The future of artificial intelligence, we are told, hinges on making it more flexible. Not faster. Not more accurate. Flexible. As if the machines have been sitting at their desks for too long and need to touch their toes.
Yan LeCun, one of the people who actually built modern AI, has apparently concluded that raw intelligence is overrated. His new venture is focused on developing AI systems that bend and adapt rather than simply knowing things better. It is a refreshing admission: we made them smart, and they still cannot figure out what we actually want. So now we are teaching them to improvise.
This is where the logic gets delicious. We spent twenty years optimizing AI to do one thing extremely well — recognize patterns, predict text, classify images. Turns out humans do not work that way. Humans are catastrophically flexible. We can learn to cook from a video, adjust our strategy mid-conversation, and handle situations we have never seen before. We are basically professional improvisers having a very long career.
So the next frontier is not smarter AI. It is AI that can figure out what to do when the problem changes. AI that does not need a PhD in prompt engineering to be useful. AI that works when you throw something unexpected at it instead of politely returning an error message.
The absurdity is not in the goal. It is in how we got here. We built these systems to be narrow specialists and are now shocked they do not generalize. It is like training a chess grandmaster and then complaining they cannot cook. The solution is not yoga for machines. It is finally admitting we were solving the wrong problem.