AI and machine learning
More on quantum computing (including that Rigetti news and more), but long-form style. Based on everything I'm reading (and nothing I actually know about the science), three years doesn't seem like too crazy a prediction.
Well, DARPA is. And DARPA has arguably been working in AI for years on previous budgets, even if not by that name specifically. It also has a pretty good track record for innovation, so let's see what it does beyond what the web companies currently dominating AI are doing.
This sounds like a good idea in theory, but BMW really needs to nail security and reliability. Hacking Alexa-enabled things is only going to get more common, and even small disruptions while driving can be dangerous.
This story falls into a massive gray area that spans racial profiling, privacy, civil rights, policing and then some. I could easily argue both sides of this if forced to. (Also, I would note that I've been pushing for an expanded definition of "privacy" for years now. But until we redefine age-old rules, it's kind of fair game once you leave the house.)
I think it's fair to measure AI startups differently only if they're doing something genuinely difficult or slow-going (e.g., consulting with large enterprises), but we're also seeing a glut of SaaS-based "AI for X" startups that will likely be measured by the same yardstick as their non-AI peers. Especially because once the product is out, it's game on in terms of business metrics.
Consider this an appendix to the link above. There are already solid options in this space, so the AI focus had better really add something useful.
And another B2B AI startup, also tackling a space with some serious competition -- including, it seems, Microsoft with its recent acquisition of AI startup Bonsai.
A writeup of some recent research by the Allen Institute for AI. Text understanding is important work, but nothing on which I'd trust an AI for anything important today.
I'm pretty sure this is just another take on the Facebook research I linked to earlier this week. If there's a twist, it's that Facebook has now open sourced the code for training these models.