Neural networks can be massive.
Our previous diagram showed only two hidden layers, but the theoretical limit is infinite.
And with these extra layers, some pretty incredible stuff can start to happen.
Without human guidance, these models can take in data (including text, images, and audio), and find complex patterns that we’d never find otherwise.
This makes them great at tasks like reading medical imaging, diagnosing diseases, and discovering drugs.
But there are limitations:
1) It requires a lot of computing power
2) It can be very expensive
3) It usually needs lots of data to learn from
Despite this, at the time of writing, there’s a worldwide arms race to pour as many resources into these models as possible.
And even if the improvement curve starts to flatten, changes will still be rapid.
Which is why it’s so important to future-proof ourselves.