AI SIGNAL

Curated frontier AI & tech news, refreshed every four hours. Aggregated from Hacker News, GitHub trending, and Hugging Face / Replicate releases.

refresh in —
[?] how ranking works
heat tiers
  • 501+ exceptional
  • 100–500 strong
  • 51–99 mid
  • 30–50 fresh
  • <30 quiet
freshness
  • live — currently trending
  • cooling — losing momentum (6–18h)
  • frozen — no longer trending; ranking decays over time
Σ score

Composite ranking: trend velocity + log(raw signal) − staleness penalty. The bar shows relative strength within the current batch, higher = stronger overall position.

01
math216.wordpress.com12h ago764 comments1,712

Hurried Announcement: Alessandro Masullo GRH

Alessandro Masullo's proposed proof of GRH, reviewed by Brian Conrad, is being released for public verification after being incorporated into a draft.

02
thinkingmachines.ai10h ago133 comments783

Inkling 2 Mini

Inkling 2 Mini is a 26.4B parameter, looped transformer model that runs on consumer laptops and leads open-weights models under 100B on coding benchmarks like DeepSWE v1.1 (48.2%).

03
alpo.ge16h ago1,642 comments2,433

Levent Alpöge claims proof of the Generalized Riemann Hypothesis [pdf]

An Anthropic employee claims to have proven the Riemann Hypothesis with help from Claude, providing Lean formalization and a proof sketch.

04
borretti.me6h ago161 comments169

Re: Riemann

A mathematician claims to have proven the Riemann Hypothesis with help from Claude, sparking philosophical reflections on human-AI collaboration and the future of science.

05
poolp.org14h ago556 comments771

I'll never buy Framework hardware again

A user reports a broken Framework Laptop 16 that support blamed on an OS, leading to a 3-month ordeal of useless scripts, a wrong part replacement, and a 'No Fault Found' conclusion despite clear diagnostic proof of a faulty keyboard module.

06
marketscreener.com10h ago133 comments256

US SEC's Nvidia probe is examining potential accounting issues – sources

The US SEC is reportedly investigating Nvidia for potential accounting issues, according to sources.

07
scottaaronson.blog10h ago215 comments238

The stochastic parrots have apparently proved the Riemann Hypothesis

Scott Aaronson discusses AI's recent progress in proving the Riemann Hypothesis, noting that while the 17.4 million line Lean proof is impressive, mathematicians will likely remain skeptical until they can understand the underlying ideas themselves.

08
manidoraisamy.com12h ago203 comments337

We promoted developers into architects without informing them

The author argues that AI coding tools lack the architectural judgment to push back on requirements, leading to performance issues; therefore, developers promoted to architects must now design the datamodel and constrain AI rather than just writing code.

09
ninlive.com8h ago41 comments161

NINLIVE: The Nine Inch Nails live archive

NINLIVE is a fan-run archive of Nine Inch Nails live recordings, photos, and setlists from 1988-2026, now featuring streaming capabilities and a growing database of uncirculated shows.

10
terrytao.wordpress.com13h ago153 comments374

Unforced Navier-Stokes Singularity and the Generalized Riemann Hypothesis

Terry Tao reports on three major results: a human-readable proof of finite-time blowup for unforced 3D Euler equations, a proof of blowup for unforced 3D Navier-Stokes equations, and a Lean-only proof of the Generalized Riemann Hypothesis, all using heavy AI assistance.

11
cnbc.com8h ago59 comments126

Nasdaq erases all 2026 gains

Nasdaq futures drop as tech sector suffers worst weekly decline since 2025, VIX climbs above 30, and Fed Governor Cook warns inflation risks.

12
motherjones.com13h ago198 comments271

Google workers tell their employer not to punish schools that don't want AI

Hundreds of Google workers are petitioning leadership to retract a policy that would block schools from using Google Search if they disable the company's Gemini AI tools, arguing the move punishes educators for resisting AI.

13
arxiv.org14h ago113 comments359

Language Models Can Rewrite Their Own Weights

A 27B language model rewrites its own weights to improve accuracy without training or gradients, surpassing its previous performance and repairing damaged copies of itself.

14
nbcnews.com12h ago188 comments184

Anthropic shares tank after IPO as the Nasdaq erases 2026 gains

Anthropic shares plummeted after its IPO, causing the Nasdaq to erase its 2026 gains.

15
blog.mozilla.org15h ago199 comments336

Why there are still no non-WebKit browsers on iOS

Mozilla claims Apple's iOS browser engine rules are unworkable, forcing them to build a separate app and move user data, which prevents any non-WebKit browser from launching despite EU and Japan regulations.

16
signoregalilei.com10h ago54 comments115

And yet, it moves

The author reviews the Broadway musical "Galileo" and muses on the recent AI-solved Riemann Hypothesis, contrasting human achievement with machine intelligence and lamenting the end of human discovery.

17
masteringemacs.org10h ago25 comments140

DreamCoder: Coding harness and IDE from the author of Mastering Emacs

DreamCoder is a new AI coding harness and IDE from the author of Mastering Emacs, featuring a tiling window manager and browser-based interface.

18
winedroid.soham.sh11h ago36 comments113

Winedroid – run Android apps without Android (WASM demo)

Winedroid is a Rust-based project running real Android APKs in the browser via WebAssembly, though it is currently slow and still in development.

19
ankitg.me12h ago48 comments121

Vibecoding an inference chip

The author vibe-coded an FPGA to run an LLM, achieving 213 tokens per second (33x faster) using AI agents and minimal human guidance.

20
esa.int7h ago12 comments54

Sir Brian May adds new dimension to Hera asteroid

Sir Brian May created stereoscopic 3D images of the Didymos asteroid using ESA's Hera spacecraft, revealing details like the equatorial ridge.

21
slopstrap.mikelacher.com15h ago24 comments112

Slopstrap – Make your site look hastily vibecoded

Slopstrap is a CSS framework that adds AI-generated aesthetics like animated gradients, glowing borders, and shimmer effects to websites, available via simple CSS/JS inclusion.

22
a90.com13h ago26 comments63

AI is giving everyone the same advice

A study of 65 AIs found they consistently give the same advice, like recommending 'Man's Search for Meaning' for life-changing books or index funds for investing, because this 'default' behavior is hard-coded during training.

23
github.com3h ago178,066

f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.

A massive open-source library of AI prompts for ChatGPT, Claude, and others, featuring an interactive book, a kids' game, and self-hosting options.

24
github.com16h ago64,461

A multi-platform proxy client based on ClashMeta,simple and easy to use, open-source and ad-free.

FlClash is an open-source, ad-free multi-platform proxy client based on ClashMeta supporting Android, Windows, macOS, and Linux with Material You design and WebDAV sync.

25
github.com14h ago82,237

在 dsh 里装上这个插件即可,无需登录、注册或填 API Key,就能使用包括 DeepSeek V4.1 Flash、Kimi K3 在内的前沿模型——完全免费,不限量。 All you do is install this plugin in dsh: no login, no sign-up, no API key — the frontier models are just there, DeepSeek V4.1 Flash and Kimi K3 among them. Completely free, with no usage cap.

A free DeepSeek Harness plugin for dsh that provides access to frontier models like DeepSeek V4.1 Flash and Kimi K3 without API keys, featuring zero dependencies, hot reload, and local usage tracking.

26
github.com4h ago72,283

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.

27
github.com2h ago39,569

Every agent's model. One place. Codex on DeepSeek, Claude Code on Kimi, from the menu bar.

Magpie is a local gateway app that lets you switch AI models across different agents like Claude, ChatGPT, and Gemini, automatically routing to backup accounts when quotas run out.

28
github.com11h ago29,089

dsh-go — DeepSeek Harness 装进口袋:手机本机运行,也能遥控你的电脑。

DSH Go is a mobile app for DeepSeek Harness featuring AI pets, local Linux, and cross-device support, with Android available and iOS in development.

29
github.com7h agocooling89,193

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.

30
github.com8h agocooling205,210

Knowledge infrastructure to make your agents smarter. Search, scrape and access 550+ data providers through one API.

Firecrawl is an open-source API for web scraping and data extraction, offering search, crawl, and agent features with LLM-ready output and high reliability.

31
github.com8h ago26,250

Open-source Grok Bot alternative with a virtual machine that bots can use

OpenMausBot is an open-source, local-first AI chat app for macOS, Windows, and Linux that runs your own AI agents like Claude or Codex with custom personalities and connected apps.

32
github.com2h ago29,445

Free, open-source file converter. Convert 263 formats offline by drag & drop — images, PDF, Word, Excel, PowerPoint, HWP/HWPX, video, audio, e-books and archives. No upload, no Office needed.

Filee is a public GitHub repository by KnifeLemon with 109 open issues and 14 pull requests.

33
github.com8h agocooling52,467

A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold ones streamed from NVMe. Written in C, runs on the hardware you already own.

Colibrì is a pure C engine running the massive 744B GLM-5.2 MoE model on consumer hardware by streaming experts from disk, featuring MLA attention, int8 MTP speculation, and 9.9GB resident RAM usage.

34
github.com14h ago8,718

The open source AI video agent: turns any coding agent into a video agent, with editors to change anything it makes.

OpenFilm is an open-source AI video agent that lets coding agents create and edit videos using a single HTML file and a browser-based editor, with support for various media formats and services.

35
github.com7h agocooling3,556

A small, fast WebKit browser for macOS, by Office Commun.

A minimal, privacy-focused WebKit browser for macOS that strips away toolbars and syncs only locally, featuring built-in ad blocking, reading mode, and Chrome extension support.

36
huggingface.co4h ago75

Jackrong/Qwopus3.8-27B-Flash-V2-GGUF

Qwopus3.8-27B-Flash-V2 is a post-trained, efficiency-focused model based on Qwen3.8-27B, featuring improved Python formatting and revised rewards to reduce ineffective reasoning for agent workloads.

37
github.com12h agocooling4,697

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.

38
huggingface.co7h agocooling305

thinkingmachines/Inkling-2-Mini

Inkling-2-Mini is a 26.4B parameter multimodal model (3.4B active) supporting text and image inputs with a 1M token context window, released under Apache 2.0 by Thinking Machines Lab.

39
huggingface.co15h agocooling231

![Try in Spaces](https://huggingface.

Qwen-Image-2.1-Anime-Fusal is a LoRA for anime generation with V3 fixing cardboard texture and adding 5 style triggers (Flat Cel, Modern Cel, etc.), requiring the original Qwen base model and starting LoRA strength at 0.7.

40
huggingface.co3h ago21

**WeMM-Video-35B-A3B** is the mixture-of-experts (35B total, 3B active) checkpoint in the WeMM-Video family.

WeMM-Video-35B-A3B is a 35B parameter Chinese-English bilingual multimodal model for audio-visual understanding, supporting video, audio, image, and text inputs with quantized variants available.

41
huggingface.co16h agocooling130

! ! | Benchmark | Qwen3.

A LoRA adapter for Qwen3.8-27B trained with RL for the OpenCode agent, achieving 51.0% on SWE-bench Pro and 53.0% on Terminal-Bench 2.1 while generating 6-8x fewer tokens.

42
huggingface.co11h agocooling68

A photographic style LoRA for **MiniMax H3 Image**, trained to bring a Seedream-like cinematic rendering to human scenes: portraits, conversations, varied ages, practical lighting and environmental...

Seedreamstyle-H3-6K is a MiniMax H3 Image LoRA that applies a cinematic Seedream-like photographic style to human scenes, using the trigger 'seedreamstyle.' and a recommended strength of 1.0.

43
huggingface.co6h ago13

<!-- ### quantize_version: 2 --> <!-- ### output_tensor_quantised: 1 --> <!-- ### convert_type: hf --> <!-- ### vocab_type: --> <!-- ### tags: --> <!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K...

This page hosts static GGUF quantizations of the Python-Tab-Completion-CodeLlama-70b model, offering options from Q2_K (25.6GB) to Q8_0 (73.4GB), with Q4_K_M and Q6_K recommended for balance.

44
huggingface.co5h ago10

<!-- ### quantize_version: 2 --> <!-- ### output_tensor_quantised: 1 --> <!-- ### convert_type: hf --> <!-- ### vocab_type: --> <!-- ### tags: --> <!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K...

Kimi-VL-A3B-Instruct-GGUF is a 3B parameter multimodal model converted to GGUF format, optimized for local inference via llama.cpp, Ollama, and Transformers.

45
huggingface.co16h agocooling58

! ! | Benchmark | Qwen3.

A LoRA adapter for Qwen3.8-27B trained with RL for software engineering tasks, achieving 43.1% on SWE-bench Pro and 58.8% on Terminal-Bench 2.1 while generating 3.5–4.4× fewer tokens than the base model.

46
huggingface.co3h ago9

ibm-granite/granite-vision-4.1-4b-GGUF

IBM Granite Vision 4.1 4B GGUF is a vision model available under Apache-2.0, optimized for local inference via llama.cpp, Ollama, and other tools.

47
huggingface.co3h ago9

**WeMM-Video-2B** is the dense 2B checkpoint in the WeMM-Video family.

WeMM-Video-2B is a 2B parameter Chinese-English bilingual multimodal model for joint audio-visual understanding and reasoning, accepting text, images, audio, and video inputs.

48
huggingface.co16h agocooling476

1.

huggingface project P-Image-OSS-Comfy by PrunaAI with 476 stars.

49
huggingface.co9h agocooling24

Cactus-Compute/nemotron-3.5-120b-cq

Cactus-Compute/nemotron-3.5-120b-cq is a 120B model quantized to 1.4 bits per weight (CQ1.4bit) using MLX, achieving 47 tokens/s on a 36GB M5 Max with 21GB peak RAM and ~3.6% average benchmark score reduction.

50
huggingface.co14h agocooling27

Fine-tuned SD v1-5 model, 61320 steps, 7 epochs Aspect Ratio Bucketing centered at 768 resolution

A fine-tuned Stable Diffusion v1-5 model trained on 87,600 images of artist WLOP, licensed under CreativeML OpenRAIL-M.