In the future, everything will be like that vapid chicken wing podcast (the one where they bring on an interesting, talented person and then waste half an hour interviewing her about sriracha)
So Ira Glass will be narrating This American Life while simultaneously reviewing different varieties of Doritos, etc.
...or the producers of The Rest is History will add the Planters Peanut Man as a third host
...or Marques Brownlee will review every product in relation to how well it works with Bose headphones
My favorite relentless one is Tracker (Amazon Prime), who spend approximately 30% of screen time dedicated to showing off a GMC pickup and Airstream, but the most egregious was one dialog line:
> As Colter enters and gives them hugs, Velma remembers that they got a gift for him. Reenie hands him the gift - very conspicuously packaged in an Amazon box with its trademark logo and blue tape - and says, "I've gotta say, next-day delivery is pretty sweet. Thank you, Amazon Prime!"
...it powers through tough grease and grime
...with no harsh smells!
The future is Fantastik®.
Then you would need to set up a server that would do all this and serve as a 'mirror' to your podcasts without the ads.
1. Interviewing a guest while they are eating insanely spicy food is an extremely novel idea and the guest's reactions and answers end up being really interesting and unexpected as a result. It humanizes famous people in a way I've never seen before because you can't just bluff your way through it.
2. The hot sauce vendors are often small companies or indie makers (at least they used to be). This is way different than reviewing 11 bags of Superman Transformers 3D Doritos Walmart Product Placement for the next blockbuster.
3. Hot sauce is interesting! Nearly every culture on every continent has hot sauce. They are made from a huge variety of interesting and unusual ingredients but are also simple and can be made at home. Hot sauce hasn't been explored in this way in popular culture.
I also have a setup like this, I transcribe with Whisper and send it to OpenAI 4o-mini to detect ads then clip those segments with pydub, but my prompt must be lacking because the success rate on detecting ads is maybe 60%
I think it's better than 60%, but I should definitely set up some evals.
I split the text by sentence, but was considering having the LLM try and put into paragraph (that might conceptually chunk commercial sentences together), but what I've got has been good enough for me.
I wanted to switch to Flash 2.5, but it looks like they increased the price a lot.
I think I could do a fair bit of ad identification just with text heuristics: "This podcast is sponsored/supported by...", etc.