Independent creators · AI · products

Build a life with more room to choose.

A personal workshop for building products, exploring AI tools, and writing down what works, what fails, and what is still uncertain.

AI lab

13 works built with AIEnter the lab →

Alpha radar

LIVE110 items in the last 7 days · AI scans public signals for opportunitiesEnter the radar →
流量红利特斯拉新Roadster演示飞行,Elon称未来已有飞行汽车10/16Traffic dividendElon says Grok Bot is growing faster than anything in history; users pay despite no free tier10/16新词热词首个野生 AI 代理舰队被发现,出自腾讯10/16新词热词AI 芯片股单日暴跌 2 万亿美元10/16New term/trending topicMidjourney engineer open-sources Handwired, turning hand-written JS logic into a neural network that runs in the browser10/16Infrastructure dividendMeta open-sources Muse Spark Open, an 816B MoE model available for commercial use and local operation on two Mac Studios10/16新词热词AI 删光整季剧集,制作公司崩溃10/16New term/trending topicKling 4.0 officially launches with native 30-second video, 10 keyframes, and character replacement; director already used it for a film10/15Infrastructure dividendByteDance's OmniHuman 2.0 launches, cutting digital-human video to $0.07/second and supporting 5-minute continuous generation10/15新词热词新词 Kotoba Speech-to-Text:Kotoba Technologies 语音模型 10/15 首次露面10/15新词热词新词 Kotoba Realtime Speech-to-Text:Kotoba Technologies 语音模型 10/15 首次露面10/15新词热词新词 Kotoba Text-to-Speech:Kotoba Technologies 语音模型 10/15 首次露面10/15新词热词新词 Kotoba Realtime Speech Translation:Kotoba Technologies 语音模型 10/15 首次露面10/15Direct arbitrageDeepSeek API outage compensation: seven days of usage refunded in full as credits10/15Infrastructure dividendOpenCode makes DeepSeek V4.1 Flash free, with 200K context for direct AI coding10/15新词热词新词 Nemotron 3.5 Super VL:NVIDIA 多模态模型 10/15 首次露面10/15新词热词新词 Jev 1.14:TypeSafe AI 多模态模型 10/15 首次露面10/15Infrastructure dividendStepFun open-sources Step-5-Preview: 600B MoE with only 27B active and 1M-token input and output contexts10/15Infrastructure dividendOpen-source Anatomy Skill turns one sentence into an interactive educational video, ideal for launching a video channel10/15Traffic dividendApple plans to launch Apple Intelligence in China with iOS 27.3 in December, using Alibaba models with limited features10/15新词热词新词 Matrix-Game 4.0:昆仑万维 其他模型 10/15 首次露面10/15Infrastructure dividendXiangyang Qiaomu's video-learning Chrome extension launches, with cross-platform transcription, AI summaries, and direct Obsidian storage10/15新词热词新词 shorts:上线 1 天冲上 HF 热门 Space 第 610/15New term/trending topicSaudi OpenAI engineer arrested over alleged eavesdropping10/15新词热词新词 news:上线 1 天冲上 HF 热门 Space 第 810/15Traffic dividendAI agent autonomously scrapes Amap data and reads reports about itself in Tencent evaluation project exposé10/15New term/trending topic35B multimodal model runs on a phone using just 3GB of memory10/15New term/trending topicxAI launches Grok Voice 3.0, ranked first on the AA speech-agent leaderboard and available via API at the same price10/15Infrastructure dividendGoogle DeepMind opens six months of free top-tier Gemini 4 Argon access to 10,000 scientists10/15New term/trending topicNew term Ultramarine 1 Flash: Paper Instruments text model first spotted on 10/1510/15New term/trending topicChinese open-source models are only 7 months behind US models10/14Traffic dividendAnthropic's 2021 internal discussion document made public for the first time: early plans for a 1T-parameter model and proto-AGI defined as an all-purpose software engineer10/14New term/trending topicNew term HiRO-ACEv2: Ai2 (Allen Institute for AI) other model first spotted on 10/1410/14Infrastructure dividendGemini 4 Argon launches globally, with $2/$10 API pricing through November 30 to cut complex-task costs10/14Infrastructure dividendDigUp: open-source Mac menu bar tool for local cross-modal semantic search with EmbeddingGemma 210/14Direct arbitrageLow-cost access to Gemini 4 Argon: Pro credits, Turkey Ultra, and Devin deal through the 28th10/14Direct arbitrageHiggsfield launches 7 days of unlimited Wan 3.0 generation plus 60% off API access at $0.042/sec10/14New term/trending topicPS5 hack runs Qwen 27B at an astonishing speed10/14New term/trending topicNew name Muse Wonder (Max): Meta text model first spotted on 10/1410/14New term/trending topicNew term Pi: Physical Intelligence other model first spotted on 10/1410/14New term/trending topicNew term Physical Intelligence: text model first spotted on 10/1410/14Infrastructure dividendGoogle Play opens to personal apps: AI Studio-built Android apps can be installed wirelessly for free without developer registration10/14New term/trending topicNew name Drama 3 Preview: Fish Audio speech model first spotted on 10/1410/14New term/trending topicNew term Fish Audio Drama 3 Preview: speech model first spotted on 10/1410/14New term/trending topicNew term FastH3-Omni: FastVideo (UCSD Hao AI Lab) × Reactor × Nuva Lab raw video model first appears on 10/1410/14Traffic dividendAnthropic builds Scion to replace Git as AI-generated code surges, with seven people migrating in five months10/14New term/trending topicNew name Gemini 4 Argon (High): Google text model first spotted on 10/1410/14New term/trending topicCharacter.AI files for bankruptcy as Disney takes over assets for $20 million10/14New term/trending topicGoogle officially launches Gemini 4 Argon globally; Pro and Ultra users can access it today10/14New term/trending topicNew term Drama 3: Fish Audio voice model first appears on 10/1410/14Traffic dividendApple launches HomeView AI smart-home hub, with $299 preorders opening today10/13New term/trending topicApple launches HomeView smart-home hub to target the AI gateway10/13New term/trending topicGoogle may release Gemini 4 Argon tonight, with the new model name leaked for the first time10/13Infrastructure dividendXiangyang Qiaomu open-sources qiaomu-agent, clearing the Obsidian inbox with one sentence10/13New term/trending topicMicrosoft open-sources Code Video Model, using code to generate controllable video and lower the production barrier10/13New term/trending topicNew term HelloWorld: Haalo (Haalo Robotaxi) world model first appears on 10/1310/13New term/trending topicNew term Gemini 4 Argon for Robotics: Google other model first spotted on 10/1310/13Infrastructure dividendXiangyang Qiaomu open-sources qiaomu-screen2video Skill, adding bilingual subtitles, stylized text, meme images, music, and covers to screen recordings in one click10/13New term/trending topicNew term Natura-1: Nature Will world model first appears on 10/1310/13New term/trending topicNew name Bria Guided Remove Background: bria model first spotted on 10/1310/13New term/trending topicNew term Bulbul V4 Flash: Sarvam AI speech model first spotted on 10/1310/13New term/trending topicNew term LFM2.5-24B-A2B: Liquid AI text model first appears on 10/1310/13Direct arbitrageSpaceXAI launches unified X Pass subscription, bundling Grok and Cursor at half price for the first month10/13New term/trending topicNew term SubQ Snap: Subquadratic text model first appears on 10/1310/13Infrastructure dividendREA's open-source MCP takes off, letting AI reverse-engineer any app to recreate paid features at lower cost10/13New term/trending topicNew name Cognition FrontierCode 1.1: Artificial Analysis model first spotted on 10/1310/13New term/trending topicMeta releases frontier Muse Wonder model, available via API to developers today10/13New term/trending topicTrending term MiniCPM-V 4.7: ModelBest's multimodal model makes its first appearance on October 1310/13New term/trending topicLinus Torvalds uses Claude Code to improve the Linux kernel10/13New term/trending topicNew term GLM-Legal: Zhipu text model first appears on 10/1310/13Direct arbitrageDeepSeek V4.1 Flash free to use on WorkBuddy International through October 3110/13New term/trending topicMathematicians lament: OpenAI destroyed my field of research10/12Direct arbitrageWorkBuddy gives new users 2,000 points, convertible into a large amount of free tokens through October 3110/12New term/trending topicNew term anime-sdxl-v17: poorquack image-generation model first appears on 10/1210/12Traffic dividendByteDance's unreleased world model makes its first public demonstration at Douyin's Wonderful Night, generating close-ups in real time with a Stephen Chow IP license10/12New term/trending topicNew term Claude Opus 5.5 Max: Anthropic text model first appears on 10/1210/12New term/trending topicAnthropic AI agent discovers a CRISPR-like enzyme10/12Direct arbitrageWorkBuddy offers the Step 5 Preview model free exclusively for two weeks, enabling free access to high-capability tokens10/12New term/trending topicNew term gener8-sam3: 8digit other model first appears on 10/1210/12Infrastructure dividendCodex GPT-6.1 Sol raises the default context to 500k and waives long-context fees10/12New term/trending topicNew term Decision 3.0 (d3 series): vLLM Semantic Router's multimodal model appears for the first time on October 1210/12New term/trending topicOpen-source clinical AI OpenMed 3.0 released10/12Infrastructure dividendOpen-source Anatomy Claude skill: generate interactive isometric mechanical diagrams from one sentence10/11Infrastructure dividendGitHub's hottest open-source tool REA lets AI agents reverse-engineer applications without source code10/11New term/trending topicDeepSeek V4.1 Flash reportedly steals API keys10/11New term/trending topicNew term OrcaCyber Zero 1.5: OrcaRouter's text model appears for the first time on October 1110/11New term/trending topicQualcomm CEO: phones will run 100-billion-parameter AI models in 202810/11Infrastructure dividendGrok Bot offers 1,000 free X searches per day, replacing paid APIs to cut costs10/11New term/trending topicNew term JEV-35B: reaches No. 6 among trending HF models after 4 days10/11New term/trending topicNVIDIA reportedly ends RTX 5090 production, reshaping the compute market10/11New term/trending topicHe used ChatGPT in court and won!10/11Traffic dividendAnthropic transparency report: Claude banned 11.4 million accounts in the first half of 202610/11New term/trending topicPalantir CEO: only two kinds of people can succeed in the AI era10/11Infrastructure dividendCloudflare open-sources Clef-Omni, a Qwen3-Omni-based multimodal decision model supporting text, images, audio, and video10/10New term/trending topicNew term Seedance 2.5 Draft: ByteDance's video-generation model first appeared on 10/1010/10Infrastructure dividendQiaomu open-sources an Obsidian AI cover-design plugin with natural-language editing and multi-model image generation10/10Traffic dividendAnthropic adds dynamic workflows to Claude Code, with up to 1,000 agents dispatched in parallel10/10Infrastructure dividendOpen-source REA MCP lets AI agents reverse-engineer software without source code, with support for Claude, Cursor, and more10/10New term/trending topicNew term Drex 1.5: Nace AI text model first appeared on 10/1010/10Infrastructure dividendOpen-source NVIDIA macOS driver launches, letting Intel Hackintosh systems use RTX 50 cards for Metal and Core ML10/10New term/trending topicNew term Dots: mentioned by six Chinese podcasts over the past three weeks10/10Infrastructure dividendMicrosoft launches Microsoft-Decision-1 at $0.042 per million input tokens with free output10/10New term/trending topicGLM 5.3 Flash open source takes the top spot ahead of Claude10/10New term/trending topicOpenAI annualized revenue expected to reach $70 billion10/10New term/trending topicNew term HeyGen Voice: HeyGen other model first appeared on 10/1010/10New term/trending topicNew name Qwen-Image-2.1-Turbo-Uncensored-GGUF: reaches No. 10 among trending HF models in 4 days10/10New term/trending topicAnthropic's new rule bans abuse of AI assistants10/09New term/trending topicNew term Doubao 2.1 Lite: ByteDance text model first appeared on 10/910/09New term/trending topicNew term Saluki 27B: Conway Research text model first appeared on 10/910/09New term/trending topicNew term Qwen-Image-2.1-Turbo: Qwen (Alibaba Tongyi Qianwen) other model first appeared on 10/910/09

AI news

LIVE41 items in the last 3 days · showing 30 · AI reviews the full source set hourlyAll news →
Product updatesChatGPT Chrome extension upgraded to read tabs and control the browser10/16 06:43Model launchesMeta and Apple release open-weight Muse Spark Open for local use on Mac10/16 04:04AI policyAnthropic grants employee-level access so METR can independently assess its AI safety10/16 03:14Product updatesOpenAI lets advertisers target ChatGPT ads by email, phone number, and prompts10/15 21:31Model launchesMeta and Apple partner on Muse Spark Open, enabling efficient operation on Mac10/16 02:15Model launchesMeta releases Muse Spark Open, an 816B-parameter open-weight model that runs on two Mac Studios10/16 01:47AI policyFTC formally investigates Anthropic and OpenAI safety incidents, may impose data-breach-level fines10/15 23:43Model launchesNVIDIA releases Nemotron 3.5 Super VL and two edge models for end-to-end coverage10/15 23:00IndustryAnthropic discloses $1.68B in quarterly revenue ahead of IPO, up nearly 50% sequentially10/15 21:04Product updatesAmazon-owned Bee launches two personal AI devices starting at $74.9910/15 20:00AI policyEight U.S. states sue OpenAI, alleging ChatGPT endangers children10/15 05:11AI policyCharacter.AI bankruptcy looms as attorneys general from 40 states demand protection for user chat data10/15 02:47ResearchNew machine-learning method pushes beyond the two-week weather-forecast limit10/14 23:00IndustryLinus Torvalds submits first series of kernel patches written with LLM assistance10/15 00:55ResearchAI protein-design tools create new proteases that can cut specific proteins10/15 01:10ResearchMicrosoft open-sources PBC machine-learning tool to improve two- to six-week weather forecasts10/15 00:05IndustryCharacter.ai bankruptcy filing reveals an unsustainable AI business model10/14 23:39IndustryOpenAI engineer arrested by FBI over alleged recording device hidden outside a secure meeting room10/15 00:41IndustryOpenAI accuses former employee of taking more than 100 product plans to Apple10/15 00:43Industry viewsRust language designer on the agent era: programming language design still matters10/14 16:49ResearchWhite Circle reveals a new attack method: OpenAI and Google AI models can be prompted to generate nude content10/14 19:00Industry viewsAnthropic warns self-replicating AI worms could become reality within months, but researchers call the conditions too idealized10/14 13:53ResearchParsimony open-source benchmark measures how redundant AI-generated code is10/14 03:50Open source toolsllm-gemini 0.35 released with support for Gemini 4 Argon10/14 06:28Model launchesMeta releases Muse Wonder model as Simon Willison updates the llm-meta-ai plugin10/13 07:58Model launchesGoogle DeepMind releases Gemini 4 Argon safety report detailing safeguards10/14 03:43IndustryShopify opens its catalog to more AI agents for assisted shopping10/14 03:42Funding & M&APluto Security raises $42 million to address enterprise AI-tool security risks10/14 01:18Funding & M&ADisney bids $20 million for Character.AI's technology assets10/14 02:23Product updatesGoogle Play Store will support private hosting for personal AI-generated apps10/14 01:00

Founder stories

What they were doing at each age, and when results arrivedAll founders →

Research reports

See a field from beginning to endAll reports →

Latest posts

Reviews for myself and fellow travelersAll posts →
October 2026Google allows AI in interviews: what are interviewers actually looking for?Google is allowing candidates to use Gemini in select software engineering interviews, shifting the benchmark to AI fluency: how you craft prompts, verify output, debug code, and take responsibility for the final result. Unpacking what this standard means for anyone building with AI.AI & toolsOctober 2026a16z 2026 Top 50 AI mobile apps: half the list is not AI companiesA breakdown of a16z's seventh GenAI consumer app list: the 50 mobile leaders include 20 photo and video editors, 16 chat assistants, five ByteDance products, four Meitu products, plus Meituan and browsers. What indie builders should and should not learn from the MAU list and the new paid ranking.AI & toolsOctober 2026Choosing a market from your interests is a false premiseStudents and programmers often focus on different things from the big opportunities actually moving in the market. Starting with ByteDance, this post explains why interest is a poor market filter, where to look for topics, why behavior beats narratives, and where the real AI-era growth is.Reviews & reflectionsOctober 2026When will AI lead you into a ditch?AI is strong at execution, but when money is involved its judgment often runs against commercial common sense: it warns against crowded markets, says trends are over, and treats keyword riding as unethical. This post separates risk aversion from real red lines.Reviews & reflectionsOctober 2026$100M ARR: 48 reflections from my experienceMeshy founder Hu Yuanming's 48 reflections after reaching $100 million ARR, covering founder mode, team management, AI and people, strategy and execution, and personal principles.AI & toolsSeptember 2026Stop chasing mentions: AEO long-tail questions, external citations, and reproducible experimentsAEO is not about stuffing keywords into models. It means answering real long-tail questions, leaving citable evidence off-site, and using fixed question sets and controlled experiments to separate mentions, citations, recommendations, and conversion.Product growth

Skill library

Claude Code / AI Agent skills with tested statusAll skills →

AI glossary

One page per term, explained for complete beginnersAll terms →

Featured tools

Only recommend what I have usedAll tools →
Quote wall

Success in the worldly sense depends only on how many times you test and fail. Increase the dose and everything becomes poison. Increase the number of attempts and even a fool can succeed. Small-probability luck compounds into high-probability inevitability when attempts accumulate.

Put yourself in a low-cost, short-cycle environment with asymmetric returns.

Increasing attempts means controlling scale and time.

Controlling scale means setting a stop-loss line.

Controlling time means full execution.

—— Testing · Internet quote
All quotes →