Conversation

This article, fromthenew.world/p/diminishing- BTW, roughly matches my take from the hardware side, although I think you also need to include l the shortage of good data to get AGI odds as low as 5%. In 2017 I would have gone with 12%, though some of the difference is elapsed time.
Quote Tweet
Replying to @Simeon_Cps
I’ve been updating my timelines slowly upwards for the last two years. worldbuild.ai/W-0000000429/ Now feels maybe 20% too optimistic about the rate of change being fast. (Which is to say pessimistic about the world being relatively safe)
GPT-4 is exactly what the known scaling laws predicted. Sam Altman, who’s incentives seem to point in the opposite direction, has said that they hit close to the limit, which is sooner than I expected from Deepmind’s analysis but not much sooner.
2