@themattsimpson Yeah it seems like job #1 when going into the field would be to learn everything you can about stats, but apparently not
-
-
Replying to @Meaningness
@themattsimpson ML is “statistical methods that work better than you’d expect given that they’re hard to characterize,” so figuring out >
1 reply 0 retweets 1 like -
Replying to @Meaningness
@themattsimpson > why they work should be the main focus of the field, but they’re mostly not interested in that; they just want dramatic >
1 reply 0 retweets 1 like -
Replying to @Meaningness
@themattsimpson > positive results, and of course don’t publish the negative ones, so we don’t even know *whether* they work better than >
1 reply 0 retweets 0 likes -
Replying to @Meaningness
@themattsimpson > you’d expect if you did some digging, and when I’ve done the digging, they turned out not to work better than expected.
1 reply 0 retweets 0 likes -
Replying to @Meaningness
@themattsimpson The dramatic seeming-successes, in each case, turn out to be because the problem is much easier than it looked at first, and
1 reply 0 retweets 0 likes -
Replying to @Meaningness
@themattsimpson once you understand the problem itself, you can see that solving it was not a big deal. The ImageNet results *might* be >
1 reply 0 retweets 0 likes -
Replying to @Meaningness
@themattsimpson > an exciting partial exception, but having done some digging on that, I’m pretty sure the problem is easier than it looks,
2 replies 0 retweets 0 likes -
Replying to @Meaningness
@themattsimpson > and that I know how to demonstrate that. So I might be able to kill off the current wave of AI hype with a year’s work.
1 reply 0 retweets 0 likes
@themattsimpson I’m torn about this because I killed one wave of AI hype in the mid-80s, and passed on killing another in early 90s.
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.