Impressive work! Apologies if this is discussed elsewhere, but it seems useful to think about if the advanced models were released, what % of individual workers they could automate. For instance, in September 2026, there would be 50,000 reliable agents thinking at fifteen times human speed, so if they work five times as many hours as humans, this could displace ~4 million jobs. This is a very small % of jobs. Now if we go to October 2027, we are at 330,000 superhuman AI researchers thinking at fifty-seven times human speed, so if we ignore the superhuman part and assume they are general, we could be at about 90 million jobs displaced. At that growth rate, the intellectual jobs could all be automated sometime in 2028.
Currently, the value of AI services is much larger than what people are paying for them. However, if the AI could automate nearly all job types and yet the compute were insufficient, the price of AI services would skyrocket (to roughly the value being created). This would mean far more revenue for the AI companies than one would expect based on the current fraction of value captured.
Impressive work! Apologies if this is discussed elsewhere, but it seems useful to think about if the advanced models were released, what % of individual workers they could automate. For instance, in September 2026, there would be 50,000 reliable agents thinking at fifteen times human speed, so if they work five times as many hours as humans, this could displace ~4 million jobs. This is a very small % of jobs. Now if we go to October 2027, we are at 330,000 superhuman AI researchers thinking at fifty-seven times human speed, so if we ignore the superhuman part and assume they are general, we could be at about 90 million jobs displaced. At that growth rate, the intellectual jobs could all be automated sometime in 2028.
Currently, the value of AI services is much larger than what people are paying for them. However, if the AI could automate nearly all job types and yet the compute were insufficient, the price of AI services would skyrocket (to roughly the value being created). This would mean far more revenue for the AI companies than one would expect based on the current fraction of value captured.