Early Signal: AI launches before everyone else

Today's edition · last 24 hours543 new signals

The short version. Today's theme is turning heavyweight tech into plug‑and‑play tools, starting with REA, which lets an AI agent reverse‑engineer any binary without needing the source code. From a rust‑based Photoshop clone to a Cloudflare skill that runs multi‑phase security audits, the lineup shows how developers can offload heavy lifting to ready‑made AI services.

▲ 15.2k today368k starsTypeScriptReader tip

REA lets AI agents reverse‑engineer any software, no source needed

Repo · morluto/rea

REA is a TypeScript toolkit that lets you attach an LLM agent to tools like Ghidra, Hopper or static JavaScript analyzers, so the agent can inspect binaries, Electron apps, .NET assemblies or websites and explain how they work. It runs locally, shows the evidence behind each conclusion, and can even generate a re‑implementation of the feature you examined.

Why it matters. Security researchers, developers and hobbyists can use REA to quickly understand undocumented functionality and prototype compatible replacements.

Try it: npx rea-agents setup

spotted 1 h ago

Biggest movers

2
▲ 4.2k today25.2k starsJava

Debloat Android TV over Wi‑Fi and get Apple‑TV UI
Repo · Gathin23/cleanyourtv
The tool connects to your Android or Google TV via ADB over Wi‑Fi, backs up the current apps, disables preinstalled telemetry and bloatware, and installs a lightweight launcher that mimics Apple TV’s home screen. It requires no root, no cables, and provides a one‑line restore script to undo changes.
3
▲ 3.1k today71.5k starsRust

Rust‑powered Photoshop clone rakes in 3k stars overnight
Repo · storytold/photocraft
PhotoCraft is an open‑source image editor that replicates Photoshop’s layers, masks, adjustment layers and PSD compatibility. It runs natively with GPU acceleration via wgpu, so edits stay offline and fast.
4
18.2k likes

Qwen releases 27B vision-language model, boosting image-to-text AI
Model · Qwen/Qwen3.8-27B
Qwen3.8-27B is a 27‑billion‑parameter model that can look at images or video frames and generate descriptive text. It also supports complex multi‑step reasoning and autonomous planning, making it useful for coding help, research, and long‑horizon tasks.
5
▲ 3k today40.1k starsJavaScript

Cloudflare's security-audit skill automates multi‑phase code audits
Repo · cloudflare/security-audit-skill
The repo provides a coding‑agent skill that runs a six‑step, machine‑readable security audit, from reconnaissance to independent verification and neutral reporting. It orchestrates isolated agents to hunt for bugs, validate findings, and output JSON and markdown reports.
6

Kazakhstan ramps up GPU compute for AI rollout
Lab · Kazakhstan Expands Compute for Its AI Future
Kazakhstan is moving from planning to actually deploying AI hardware, adding NVIDIA‑powered GPU clusters across the country. The new infrastructure supports local models, startups and government applications.
7
77 upvotes

New Benchmark Tests Deep Search Agents on Real-World Web
Paper · Deep Search in Real-World Web Environments
The paper introduces a benchmark that evaluates how AI agents perform end‑to‑end research across the live internet, handling diverse sources, media types, and constantly changing content. It measures an agent’s ability to locate, retrieve, and synthesize information beyond static retrieval datasets.
8
▲ 2.9k today20.2k starsPython

DeepInfer delivers high-throughput inference for DeepSeek models
Repo · deepseek-ai/DeepInfer
DeepInfer is an open‑source Python framework that runs DeepSeek‑V4.1‑Flash models at production speed. It bundles optimized kernels, parallel attention and a zero‑overhead batch scheduler so you can serve large language models offline without hand‑crafting low‑level code.
9
930 points420 comments

SimCity, but with Permits
Hacker News · SimCity, but with Permits
10
8.9k likes

Lightricks/LTX-2.5
Model · Lightricks/LTX-2.5
Model · image-to-video

The repo market

Open-source projects ranked by stars gained today, top 9 of 381

Project	Today	Total
Ebony-Vinyl/dsh-our-free-modelFree Frontier LLMs Plug‑in for dsh Gains Rapid AdoptionThe dsh-our-free-model plugin adds cutting‑edge large language models to the dsh interface. It requires no login, API key or extra configuration, giving unlimited access to models like DeepSeek V4.1 Flash and Kimi K3.	▲ 2.3k	37.4k
DietrichGebert/ponytailPonytail AI plugin slashes code size by half for devsPonytail lets an AI agent rewrite requested code into the fewest necessary lines, reusing existing components instead of generating fresh code. In tests it cut code volume by 53%, reduced development time by 41% and lowered token usage.	▲ 1.9k	174k
debpalash/VoiceStudioOpen-source ElevenLabs clone lets you dub, clone, and transcribe locallyVoiceStudio runs on your computer and can copy any voice, generate speech in 646 languages, dub videos, transcribe audio and create audiobooks. It ships as a desktop app with a one‑line installer and works without sending data to the cloud.	▲ 1.9k	82.1k
stablyai/orcaOrca is the place for working with a fleet of agents (e.g. Grok Bot, Claude Code). Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.	▲ 1.7k	103k
Graphify-Labs/graphifyTurn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.	▲ 1.4k	133k
wheresryan22/anatomyAn agent skill for creating detailed, interactive isometric illustrations of real-world machines and complex systems. Generates polished SVGs with optional WebGL shaders and 3D effects.	▲ 1.3k	19.3k
pengchujin/aerial-tv给国产电视的一块干净桌面：参照 tvOS 设计的开源 Android TV 桌面，没有广告，开机自动接管。A tvOS-style launcher for Xiaomi, TCL and other China-market TVs.	▲ 1.1k	3.4k
google-research/harness2	▲ 1.1k	18.2k
Niko1221/StrataQwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.	▲ 1.1k	30.7k

New models

Trending on Hugging Face, top 9 of 61

Qwen3.8-Flash-Next offers faster long-context LLM inference

Model · Qwen/Qwen3.8-Flash-Next

Qwen3.8-Flash-Next is an open‑weight language model that can handle up to a million tokens while keeping latency low, thanks to its new hybrid attention called Qwen Sparse Attention. It runs with the standard Hugging Face Transformers stack and is compatible with vLLM, SGLang and TokenSpeed.

6.4k likes

Unsloth releases 4‑bit Qwen3.8‑27B GGUF for desktop AI

Model · unsloth/Qwen3.8-27B-GGUF

The Qwen3.8‑27B GGUF is a 27‑billion‑parameter model quantized to 4‑bit, runnable on a laptop through Unsloth Desktop. It handles text, images and video, supports tool calling and can be fine‑tuned for coding or long‑horizon agentic tasks.

5.3k likes

facebook/sam3

Model · facebook/sam3

Model · mask-generation

4.9k likes

baidu/Unlimited-OCR

Model · baidu/Unlimited-OCR

Model · image-text-to-text

4.9k likes

abenzerps/Qwen-Image-2.1-Uncensored-Quantized

Model · abenzerps/Qwen-Image-2.1-Uncensored-Quantized

Model · text-to-image

4.8k likes

deepseek-ai/DeepSeek-V4.1-Flash

Model · deepseek-ai/DeepSeek-V4.1-Flash

Model · image-text-to-text

4.7k likes

pyannote/speaker-diarization-community-1

Model · pyannote/speaker-diarization-community-1

Model · automatic-speech-recognition

3.3k likes

google/embeddinggemma-2

Model · google/embeddinggemma-2

Model · feature-extraction

2.9k likes

selfit-camera/Omni-Image-Editor

Space · selfit-camera/Omni-Image-Editor

App / demo · gradio

2.9k likes

From the labs

Straight from the companies building them, top 9 of 50

OpenAI
OpenAI teams with Poland’s NASK to boost AI safety

OpenAI and the Polish research network NASK announced a partnership to work on AI safety, security, and technical research. The collaboration will focus on building safeguards and strengthening the resilience of the digital ecosystem.
Hugging Face
Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
OpenAI
Build more responsive apps with the Decisions API

Fast, focused, low-cost decisions for any application.
AWS ML
Amazon Polly delivers a major upgrade to its Generative engine

The Amazon Polly Generative engine gets a major upgrade: more expressive, more accurate speech with low latency, plus 14 new voices across 9 locales and general availability of Hindi and Finnish. This drop-in upgrade requires no code changes and is now general
AWS ML
How IskayPet unified corporate knowledge with Amazon Bedrock Managed Knowledge Bases

Learn how IskayPet, a leading pet care company in Iberia, built a multilingual corporate knowledge assistant on Amazon Bedrock Managed Knowledge Bases. The assistant answers employee questions across retail, veterinary, onboarding, and procurement in four lang
AWS ML
Announcing new ways to customize models with SageMaker AI, without managing GPUs

Amazon SageMaker AI serverless model customization now includes training sessions, generally available today. You write and coordinate your own post-training loop (custom loss functions, reward signals, or multi-turn reinforcement learning) while SageMaker AI
AWS ML
Write your own training loop, without managing GPUs, with Amazon SageMaker training session APIs

Amazon SageMaker AI training session APIs let you write your own LLM training loop (SFT, RL, or custom objectives) in a Python script while the service runs the GPUs. This post introduces the programming model and three SDK clients, then walks through a GRPO l
AWS ML
Deploy AI in a HIPAA-eligible environment with Amazon Quick

Amazon Quick is a HIPAA-eligible service, but eligibility is not the same as compliance. This post explains who HIPAA applies to, how PHI flows through Amazon Quick, what the platform provides for regulated environments, and a step-by-step guide to deploying c
AWS ML
Stream low-latency speech from AI agents using Kokoro TTS on Amazon SageMaker AI

Voice agents often wait for an agent's full response before starting speech synthesis, adding latency users feel directly. This post shows how to stream each sentence to a Kokoro-82M text-to-speech model on Amazon SageMaker AI as the agent produces it, and how

Research

The papers researchers upvoted most today, top 9 of 35

Paper
New Method Lets MLLMs Generate Cinematic Videos from Sparse Prompts

It uses factor‑aware predictive learning to turn brief textual or sketch inputs into full‑length, movie‑style video clips. The system separately controls characters, scenes, actions and camera while automatically filling in the missing frames.
69 upvotes
Paper
Chunk-Level Parametric Memory Lets LLMs Learn Without Forgetting

Deep Fusion Memory adds a trainable chunk‑level store that LLMs can write to and read from during inference, sidestepping costly external retrieval. The memory updates with new data while preserving what the model already knows.
28 upvotes
Paper
Policy-Mask trims diffusion features for classification

Policy-Mask automatically selects timesteps, layers and channel groups from a frozen diffusion model to use for discriminative tasks. It learns a lightweight, budget‑aware policy so you don’t have to hand‑pick features across the denoising trajectory.
25 upvotes
Paper
Meta Expert: Cross-Layer Expert Sharing for Storage- and Retrieval-Efficient MoE

As model parameters scale up, sparsely activated Mixture-of-Experts (MoE) approaches have become essential for model scaling. However, mainstream models currently adopt hierarchical architectures, where experts in different layers, although topologically independent, exhibit substantial similarity a
19 upvotes
Paper
Code2Video-Compass: A Comprehensive Benchmark for General-Purpose Code-Based Video Generation across T2V, I2V, and V2V

Code-based video generation offers fine-grained control, targeted editing, and component reuse compared with direct video generation. This approach requires multimodal large language models (MLLMs) to translate visual and temporal requirements into executable programs that produce complete videos. H
16 upvotes
Paper
Heartbeat Thinking: Speculative Reasoning and Tool Execution While Listening in Realtime Voice Agents

Tool-using voice agents often leave users waiting after speech ends, even when actionable information arrives earlier in the request. Speculative approaches begin work during speech, but chunk-bounded or interrupt-on-update generation can truncate a tool call before its arguments are complete. This
12 upvotes
Paper
Verus-Insight: Towards Explainable Artifact Detection in AI-Generated Videos

Recent advances in video generation have made synthetic videos increasingly realistic, placing greater demands on reliable AI-generated video detection. Existing artifact-based detectors typically identify generation artifacts as evidence for source prediction, yet they may either fail to recognize
10 upvotes
Paper
Editing Images into Reusable Layers via Compositional Edit-Prior Distillation

High-quality image editing requires precise modification of target regions while preserving the content of unrelated regions. Existing editing models produce flattened outputs that limit precise local manipulation and layer reuse, while decomposition methods typically optimize layer recovery separat
10 upvotes
Paper
Rethinking Sequence Compression with Test-time Regression

Sequence compression is a fundamental operation and a critical information bottleneck in efficient language and multimodal models, and yet prevailing approaches based on pooling or learned projections remain underperforming. We revisit sequence compression through the lens of test-time regression: i
9 upvotes

The conversation

What builders are arguing about on Hacker News, top 9 of 16

Hacker News
Claude Duet

550 points514 commentsRead the thread
Hacker News
WorldTUI: I trained a 360M model to pretend to be a terminal

517 points159 commentsRead the thread
Hacker News
PHP2Shell: AI uncovers critical RCE in PHP

449 points184 commentsRead the thread
Hacker News
Publish fake stats, get cited in LLMs

366 points204 commentsRead the thread
Hacker News
Every AI model has the same "controversial" take on humans

331 points348 commentsRead the thread
Hacker News
Mistral Large 4

323 points102 commentsRead the thread
Hacker News
OpenAI models are coming to Google Cloud

323 points145 commentsRead the thread
Hacker News
Patch the Planet: global response coordination as agents uncover vulnerabilities

265 points183 commentsRead the thread
Hacker News
LLMs are making formal software verification feasible at a much larger scale

254 points119 commentsRead the thread

Get the signal first.