An audio transcription system transcribes one hour of audio in 13 seconds on a MacBook Air.
The chart will appear after repeat observations. The current metric comes from the source.
141 source pointsReal observations only. History before source connection is not reconstructed.
Whistle is a speech-to-text model with a size of 16.9 MB.
4 days agoGitHubStrata is an inference engine that allows Qwen3.8-Flash-Next to run on consumer hardware with a one-click install for Windows and Linux. It supports OpenAI and Anthropic APIs on localhost and includes optional image input.
18 days agoLessWrongThe article investigates the origin of a popular anecdote about a pottery experiment where students evaluated based on quantity produced better quality work than those focused on quality. The author traced the story to its source in the book 'Art & Fear' by David Bayles and Ted Orland, finding it was not a scientific study but an anecdotal account. The experiment's methodology and results are described as lacking rigorous scientific validation.
22 hours agoLessWrongRobin Hanson did not invent the concept of a martingale process, but he popularized the idea that probability assignments should form a martingale. This concept is linked to the Efficient Markets Hypothesis, which states that short-term price movements cannot be predicted. The idea was developed earlier, but Hanson contributed to its visibility.
4 days agoLessWrongThe author conducts a personal evaluation of new AI models by asking them to design a novel game with unique mechanics. Previously, models often suggested 'time manipulation mechanics,' but newer models like o1 OpenAI and Opus 4.6 proposed original ideas, including a smell-based game and a game where rules are defined retroactively during play.
yesterdayHugging FaceHugging Face model card for abenzerps/Qwen-Image-2.1-Uncensored-Quantized. Task identifier: text-to-image. Follow the source link for details.
22 days ago