UNGETSU (@ungetsuli) header
UNGETSU (@ungetsuli) profile picture

🎓 00 INTJ @NTUsg 🧑🏻‍💻 Building AI Products 🎮 Prev @GenshinImpact & @TencentGames 📍 Japanese→MBB→Gaming→ AI Startup

Shanghai
Joined May 2026
Photo by UNGETSU (@ungetsuli)Photo by UNGETSU (@ungetsuli)Photo by UNGETSU (@ungetsuli)Photo by UNGETSU (@ungetsuli)Photo by UNGETSU (@ungetsuli)Photo by UNGETSU (@ungetsuli)
UNGETSU retweeted
这可能是市面上写给独立开发者最详细的产品宣传片制作教程。 苦写几个月代码,上线却卡在宣传片?很多朋友问我 Auday 的宣传片怎么做的,其实就是靠 Claude 极强的视频能力,去操作 Seedance 2.5 生成素材,再配合真实界面和配音合成,如果你也有类似需求不妨来看看。
95
76
20
713
834,007
UNGETSU retweeted
My iPhone Duo video went viral, so I'm giving away 100 iPhones on X 😛
91,515
92,118
6,531
139,472
6,203,348
UNGETSU retweeted
Auday 1.0.1 is now live! This update adds Japanese and Korean support for AI-generated content, including summaries, Highlights, daily Reflections and Chat replies. You can switch languages in Settings → AI Primary Language. It also fixes the silent-recording issue affecting some Apple Watch Ultra users, improves transfer reliability between Apple Watch and iPhone, and brings a major overhaul of local storage for smoother loading and browsing as your recording history grows. On launch day, Auday reached #1 in the App Store’s paid Lifestyle charts in multiple countries. The response has far exceeded our expectations, and every download, message and suggestion has been a huge encouragement. Behind Auday is a solo developer, working late to collect your feedback, investigate issues and keep improving the app. Thank you for your support and patience. We also want to explain the choices behind transcription. To support a wider range of iPhones while keeping resource use low and minimizing slowdowns, our current on-device speech recognition model prioritizes speed and efficiency. In tests on iPhone 17 Pro, several hours of audio often took just a few seconds to transcribe. Actual performance varies with the device and recording. That speed comes with some accuracy trade-offs. Auday AI uses surrounding context when interpreting transcripts and generating content to help identify and work through recognition errors, reducing their impact on summaries, analysis and conversations. For the next update, we plan to introduce an optional speech recognition model with higher accuracy and slightly higher resource requirements, alongside the current lightweight model. You’ll be able to switch between them and choose the balance of speed and accuracy that best suits your iPhone and your needs. We’ll also continue adding support for more languages. To thank everyone for the support, we’re extending our 50% launch discount for one more week. Buy once, with no Auday subscription. AppStore: apps.apple.com/us/app/auday-… Thank you for being part of Auday’s beginning. We’ll keep listening and making it better.
35
27
51
567
741,833
UNGETSU retweeted
The best AI recording hardware? You may already own it. Introducing Auday—now live on Product Hunt. Turn your Apple Watch into your all-day AI recorder. Beyond voice, Auday combines health, real-time stress, location, and other signals to understand your everyday life—not just answer your questions. Record on your wrist. Revisit your day with transcripts, timelines, highlights, reflection, and AI chat. Almost everything runs locally on your devices. Bring your own API key for AI. We don't even have a backend server. App Store launch offer: half price. Just $9.99 for a one-time purchase. No subscription. We'd love your support and feedback on Product Hunt. Take a look, join the conversation, and tell us what you'd want your watch to remember. Product Hunt: producthunt.com/products/aud… App Store: apps.apple.com/us/app/auday-… Website: auday.ai/
🤖 Made with AI
99
115
139
1,709
2,118,014
UNGETSU retweeted
Current Ai Race Situation:
1,478
4,416
712
61,993
9,375,706
UNGETSU retweeted
Google's team just dropped a 9-page PDF on Harness Engineering - one formula replaced prompt engineering: the twist: same Claude Sonnet, same benchmark - change only the harness the formula: Agent = Model + Harness here's the playbook in 6 steps: step 1 → add guides - AGENTS.md, rule files, constraint docs - each line is a past agent failure turned into a permanent fix step 2 → add sensors - linters, tests, validation scripts the agent runs on its own output before a human sees it step 3 → build the agentic loop - plan, execute, verify, fix - bounded retries, budget caps, escalation when stuck step 4 → externalize memory - the model forgets every session. the harness remembers state, decisions, and artifacts across all of them step 5 → enforce permissions - which tools, how many writes, what needs approval. the model doesn't enforce safety. the harness does step 6 → wire observability - track every tool call, cost, retry. trip wires fire when behavior drifts the result: your agent stops being a demo and starts being infrastructure - every failure makes the system permanently better, not just the next conversation this 9-page PDF is what comes after prompt engineering and context engineering bookmark and build the full system from the article below ↓
94
554
19
3,064
460,201
UNGETSU retweeted
We built vgpu.sh to ship shaders on vercel.com. Now it's open source. ▪︎ Minimal agent-first WebGPU library ▪︎ Run in the browser or headless Node.js ▪︎ Render in CPU sandboxes and CI tests ▪︎ Create reusable .wgsl modules
155
387
137
4,893
930,535
最近在尝试把 FUI 风格用到数据可视化大屏里 👀 中间的3D地图区域调用了我自己的three-scope-map-skill(已开源),再用Codex把地图、数据图表和整套 FUI 界面组合到一起。 还挺喜欢这种「科幻控制台 × 数据可视化」的感觉 ⚡️ #FUI #DataViz #ThreeJS #Codex #UIUX
18
26
3
283
25,310
50 Sitios Web de los que No Podrás Salir Durante Horas Una Vez que Entres 1.) 'zoom.earth' — Observar el mundo con imágenes satelitales en vivo 2.) 'flightradar24.com' — Ver todos los aviones en el cielo en este momento 3.) 'uniaverso.com' — Academia de IA para todas las edades 4.) 'windy.com' — Mapa en vivo de vientos y tormentas 5.) 'lightningmaps.org' — Observar en tiempo real los rayos que caen en el mundo 6.) 'earthquake.usgs.gov' — Lista en vivo de los últimos terremotos 7.) 'submarinecablemap.com' — Cables submarinos que transportan internet 8.) 'globalforestwatch.org' — Observar la deforestación de los bosques desde satélite 9.) 'worldometers.info' — Estadísticas del mundo actualizadas segundo a segundo 10.) 'internetlivestats.com' — Número de tweets y búsquedas en este momento 11.) 'thetruesize.com' — Comparar los tamaños reales de los países 12.) 'oldmapsonline.org' — Mapas de hace siglos 13.) 'davidrumsey.com' — Archivo de 150 mil mapas históricos 14.) 'openstreetmap.org' — Mapa mundial dibujado por voluntarios 15.) 'window-swap.com' — Mirar el mundo desde la ventana de alguien al azar 16.) 'virtualvacation.us' — Paseos virtuales por ciudades 17.) 'mapcrunch.com' — Teletransportarse a un punto aleatorio del mundo 18.) 'atlasobscura.com' — Catálogo de los lugares más extraños del mundo 19.) 'neal.fun' — Experiencias interactivas de conocimiento 20.) 'htwins.net/scale2' — Viaje a escala desde el átomo hasta el universo 21.) 'eyes.nasa.gov' — Explorar el sistema solar en 3D 22.) 'stellarium-web.org' — Mapa real del cielo en el navegador 23.) 'apod.nasa.gov' — La foto astronómica del día de la NASA 24.) 'images.nasa.gov' — Todo el archivo visual de la NASA gratis 25.) 'pudding.cool' — Artículos visuales narrados con datos 26.) 'ourworldindata.org' — El estado del mundo con datos reales 27.) 'gapminder.org' — Lo que creemos saber erróneamente sobre el mundo 28.) 'informationisbeautiful.net' — Visualizar datos complejos 29.) 'data.worldbank.org' — Datos abiertos del Banco Mundial 30.) 'data.tuik.gov.tr' — Base de datos oficial de estadísticas de Turquía 31.) 'archive.org' — Archivo de millones de libros, películas y software 32.) 'gutenberg.org' — 70 mil libros gratuitos con derechos de autor expirados 33.) 'openlibrary.org' — Registro de todos los libros del mundo 34.) 'loc.gov' — Archivo digital de la Biblioteca del Congreso de EE.UU. 35.) 'europeana.eu' — Archivo del patrimonio cultural de Europa 36.) 'dp.la' — Colección de biblioteca digital de América 37.) 'artsandculture.google.com' — Recorrer museos desde casa 38.) 'rijksmuseum.nl' — Descargar obras de arte en alta resolución 39.) 'wikiart.org' — Archivo de 250 mil obras de arte 40.) 'publicdomainreview.org' — Tesoros visuales olvidados de la historia 41.) 'openculture.com' — Archivo gratuito de cultura y educación 42.) 'metmuseum.org' — Colección abierta del Museo Metropolitano 43.) 'musicmap.info' — Árbol genealógico de los géneros musicales 44.) 'radiooooo.com' — Seleccionar país y década para escuchar esa era 45.) 'listen.hatnote.com' — Convertir ediciones de Wikipedia a audio 46.) 'wikipedia.org/wiki/Special:R…' — Pozo de conocimiento aleatorio 47.) 'timeanddate.com' — Tiempo, amaneceres y eventos celestes 48.) 'sciencedaily.com' — Flujo actual de noticias científicas 49.) 'arxiv.org' — Preimpresiones gratuitas de artículos científicos 50.) 'observablehq.com' — Visualizar datos con código en vivo Internet no se limita a esos cinco sitios que el algoritmo pone frente a ti.
123
5,172
128
13,221
980,000