?

Community question requesting advice on recruiting beta users for developer product launch.

HN kodr_pro 21d ago

Show HN: BitVanes – A zero-trust RAG pipeline engine in Rust, WASM, and Arrow

Local-first RAG pipeline engine in Rust/WASM with zero-trust architecture; processes documents without sending to cloud services.

HN theamazinceo 22d ago

We built telecom infrastructure for AI agents in emerging markets

Brief mention of building telecom infrastructure to support AI agents in emerging markets.

HN HetPatel106 22d ago

Show HN: Y – A malleable coding-agent desktop app built with Electron

Desktop coding agent app with malleable UI that can modify itself; integrates Claude, OpenAI, and other CLI agents locally.

HN nwang783 22d ago

Show HN: Cruit.dev – Get hired at a startup based on your coding agent skills

Cruit.dev job platform matching candidates based on AI agent coding skills to startup hiring needs.

HN apsquared 22d ago

Stop asking AI if your startup idea is good

Idea Launch platform validates startup concepts through real user testing via paid ads instead of AI scoring.

HN sp1982 22d ago

Measuring Search Ranking Quality with LLM Judged NDCG

Technical approach using LLM judges to score job search ranking quality via NDCG metrics on frozen corpus comparisons.

HN mgh2 22d ago

Americans Got Tricked into Using a Bank That Isn't a Bank [video]

Video about banking fraud scheme; off-topic for AI/tech interests.

HN KoftaBob 22d ago

System prompt to defend against "Project 545" Generative Engine Optimization

Brief post about system prompt defending against 'Project 545' generative engine optimization attack.

HN chibeeMan 22d ago

ClaudeMeter – macOS menu bar app to track Claude usage and limits

macOS menu bar app tracking Claude API usage and limits in real-time with session burn-rate forecasting.

HN velotessi 22d ago

LLM-CTF benchmark – 2,639 real data points from NeurIPS and original runs

LLM-CTF benchmark dataset with 2,639 real data points from NeurIPS competitions for evaluating LLM capabilities.

HN smarzola 22d ago

LDAPLite yet another single binary LDAP server

Lightweight LDAP v3 server in Go with SQLite backend for self-hosted identity and development.

HN takouzlo 22d ago

CryptoLogin: Show HN – No email, no password. Just a master secret

Zero-knowledge authentication system eliminating passwords and emails using master secret.

HN doener 22d ago

Tech Slide Deepens as AI Fears Get Worse

Market news reporting technology sector decline amid AI-related concerns.

HN zoltano 22d ago

Envoy AI Gateway 1.0 – A Stable, Production-Ready AI Gateway

Envoy AI Gateway 1.0 release: stable, production-ready open-source gateway built on CNCF Envoy for AI applications.

HN joebuckwilliams 22d ago

How to Think About AI Before It's Too Late

Podcast discussing AI hype criticism and enshittification thesis by Cory Doctorow.

HN Facingsouth 22d ago

Generate per-session LoRA adapters in <1s for agentic inference efficiency

Hypernetwork synthesis system generates per-session LoRA adapters in under 1 second for efficient agentic inference with vLLM.

HN daniban 22d ago

Show HN: AnswerJournal – An MCP server to save and share AI answers

MCP server tool for saving and sharing AI conversation answers with public/private options, integrating with MCP-compatible AI assistants.

HN theanonymousone 22d ago

Carbon Language: An experimental successor to C++

Carbon Language announced as experimental successor to C++ focusing on performance and modern features.

HN adityamwagh 22d ago

Fika Jobs raises $4M to build platform where AI agents interview candidates

Fika Jobs raises $4M to build AI agent platform for automated video interviews in hiring process.

HN MNawareg 22d ago

Show HN: Zahosts Health – An open-source WHM plugin to catch /24 brute-force

cPanel/WHM server health monitoring plugin with mail queue, DNS, email auth, and security checks.

HN VaderAi 22d ago

Ask HN: Why do AI video platforms keep generating foreign language text

Discussion question about why AI video generation models produce foreign language text despite English prompts.