it will save the world as a joke. it promises
-
-
Replying to @VesselOfSpirit
if you don't make any assumptions then nothing can be funny. because the assumptions can't be violated
1 reply 2 retweets 7 likes -
Replying to @VesselOfSpirit
it's serious about saving the world as a joke but not about promising. it's too hard a thing to be promisable
2 replies 0 retweets 4 likes -
-
-
-
Replying to @VesselOfSpirit
I think if I were trying to save the world as a joke, I would first get in touch with Nick Hay and Adam Safron and talk to them about value learning.
1 reply 0 retweets 0 likes -
-
Replying to @VesselOfSpirit
Those two because they're LW adjacent, they have some credentials in algorithms and neuroscience respectively, and because they've been recently publishing on neurally-guided value learning, for example https://arxiv.org/abs/1811.03493
1 reply 0 retweets 1 like -
Replying to @preinfarction @VesselOfSpirit
Value learning as a topic because it's not advancing AI capabilities (or AI's ability to self represent and recursively self modify, as MIRI's reliable agent project is), while still being a necessary component of FAI design (regardless of whether reliable agent design is net +).
1 reply 0 retweets 1 like
In addition to being one of many necessary components, I think it's getting less attention that it deserves right now, and it could benefit even from people who can't do original algorithmic research.
-
-
Replying to @preinfarction @VesselOfSpirit
And the reason I'd look at neurally-inspired value learning is that the dominant paradigm of inverse reinforcement learning from behavior doesn't seem adequate in principle to generally pinpoint values:https://www.lesswrong.com/posts/DfewqowdzDdCD7S9y/agents-that-learn-from-human-behavior-can-t-learn-human …
0 replies 0 retweets 1 likeThanks. Twitter will use this to make your timeline better. UndoUndo
-
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.