These graphs shows two important things:
1.) LLMs make it easier for a wide range of people to produce higher-quality work more consistently.
2.) We need to rethink what we should be testing and evaluating in education.
Frontier models can ace every exam you’ve ever taken. So what use is there in continuing to evaluate students with those same questions?
What will it take to get an A today given that LLMs can produce B+ to A quality work in less than five seconds?
Here’s what I look for during interviews (where the candidate is free to use LLMs, even during the technical round):
imagination
social intelligence
persuasion, sales
judgment under uncertainty
discipline and follow-through
the ability to take responsibility for a decision
If you use AI to do your thinking for you and have no plan to backfill with the skills listed above, you’ll have a difficult time in 2027 when the new baseline for quality has been completely reset.