2d ago · 6 min read · Part 92 of Forgotten Theories, a series re-testing old scientific claims with modern tools. Find the rest under the #forgotten-theories tag. Konstantin Tsiolkovsky was a deaf schoolteacher in Kaluga who had been writing about space travel since the ...
Join discussion3d ago · 9 min read · The benchmark accuracy numbers for AI speaker diarization are deceptively good. Here is what actually happens on production audio — and why these problems persist despite years of improvement. Speaker diarization — the problem of figuring out "who s...
Join discussion3d ago · 9 min read · Whisper Large-v3 benchmarks look exceptional on clean audio. The question nobody publishes is what happens when the audio is not clean. OpenAI's published benchmarks for Whisper Large-v3 show word error rates under 3% on standard test sets. Those nu...
Join discussion3d ago · 14 min read · The most common architectural mistake teams make when building AI agents isn't choosing the wrong model or writing bad prompts. It's choosing the wrong number of agents. Multi-agent systems have an almost gravitational pull for engineers. They feel a...
JJohn commented
