AI SIGNAL
Curated frontier AI & tech news, refreshed every four hours. Aggregated from Hacker News, GitHub trending, and Hugging Face / Replicate releases.
[?] how ranking works
- 501+ exceptional
- 100–500 strong
- 51–99 mid
- 30–50 fresh
- <30 quiet
- live — currently trending
- cooling — losing momentum (6–18h)
- frozen — no longer trending; ranking decays over time
Composite ranking: trend velocity + log(raw signal) − staleness penalty. The bar shows relative strength within the current batch, higher = stronger overall position.
Gemini 4 Neon
Google's Gemini 4 Neon is a frontier-level AI model released Nov 18, 2026, offering fast end-to-end task completion for software engineering, knowledge work, and science workflows, priced at $3 per million input tokens with a 1 million-token context window.
Unauthenticated remote code execution in Nginx
Two critical unauthenticated RCE vulnerabilities (CVE-2026-123633 and CVE-2026-123007) in NGINX require HTTP/2 to be enabled and affect versions dating back to 2015, allowing attackers to execute code and compromise servers.
Seven new models at scalable price-performance
Microsoft announced seven new world-class AI models at Microsoft Ignite 2026, focusing on scalable price-performance improvements.
Inside Deutsche Bank
A major investigation reveals Deutsche Bank's flawed culture and systemic blind spots through a leak of 15+ million files, exposing scandals involving Jho Low, Russian oligarchs, and weapons financing.
Genie 4 is now available
Google DeepMind released Genie 4, a new world simulation model with real-time audio and editing, available to Google AI Ultra subscribers via Project Genie, alongside Genie 3 upgrades and a community gallery.
Scottish HTTP Codes
A humorous Scottish translation of HTTP status codes, e.g., 404 is 'Nae fuckin here, pal' and 500 is 'Ah've fucked it'.
DNS resolvers are the wrong place to fight piracy
Swiss non-profit Quad9 is leaving France on Nov 24, 2026, after a Paris court fined it for refusing to block pirate sports domains, forcing it to withdraw to protect its privacy and security-focused service.
Russia reportedly planned mass-casualty attack in Poland to be blamed on Ukraine
Poland thwarted a Russian false-flag operation near the Kaliningrad border intended to blame Ukraine and stop aid, involving mass casualties and a Gleiwitz-style attack, by reinforcing security forces and taking pre-emptive action.
AI agents ask better questions than KPMG accountants in audit trial
KPMG's Project Exponential used 100 AI agents to audit an FTSE 250 company, finding the AI asked better questions and investigated deeper than human auditors, though human oversight remains essential.
Advancing chip design with OpenAI models
OpenAI's models improved custom chip Jalapeño by 13% area and 10% area in vector/matrix units, and achieved up to 14% PPA gains on AMD GPU IP by rewriting SystemVerilog code, with GPT-6.2 Astra outperforming previous models on RISC-V cores.
The Cambridge Dictionary has named ‘human’ as the Word of the Year 2026
Cambridge Dictionary named 'human' Word of the Year 2026 due to global debates on human identity amidst AI advancements, with figures like Bill Gates and Pope Leo XIV defending humanity against dehumanization and machine replacement.
KVM: x86: Unconditionally reset MMU context when emulating INIT (CVE-2026-99312)
KVM fixes a bug where an INIT instruction could leave a stale MMU context if paging was disabled, forcing a reset in all scenarios including SMM and real mode.
Energy drinks linked to 11 times more reports of cardiac arrest
Researchers found FDA reports of cardiac arrest, heart attacks, and arrhythmias were 11, 4.5, and 4 times more likely with caffeine/stimulant products, with 45% of cases involving adults 18-34, though the study notes this does not prove causation.
DaisyRose-27B: 1,200-task RL took Qwen3.8 from 5.6% to 12% on Terminal-Bench 4.1
DaisyRose-27B, a Qwen3.8-27B model, improves Terminal-Bench 4.1 from 5.6% to 12% using 1,200 RL tasks and 144 H200 GPUs.
F-22 Next-Generation micro–Data Transfer Cartridge (mDTC)
The U.S. Air Force is seeking suppliers for PC cards or CF cards with 256MB+ storage capacity for F-22 mDTC modernization.
Interviewed at a YC startup, found out they used my LinkedIn without permission
AI is killing All About Berlin
The author reports losing 75% of traffic to Google AI Overviews, which now show AI-generated answers with ads instead of direct links, threatening the sustainability of his 8-year Berlin guide site.
Show HN: ExitPath – Open-source ways out of Slack, Notion and Microsoft 365
ExitPath is an open-source tool that rates migration difficulty for 16 apps like Slack, Notion, and GitHub, suggesting self-hosted alternatives such as Mattermost, AppFlowy, and Forgejo.
Codeberg.org Not Reachable
Codeberg.org is unreachable due to a domain suspension by Namesilo, with the team waiting for the abuse team's final decision and choosing not to enable a temporary workaround on codeberg.eu.
Show HN: Take Voyager 1 apart, down to a single memory chip
This is an interactive 3D visualization of the Voyager 1 spacecraft allowing users to explore its components and timeline from launch to interstellar space.
CrossPoint Reader
CrossPoint 2.0 is a free, open-source e-reader software for compact devices featuring custom fonts, OPDS 2.0, Bluetooth page turners, and offline dictionaries.
The Magic of Partial Evaluation
Partial evaluation evaluates static parts of code at compile time, leaving dynamic parts for runtime, and can generate compilers for free.
The x86 CPU: From 4004 to 486
This article traces the x86 architecture's evolution from the 1971 4004 calculator chip to the 1989 486, explaining how historical decisions and legacy constraints created the arbitrary instruction set seen today.
NGINX2Root by Xint – From Nginx visitor to root of the server host
NGINX2Root exploits unauthenticated access to Nginx to gain root privileges via CVE-2026-123007, CVE-2026-123634, CVE-2026-123633, CVE-2026-121810, and CVE-2026-113425.
Energy-Based Reasoners: 385x more efficient inference
Energy-Based Reasoners (EBRs) use gradient descent on a learned energy function to guide reasoning, achieving 99.9% Sudoku accuracy with 1.8M parameters and 345× less compute than the best model.
A Primer on Fourier Series (2025)
A new textbook on Fourier Series by Dipierro, Pfefferlé, and Valdinoci is suitable for undergraduate and graduate courses and includes exercises with full solutions.
Show HN: Type Fighter 2
Type Fighter 2 is a word typing game where players choose words using arrow keys and confirm with Enter or Space, featuring a score and high score system.
Can Peer Review Survive AI-Generated Text? [pdf]
Federal Reserve economists argue AI detectors should flag AI-generated referee reports, which they treat as unacceptable plagiarism, to save peer review from collapse.
Talos Linux
Talos Linux is an immutable, API-driven OS for Kubernetes with no SSH or shell, offering free open-source and paid enterprise tiers with FIPS compliance, indemnity, and 24/7 support.
Why is Gemini 4 Carbon taking so long?
Gemini 4 Carbon is significantly slower than Opus and Astra because it performs excessive, unnecessary work, taking about seven times as many steps and auditing code far beyond what the task requires.
Claude was trained on protein sequences at scale
Claude Opus 5.5 and Fable 5.2 outperform other models on predicting next amino acids in little-studied bacterial proteins, suggesting they were trained on protein sequences at scale.
Voyager 1 hits one light day from Earth
NASA invites media and public to join Voyager 1 one-light day activities as the probe reaches a distance of one light day from Earth.
Why Fibonacci?
The article explores Fibonacci spirals, with a commenter noting that 72, halfway between 55 and 89, produces an interesting visual effect.
Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.
n8n is an open-source workflow automation tool hosted on GitHub, allowing users to connect various apps and services to automate tasks.
The open source coding agent.
GitHub project anomalyco/opencode with 227,763 stars.
An AI skill that provides design intelligence for building professional UI/UX across multiple platforms.
An AI skill for generating professional UI/UX design systems and reasoning rules across multiple platforms, built with Python 3.x and featuring 67 UI styles.
Beta: The truly free, open-source, local-first, cross-platform video editor. Scriptable through a JSON-RPC API & MCP.
Concat is a free, open-source, cross-platform video editor replacing CapCut with no watermarks, featuring multi-track editing, auto-captions, and text-to-speech, running entirely locally on your machine.
dsh-purge Community · DeepSeek Harness 社区维护版:宿主兼容、状态诊断、提示词管理、更新与本机备份回滚。基于小杨 / YuJunZhiXue 的 dsh-purge。
dsh-purge Community is a local DeepSeek Harness plugin (v1.6.86) that reduces friction for agents with 86 patches and 25 state checks, enabling local image generation via Qwen Studio.
Bring PS5 GTA6 to 120FPS with DLSS 5. An advanced Windows video player and capture enhancement tool featuring real-time preview for videos, images, and capture card sources. Combine AI-powered Super Resolution, DLSS5 image enhancement, and Frame Generation to deliver a next-generation visual experience.
Veyra-NRVideo is a real-time Windows GPU video enhancement tool supporting super resolution, denoising, HDR, and frame generation for game streaming and local video.
! | Benchmark | Qwen3.
A LoRA adapter for Qwen3.8-27B trained with RL for the pi coding agent, achieving 64.7% on SWE-bench Pro and 56.6% on Terminal-Bench 2.1 with 4.5x fewer tokens than the base model.
Use Claude Code, Codex, Antigravity, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)
Free Claude Code is an open-source tool aggregating 49 free AI coding providers and 9 agents like Claude Code and Cline, offering a unified UI, terminal optimizations, and support for multiple platforms.
MohammadJRanjbar/ParsVoice-XTTS
ParsVoice XTTS-v2 is a Persian TTS model fine-tuned on 2,200 hours of data from 1,800+ speakers, achieving 3.6 MOS and 4.0 SMOS scores, and supports zero-shot voice cloning with a 6-15 second reference clip.
Never stop coding. Free MIT AI gateway: one endpoint, 375 providers (150+ free), 1,400+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 900+ contributors
OmniRoute aggregates 375 AI providers and 150+ free tiers into one endpoint, using RTK compression to save 15-95% tokens and offering 1.62B free tokens monthly across 34 pool keys.
82M parameters, 16 languages, 118 voices, 24 kHz. ONNX with int8 weights (**82.7 MB**) or int4 weights (**47.5 MB**); both run on CPU, no PyTorch needed. And the float32 PyTorch weights.
KittenTTS-2-Mini is a 82M parameter TTS model supporting 16 languages and 118 voices, available as lightweight ONNX (int8/int4) or PyTorch weights, with a pip install and simple API for generating audio.
Your Android phone and your PC, working as one. Windows and Linux. Local-first, end-to-end encrypted, no account.
NectarLink is a free, open-source Android-to-Windows bridge app that syncs notifications, messages, clipboard, files, and screen mirroring locally without cloud storage, available for Windows, Linux, and Android.
Five GGUF builds of Hy4-flash: tencent/Hy4-flash | file | model size | bpw (overall) | bpw (excl.
Hy4-flash-GGUF provides five quantized builds (Q4_K_M, SPARK-4Bit, SPARK-4Bit-Fast, Q3_K, IQ2-AH) for efficient large model compression, with specific benchmarks and usage instructions for Mac M5 Max and RTX Spark devices.
! The GGUF version of harness-hub/Qwen3.
A LoRA adapter for Qwen3.8-27B optimized for llama.cpp, achieving 64.7% SWE-bench Pro and 55.8% Terminal-Bench 2.1 with 4.5× fewer tokens by keeping thinking off.
25 MB lightweight cross-platform database client for 100+ databases, including MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, SQL Server, and Dameng. Built-in AI, MCP Server, CLI, desktop and Docker. | 轻量级跨平台数据库管理工具,支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、达梦等 100+ 数据库,提供桌面端、Docker、CLI、内置 AI 助手和 MCP。
GitHub project t8y2/dbx with 30,701 stars.
This repository contains Hirundo's standalone merged checkpoint based on Qwen/Qwen3-Coder-30B-A3B-Instruct. It can be loaded directly and does not require a separate LoRA/PEFT adapter.
Hirundo's behavioral unlearning of Qwen3-Coder-30B reduces censorship from 46.6% to 4.0% on CCPC-500 and improves secure coding across personas, with no loss in general coding ability.
**Huge thanks to Hugging Face for generously sponsoring the compute used to abliterate these models! This was done on HF Jobs, which makes it incredibly easy to run tasks such as these :)** Join the...
Abliterated DeepSeek-V4.1-Flash-1017 GGUF quants with refusal behavior removed via directional ablation, available in Q8_0, Q4_K_M, Q3_K_M, and Q2_K sizes, requiring a specific llama.cpp branch to run.
Signal briefs — our take on the top picks
- HNWhy is Gemini 4 Carbon taking so long?
- HNGemini 4 Neon
- HNEnergy-Based Reasoners: 385x more efficient inference
- HNAI agents ask better questions than KPMG accountants in audit trial
- HNPoland says it thwarted false flag operation near border with Kaliningrad
- HNSeven new models at scalable price-performance
- HNInside Deutsche Bank
- HNUnauthenticated remote code execution in Nginx
- HNQuad9 Is Leaving France: Why DNS Resolvers Are the Wrong Place to Fight Piracy
- HNCodeberg.org Not Reachable
- HNHarvard researchers find AI coding agents shifted the bottleneck to code review
- HNInterviewed at a YC startup, found out they used my LinkedIn without permission