screwlisp@screwlisp@gamerplus.org
@GrantMeStrength@hachyderm.io
No they still have the deep learning thing.
You train your DL to feed your dogs. You run a test. Well, it forgets to feed the dogs in May but otherwise it was OK.
You add some training data indicating that dogs would in fact be fed. You run your test again. Lo, the dogs are fed in May this time. (Or you add more training examples of not starving the dogs in May again).
<- Like how a transformer can never, ever do cryptography.
@vnikolov@ieji.de
screwlisp@screwlisp@gamerplus.org
@GrantMeStrength@hachyderm.io
In my dog example, the point is that piling on additional specific training doesn't mean that the system has stopped doing crazy things in the future. Just if you fill its training data with FEED DOGS THROUGHOUT MAY, when you ask it about feeding dogs throughout may you're likely to see more FEED DOGS THROUGHOUT MAY included.
@vnikolov@ieji.de
