AI 2027 is a good read. Both endings seem wildly unrealistic, but as a warning it did make me think. So it worked. Just seems greatly accelerated beyond a few years from now.
Gaëtan Perrault @manton TY for sharing this, the whole presentation is well done.
I do worry about some of the timelines as they don't seem to respect some of the reality limits currently in place.
January 2027: "We think that by this point Chinese intelligence would have compromised OpenBrain in various ways for years"
But that's now... Or before now. For this statement to be true, it would be happening right now, we would all know it was happening and nobody was stopping it? ... /1
Gaëtan Perrault @manton and the timelines on compute growth seem pretty weird as well.
They talk about China taking over Taiwan in 2027 in order to claim TSMC. But there's only one TSMC and their order book is completely backlogged through that time range.
To say nothing of trade wars limiting the movement of chips and rare earths required to produce new ones. So there's currently a max available compute by 2027 and it's not clear that Agent-4 fits in that space... /2
Gaëtan Perrault The whole piece also seems to ignore the real world "things" that humans interact with. The AGI is a super genius that can program "anything", but it doesn't have any real world tendrils.
It's not building housing or factories or railways, it's just a virtual entity slowly building virtual power with existing power structures. While somehow sucking up tons of energy that nobody notices?
I like this piece as food for thought. But some of that food is definitely the missing pieces. //
Manton Reece @gatesvp Right, it’s a big leap from AGI to controlling so much physical infrastructure. That is absolutely something we have to watch out for, but I can’t see it playing out like this in just a few years.
Gaëtan Perrault I know it's been a few days, but this story stuck with me. I realized what was getting to me, it was this thing I heard in my early Silicon Valley days.
"We tend to overestimate what can get done in two years, but underestimate what can get done in ten years."
2 years feels aggressive for a lot of this. But 10 years doesn't 🤔