I built a tool to track what AI leaders say and how their views change on video

gcsydney

Hi HN, I'm Gaurav. I've been building a tool that turns YouTube videos and podcasts into searchable transcripts. But somewhere along the way, I became more interested in a different problem. How do you keep track of what influential people actually say, and how their thinking changes over time? Take Sam Altman, for example. In October 2025, talking about compute, he said, "there will come a glut for sure." In June 2026, he said, "there will be a shortage forever." Two statements, eight months apart. There may be perfectly reasonable context behind both, but I found it interesting that there wasn't an easy way to see them together, with the original recordings. It currently follows 20 people in AI, including Sam Altman, Dario Amodei, Alex Karp and Marc Andreessen. Every day, it looks for new interviews, podcasts and talks where these people are actually speaking, rather than videos of other people talking about them. For each appearance, you get a short summary, key quotes and links to the exact moments in the original recording. The feature I'm most excited about is *"What changed".* It compares what someone says today with their earlier statements on the same topic. You can see both quotes, their dates and the original sources. I'm not trying to catch people contradicting themselves. People should change their minds when new evidence emerges. I just think it would be useful to have a record of how their thinking evolves. Building it has been more challenging than I expected. A surprising number of videos with someone's name in the title don't actually feature them speaking. Reuploads are everywhere. And getting AI to reliably attribute statements to the right person is harder than simply generating a transcript. I've built filters to handle much of this, and every published "What changed" comparison is still manually reviewed. It's far from perfect. Speaker identification needs work, some podcasts aren't transcribed automatically, and I'm starting with English only. On the economics side, a transcript and summary currently cost roughly one US cent in model processing. That makes tracking a fairly large history surprisingly affordable. Everything is free to read. You can also follow up to three people, receive updates when they speak, and ask questions like "What has Dario Amodei said about open-source AI recently?" with answers linked to the original recordings. I'm still figuring out which part is genuinely valuable: keeping up with people, searching their past statements, or spotting changes in their views. I'd love some feedback from HN. Who else should I track? Would you actually use the "What changed" feature? And if you find something incorrectly attributed or summarised, please report it. I'd rather find the failures early.

saintfire

Quoting a Kai Lentit video as Dario Amodei changing his mind on regulation is peak slop. Cool idea, I guess, but its just more slop to track the slop.

toobulkeh

Speaking nothing of the implementation, this is a very valuable concept. I’m curious if statistically there are trends to identify in beliefs or how beliefs spread over societies. You could likely use that data to determine thought leadership and information diets and measure things like distribution, how far the Bay Area is ahead, and other meaningful data for investments, etc.

deadbabe

AI leaders aren’t important, why not track actual politicians?

iafine

YouTube's auto-generated captions are a great product, but one problem is that they can't tell speakers apart in podcasts with multiple people. I'm building something that also works with YouTube transcripts, and it has the same problem.

qwerty_clicks

Interesting concept but useless without signing up for who knows what just to start trying to see what things all about.

patrick_rtk

super nice idea , should do that on politics

Lerc

I think this would be a good thing to have, but it has to be comprehensive. Selective ommission and reduction of context would allow any narrative you chose to write. In other words for this to work you really need to cover enought scope that it would be surprising to find a public statement that was not covered here. That's quite a lot of work, but anything else would be susceptible to cherry picking.

ks2048

Alexandr Wang "says his favorite Muse use case is having it coach him in MMA by watching gym cameras and giving feedback between rounds" Okay, you've convinced me I don't want to track what these guys are saying. On a serious note, it looks good. Although to really track what they say, you'd probably want text too, e.g tweets and blog posts.