Umless is an open-source dictation app that turns natural, hesitant speech into polished text and types it into the active application. It combines Parakeet or Whisper speech recognition with a locally run Qwen language model, including a fine-tuned compact model, to remove fillers, resolve self-corrections, and adapt writing to the app: prompts for AI tools, formal messages for email, or casual text for chat. On macOS, it can also read nearby screen text to spell names correctly and use conversation context; a command mode can draft or revise text based on spoken instructions. It can learn corrected spellings, while offering settings to disable that feature or screen reading.
All speech processing and screen reading run on-device, with no account, cloud service, or telemetry. The Apache-2.0 project supports macOS, Windows, and Linux, though screen reading is currently macOS-only and the other desktop versions are newer and less tested. Users hold a shortcut to record and release it to insert text. Three model tiers trade storage and speed for accuracy and language coverage: Lite is English-only and occupies about 678 MB; Standard supports 25 European languages and occupies about 1.95 GB; Max supports 99 languages and occupies about 3.53 GB. In an 11-clip test, cleaned-text word error rates ranged from 5.7% for Lite to 1.8% for Max. Installation builds from source and downloads the chosen models, so setup time and hardware requirements vary.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
A Hacker News poster asks which practical AI skills individuals can learn online to benefit economically from LLMs and agents. They want specific, hands-on recommendations suited to someone with a moderately technical background, without requiring a degree. (news.ycombinator.com)
Former OpenAI researcher Will Depue claims hidden initialization patterns in six Chinese open-weight AI models match Nvidia Blackwell hardware, including systems restricted from export to China. He says the matches were reproduced on real chips and independently checked by researchers, though the claims are based on his forensic analysis. (twitter.com)
Antithesis built a system that beats NetHack on nearly 80% of game seeds within five minutes, across character builds. Successful runs take a median of under two minutes, advancing the company’s goal of creating smarter fuzzers and autonomous systems. (antithesis.com)
OpenCyclone is a free, open-source Android agent that uses your existing apps to carry out tasks through a hidden screen. It asks for approval before actions like sending messages and lets users connect their own AI models. (opencyclone.vercel.app)
Sam Bankman-Fried’s $500 million investment gave Alameda nearly 14% of Anthropic, a stake that would be worth about $37 billion at the AI company’s expected $1.5 trillion IPO valuation. The stake was sold in bankruptcy to repay FTX customers and creditors after Bankman-Fried’s businesses collapsed. (ft.com)
llama.cpp and ggml have reached v1.0, establishing stable public C APIs and clear versioning rules for breaking changes. The release strengthens the foundation for running local, multimodal AI across a wide range of hardware. (huggingface.co)
Today's best Hacker News stories, summarized and screenshotted, one email a day.