sanjar-notes / textual_computation

Understanding text, textual computations and their ramifications
0 stars 0 forks source link

Is text data a good starting point to learn and reason about AI/ML #3

Open sanjarcode opened 2 years ago

sanjarcode commented 2 years ago

Text doesn't have the learning overhead of something like computer vision/audio.

The media --> encoded information step takes is difficult.

But text --> encoded information is very simple.

Text can also help with coming up with more algorithms than media because of it's somewhat rigid abstract mathematical structure, which may or may not apply to models with media input, but can serve as inspiration for them.

sanjarcode commented 2 years ago

Maybe this is the reason why companies have spent considerable resources into textual AIs instead of media AIs.

Textual algorithms being rigid, are easy to develop and may be 'softened' by making their discrete structure 'continuous' and then applying them to media (audio/visual etc).

sanjarcode commented 2 years ago

Just explore if this is true?