Gregory's ServerI'm hoping to turn this into a series of YouTube interviews with people building cool data projects where we nerd out about what they've built and how they built it, so I'm optimistically thinking of this as episode one! https://www.youtube.com/watch?v=t_S-loWDGE0
|
9 comments
 A YouTube comment asked the price difference between Gemini 1.5 Flash and OpenAI's Whisper Whisper API is $0.006 / minute, so an hour of audio = 36 cents Gemini 1.5 Flash is $0.075 for 1 million tokens, 25 tokens/second of audio so an hour is 0.675 cents Over 50x cheaper! @simon that whisper price quote isn't competitive https://groq.com/pricing/ and I'd be curious on the accuracy differential @derenrich problem with Groq is they haven't actually launched their billed API yet, so you're stuck with whatever their free tier will let you do  @simon Is the quality of Gemini and Whisper in Speech-to-text on the same level though? @xsc from what I've seen so far they do feel similar in quality - and Gemini can do extra tricks like diarization and tone-of-voice analysis that Whisper can't I remain paranoid about the risk of Gemini accidentally acting on instructions within the audio, but I've not (yet) seen that happen - so possibly more of a risk with deliberately malicious audio |
The VERDAD prompts are pretty complex - Rajiv shared this example of a conversation he had with Claude 3.5 Sonnet to further iterate on the existing prompt used with Gemini 1.5 Pro https://gist.github.com/rajivsinclair/8fb0371f6eda25f9e5cc515cd77abd62