Seen several replies to my post on the future of DL saying "who needs programming, LSTM is enough for everything (btw the brain is a LSTM)"
-
First, LSTM is not Turing complete. I'll grant you that RNN architectures can be Turing complete, for instance ConvLSTM is Turing complete
Second, what matters isn't that *in theory* you can express anything with a ConvLSTM. What matters is if you can do it in practice
-
-
Really it boils down to this. Finding & training a ConvLSTM that can render Markdown to HTML is near-impossible. But it's 20 lines of Pythonpic.twitter.com/ATke0YcDtC
-
Let's face it: we need richer ML primitives. We need more than matmul. I don't see why it's so difficult to grasp for some
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.