11 ms·
That is an accomplishment! I'll have to get it to explain, since most of that shit is incomprehensible to me...
by johnea 3y ago
That is an accomplishment!
I'll have to get it to explain, since most of that shit is incomprehensible to me...
- BoorishBears 3y agoI actually worked on this for a toy app: you can create tokens for images, and overlay those in the embedding space you do words That allows LLMs to treat images like "words" and speak on their contents. - That being said looking at the demo I'm not sure Bard isn't just using the video transcripts. I was using them to guide seeking, but you could answer all the test questions shown in the video without image understanding