I can’t help but notice the timing of this. ML is starting to fizzle out as a “savior” of the industry after lofty promises were never fulfilled.
Goodfellow was hired right at the peak of ML hype, so I have to assume his salary/total compensation was extraordinarily high, so he was very expensive to keep around.
Apple would’ve made an exemption for exceptional talent. They didn’t make the exception for him.
Don't know about your first point on ML, but I think the second point is the more salient one. If Goodfellow was that productive and that much of a value add they would have found a way to keep him.
I don’t mean to disparage ML generally. I think it has a lot of promise in academia, but its utility in the consumer electronics industry (especially one that’s pivoting towards entertainment services) seems pretty limited.
ML has reached a level of maturity and ubiquity that it is having an effect on almost every area of computing. I have personally seen applications for ML that have moved the needle of consumer electronics applications. A lot of this is old hat stuff at this point but there is also a ton of value. Things like finding the white balance in photos, the next letter on a keyboard or how quickly to charge a phone are all areas that benefit from simple ml.
I would for one would really appreciate it if Apple photos had a decent search function.
Maybe these don’t move the needle as much as Wall Street analysts hope but these marginal improvements add up in a highly competitive space and I can’t see a world where mobile manufacturers turn their backs on ML.
"ML is starting to fizzle out as a “savior” of the industry after lofty promises were never fulfilled."
What? Siri, text completion, all the health- and activity-tracking features of Apple Watch, computational photography, Photos app, Face ID, and many, many more depend on machine learning. And let's not forget the Neural Engine in M1.
Especially in biosciences. But it’s probably the case that the hardware that Apple produces don’t use fancier methods such as DL/RL that much and depend on more traditional methods. You need capable GPUs for the state of the art stuff (e.g. DLSS by Nvidia).
It used to be AI then was renamed to ML (I know, ML is supposed to be a component of AI). AI has been touted since, let's see, my first encounter with it was in the early 1990's. It sucked. Then again in the early 2000's. It sucked. ML today is almost somewhat barely usable with hurculean efforts, IOW...it still sucks.
this is not a very informed take. ML is increasingly used across all kinds of use cases. Some of the wild uninformed visions haven't materialized cause they never do in any tech fad, but saying ML is fizzling is just totally inaccurate.
> Apple would’ve made an exemption for exceptional talent. They didn’t make the exception for him.
Mmm. Don’t count on it. Apple will bend a lot for you, but stuff like office location was never very flexible even before the pandemic. If your team switched offices to an inconvenient location, tough luck. Ultimately it was on you to adapt.
If you were a top performer and threatened to quit over it… well, Apple has a deep bench.
A probable take I heard from an insider: companies used to boast about their new AI developments for investors, but now are very hush-hush to avoid regulatory scrutiny
Goodfellow was hired right at the peak of ML hype, so I have to assume his salary/total compensation was extraordinarily high, so he was very expensive to keep around.
Apple would’ve made an exemption for exceptional talent. They didn’t make the exception for him.