pull down to refresh

This is nice. Saved yourself half an hour of Photoshop and potentially a 70Ksat/mo subscription.
LLM models invite skepticism because of hallucinations and privacy concerns, but when used judiciously, they do have the ability to make our lives easier.
Just have to understand the limits I think. And to do that, you have to test the limits, but be honest and critical when you assess the results. I really think that that's all there's really to it.
I was skeptical of whisper as a local speech-to-text engine because anyone that has checked YouTube captions knows how annoying the mistakes can be. Turned out its better than whatever goog uses for yt but of course slower - probably a tradeoff they made to minimize cost.
Cool! Haven’t heard of whisper…until today. Will go and check it out
reply
https://github.com/openai/whisper - let me know how you fare.
I'm currently using https://github.com/argmaxinc/WhisperKit on my Mac simply because that has optimization for Apple's NPU built-in; the CPU-only runs are really, really slow.
reply