technology
OpenAI's speech-to-text model. In episode 01 it is the 2023 baseline: Chad's first transcription tool for the archive, good enough to start and too error-prone to build on. Episode 01, at about 18 minutes.
Across the episodes
01
Enters as the 2023 baseline whose error rate made the project uneconomic at first
Marks the 2023 starting point: transcripts that misheard names and lacked speaker labels, so every hour needed heavy manual cleanup.
"I started with one series and started to take it and put it through Whisper, which is OpenAI's speech to text."Chad Nantais, Syntopic, Episode 1 (11 Oct 2026), 17:58
Connections
limited byLarge Language Model
A poor transcript poisons every later extraction step: garbage in, garbage out.