OpenAI have released a new speech-to-text model that also supports diarization (speaker separation and attribution): GPT-4o Transcribe Diarize (aka “gpt-4o-transcribe-diarize”). It’s priced the same (estimated) cost of regular gpt-o4-transcribe and Whisper. Immediate findings: - chunk limit of 1400 seconds chunk limit of 1400 seconds - not avail... Read more... 24 Oct 2025
OpenAI have released a new speech-to-text model that also supports diarization (speaker separation and attribution): GPT-4o Transcribe Diarize (aka “gpt-4o-transcribe-diarize”). It’s priced the same (estimated) cost of regular gpt-o4-transcribe and Whisper. Immediate findings: chunk limit of 1400 seconds not available over the Realtime API ... Read more... 24 Oct 2025 - 2 minute read
OpenAI has released their browser: ChatGPT Atlas. Superficially, this feels like Chrome - which could be due to Chrome engineer Dan Fisher having had a hand in it. Conceptually though, this follows, for example, the Dia browser in that it builds AI features in. Most prominently, there is an “Ask ChatGPT” sidebar that opens a chat side panel - mu... Read more... 22 Oct 2025 - 1 minute read
Motivation Read more... 19 Oct 2025 - 4 minute read
Motivation For a recent writing project that built on an earlier piece I had written in a hurry, I wanted to see if GenAI could help the way it already does for programming. Because my ideal venue was arXiv (an open‑access preprint library), I wanted a LaTeX‑based workflow for typesetting and citations, rather than the—on these fronts limited—Ch... Read more... 19 Oct 2025