What was your most incorrect thesis?
Conversation
I thought ML would be adopted for a lot more problems. Much more emphasis would be placed on human labeling, active learning, other practical ways to make the system efficient.
1
1
I thought speech and translation would be way better than it is, both at Google and in more open domain environments
1
1
I thought video models for things like human pose estimation would replace hacky 2D snapshot based “video surveillance”
1
1
I thought this would all be done through software on good but commodity hardware. The self driving and driver assist being as slow is at is — that I excepted and I think that industry is fine. Extreme promises from Elon was always just hype
1
Perhaps I was naive, about the willingness to spend $$$ on getting deep learning working for hard problems, including the active-learning human component.
Investment has not been small... but apart from self-driving, not really into new problems. And less into R&D than I'd like
1
I read something like this in 2016 on 's blog. Thoughts we'd be a lot further along to be honest...
1
I have some theories why ML adoption and ML-driven transformation has gone slower than expected over the past 6.9 years. Should write a blog on it... will probably get some push back and I'd enjoy that