Rob Miles (in SF) Profile
Rob Miles (in SF)

@robertskmiles

Followers
31K
Following
11K
Statuses
13K

Explaining AI Alignment to anyone who'll stand still for long enough, on YouTube and Discord. Music, movies, microcode, and high-speed pizza delivery

London or SF
Joined April 2010
Don't wanna be here? Send us removal request.
@robertskmiles
Rob Miles (in SF)
2 years
I'm launching a new YouTube channel: AI Safety Talks! It hosts high quality talks from AI Alignment researchers, giving a bit more depth and technical detail than my own videos! Go subscribe and hit the bell!
22
48
320
@robertskmiles
Rob Miles (in SF)
1 day
@heartbulbous @JeffLadish Whaaat? Do you have pictures?
1
0
2
@robertskmiles
Rob Miles (in SF)
1 day
RT @JeffLadish: I’ve never felt more seen by a meme, thanks @robertskmiles
Tweet media one
0
96
0
@robertskmiles
Rob Miles (in SF)
1 day
@RatOrthodox @JeffLadish If it were possible to show both brain and heart going in a direction that wasn't one of the directions either of them is going in the first image, I would have done that
0
0
8
@robertskmiles
Rob Miles (in SF)
1 day
Life in the bay: A: Hey Rob, what have you been up to today? Me: Mostly just rest! I haven't been doing that enough lately A: Oh yeah I'm learning Rust too, how are you finding it? Me: I said REST A: You're designing an API?
10
11
648
@robertskmiles
Rob Miles (in SF)
1 day
@AndrewCritchPhD Is it fair to say that the overwhelming majority of people who think they know how to steer AGI, actually don't?
1
0
34
@robertskmiles
Rob Miles (in SF)
3 days
RT @slatestarcodex: People had lots of questions about my drowning child tweet yesterday - eg do I believe in infinite moral obligation? Ra…
0
156
0
@robertskmiles
Rob Miles (in SF)
9 days
@mark_riedl I set myself a timer to check in on this. After a quick check I think I'd call it 'inconclusive'. You have stories like this But I'm not aware of a specific AI-written screenplay being produced and considered good
0
0
1
@robertskmiles
Rob Miles (in SF)
9 days
@TechnoPulp @sebasbaur @yishan @ESYudkowsky We don't have good ways to know the goals, just the actions. If you hit the reward button every time it does a good thing, is its goal "do good things", or "cause the button to be hit"? As long as there's no opportunity to seize control of the button, the behavior is the same
0
0
4
@robertskmiles
Rob Miles (in SF)
10 days
@TechnoPulp @sebasbaur @yishan @ESYudkowsky (the cancer example isn't great because no human wants to cure cancer as a terminal goal. Ask why repeatedly on that and you might end up with something like "I want there to be less suffering because suffering is obviously bad, what do you want from me". That's a terminal goal)
0
0
8
@robertskmiles
Rob Miles (in SF)
10 days
@TechnoPulp @sebasbaur @yishan @ESYudkowsky I make a distinction between terminal and instrumental goals, idk what "regular goals" are really. The distinction is, is this goal a way to achieve another goal, or not
0
0
1
@robertskmiles
Rob Miles (in SF)
10 days
Some animals can understand a pointing gesture (dogs, elephants, maybe cats and maybe chimpanzees?), while others never seem to get the concept, and will just look at your finger. Does anyone have any really good video of an animal failing to understand pointing?
11
0
88
@robertskmiles
Rob Miles (in SF)
10 days
@TechnoPulp @sebasbaur @yishan @ESYudkowsky (I talk about this in more detail in this video: )
1
0
9
@robertskmiles
Rob Miles (in SF)
11 days
@TechnoPulp @sebasbaur @GroundhogStrat @yishan @ESYudkowsky The assumption is just that it will have at least one This may be the one we 'gave it', which could still be bad because it may not be what we intended to give it, or that whole process may not work and it could end up with some distorted version, or something basically unrelated
2
0
2