Mukund Jha
👤 SpeakerAppearances Over Time
Podcast Appearances
My definition is when a model or a general AI system can actually do most of the work that humanity can do today.
Right.
Knowledge work.
Right.
And it's sort of sentient.
Right.
And it happened, right?
I mean, those are very minor things.
It's a start.
Yes, there are many such cases.
For example, Anthropic has just published a study on alignment in which
like model go back to test and he lied to himself to just pass that test.
So it had knowledge that I was being tested and was able to fake that test.
So I think that is a big risk right now.
Nobody knows what is on the other side of AGI, ASI.
Are we in a terminator scenario?
Are we in a matrix scenario?
We would have thought that this is like a very fictional thing.
But people who are closer to the field, they truly understand that it is a real possibility.
Right.