What new LLM model types would like to see in the future?

Community
1

I’ll start:

I’d love to see:

  • text or image to 3d objects (i know openai did one a long time ago, but there’s hasn’t been any updates in what feels like forever)
  • text or 3d objects to animations
  • image to 2d and 3d vectors (similar to turn tables from adobe)
  • voice to facial expressions animations with semantics
  • text, image or voice to physics engine
  • tactile inputs to text
3 Likes
2

Text → Image Vector would be huge for me. 3D would be insane. I’d be happy with SVGs.

Image → 3D World would be cool as well. for VR applications. I’ve seen this a bit in the wild.

This would definitely be cool, especially if it tied in with RealTime API. But, at this point I’d be happy with some custom voices :rofl:

2 Likes
3

text to taste; or taste to text - may require tongue-on-a-chip technology

text to encryption text - I tried this once but found it too complicated. Basically this means we could convert any text into a reasonable looking text on another but related topic, but with the right code word, it would decode to the original text. With the likelihood that normal encryption will be broken in the next decade, this may be an alternative since nobody would know what the real intended text was- unless they know the code word.

2 Likes
4

text to encryption is fun, I saw it but got lazy in getting data to train a LLM on it

1 Like