Question to help with understanding

[umaar]

Hey this project looks great! Thanks for releasing it. I'm probably being slow here, but trying to understand, what is the unique selling point of semantic search in relation to the example queries? Some of the example queries seem to just return exact phrase matches, or partial word matches - but not sure how semantic search plays a role here. I see some queries return results that don't appear to have specific word matches, so presumably, that's the benefit of this approach? E.g. searching for 'space' returns a result that has 'universe'. If that's intentional, that's pretty cool!

Curious about a few other things:

• If we set this up ourselves, can you share some example OpenAI costs you've incurred for an example playlist?
• How easy would it be to adapt this so we can write queries that use playlist video transcripts as context. My basic understanding is you can inject transcripts into the prompt itself, which OpenAI can use to derive an answer from, but I guess that's not practical to send a ton of transcripts in each query e.g.
  • User asks: "Why are wages suffering?"
  • This web app responds: "People are seeing wages destroyed by inflation, food and gas prices."
  • Which it hopefully learnt from: 'E86: Macro outlook': "...down they're seeing their wages get destroyed by inflation food and gas prices being much higher so there's a really good reason why people are sentiment is so negative out there..."

Such a great project, thanks again.