Scaling compute can only be taken so far. True advancement in AI comes from optimizing compression of human knowledge. Altman missed the first principle by a mile.
Whereas, Marvin Minsky promised a PhD from an AI at the end of a decade, he would have done it again, but he passed away. He did this in 1950, 1960, 1970, 1980, 1990, he forgot about it in 2000, but made the claim again in 2010.
Ironically it's largely Minsky's fault that it hasn't happened. Too many people listened to him.
It probably will happen by the end of this decade, or not too long afterward. Within the next few generations, a CoT mechanism like R1's (and presumably o1's) will be capable of originating, directing, and documenting original scientific work, doing so with no more guidance than a typical PhD candidate would receive from their advisor.
ForOldHack|1 year ago
Did you miss this principal?
CamperBob2|1 year ago
It probably will happen by the end of this decade, or not too long afterward. Within the next few generations, a CoT mechanism like R1's (and presumably o1's) will be capable of originating, directing, and documenting original scientific work, doing so with no more guidance than a typical PhD candidate would receive from their advisor.