|
@fchollet | |||||
|
The most cost-effective way to defeat the Terminators is to introduce a NaN in a SkyNet training data example
|
||||||
|
||||||
|
Rob Miles
@robertskmiles
|
1. stu |
|
Is AGI going to be so carefully designed that we're safe, or is it going to be so buggy that we're safe?
You kind of have to pick
|
||
|
||
|
Cassiano
@angusbirds
|
1. stu |
|
if AI design follows the same path and abide by the same rules of software design in the wild, the answer is pretty much "so buggy we're safe"
|
||
|
|
||
|
Maryllia Kiss
@marylliaKiss
|
1. stu |
|
what if it applies data imputation first?
|
||
|
|
||
|
Robert Peharz
@ropeharz
|
1. stu |
|
Or even worse, what if it uses a decent tractable probabilistic model, which can naturally deal with missing data by means of marginalization (assuming -- of course -- that NaNs encode the MAR scenario)? I've heard these things have become pretty scalable recently... 🤔
|
||
|
|
||
|
Pranav Hari💻👷
@pHequals7
|
1. stu |
|
Sike
Skynet is russian
It uses catboost
|
||
|
|
||
|
Ric Angius Fosci 📈⌛️
@ricangius
|
1. stu |
|
Still gonna try a water bucket first
|
||
|
|
||
|
David Castillo
@DCasBol
|
1. stu |
|
But what if it's all random forests internally? 😱 We're doomed!!
|
||
|
|
||
|
Danilo Bzdok
@danilobzdok
|
1. stu |
|
...or convert some data columns to string type.
|
||
|
|
||
|
Oscar Rojas, CFA
@oscarrojas_1
|
1. stu |
|
|
||
|
Miguel
@Deplorable_Mi
|
1. stu |
|
It just occurred to me that SkyNet likely CREATED John Connor (with regards to him being a warrior and SkyNet nemesis) when they sent terminators back to kill him.
The rest of the world probably figured this out 20 yrs ago... Better late than never, I guess...
|
||
|
|
||