Finishing up on the work I started in Spring 2025 (“Physics … in 30 Lectures” Series Reflection), I finished putting together the Physics 4B in 30 Lectures, which is available at the link.
With these major projects done, I had some time to think about what to finish next. For some time, I thought I would put together “companion exercises” videos, which frankly won’t help my students that much—they already have access to homework problems and homework help videos—but would add content on the YouTube channel.
Instead, I decided to work on the transcripts. So far, all my videos only used auto-transcribed transcriptions. It’s technically an ADA violation, but the amount of time that would be necessary for the hand-corrected transcriptions was too much, and I was mainly worried about doing all that work to fix the transcripts, only then for me not to use the video anymore. Well, now I have a set of 90 videos which I’m hoping I will use at least for 5 years or so, so I think this is the perfect time to put in the necessary effort to fix this ADA violation.
I started working on Physics 4A lecture video transcripts, and the progress was really slow—I think it took the whole day and a little more for a single relatively shorter video. But I think I found a good workflow that utilizes ChatGPT. It turns out, as it could work on the lecture questions from the auto-transcribed transcripts, it can also fix what the transcript is likely to be from the auto-transcription alone, without directly accessing or re-transcribing from the audio.
And ChatGPT 4o is especially good at this work. See below example. Now ChatGPT did have some access to the context which might have helped, but, you know, I, as the lecturer, know all the context, and it still would have taken me quite a bit of time to do the level of correction that ChatGPT did automatically.
In this workflow, it takes me maybe about double the video length to fully correct and review—to the extent needed—the corrected transcript, and I think I’m actually going to go back to re-do some of the transcripts that were done earlier … with some faults. But even with re-doing two … two and a half lecture videos, I think I’ll have enough time to have some lead time built in, before the semester starts. I’ll probably be able to be about 2 weeks ahead of students, as I fix the transcripts. And I have some plans for the transcripts once it’s done; it’s not just for deaf students; there are other materials that I’m thinking about making, and I would need relatively accurate transcripts of the lecture videos to start working on that.
P.S. I actually started this work using ChatGPT o3, because I’m used to using that model for coming up with the lecture questions, but actually the o3 model was doing “too much thinking”, mainly changing the sentence structures around too much, despite numerous instructions to be “verbatim”.