Exciting work by Wu, Mitchel et al (arXiv >cs>arXiv: 2302.04449.

Deep Mind can learn Atari games, but it takes very many trials. By using an AI language model to read the manual for the game, Wu et al were able to demonstrate that the game could be. learned in many fewer trials.

The relevance is that it take many trials for a macaque monkey to learn a cognitive task for reward. But a person can do the task immediately by simply following the verbal instruction given by the experimenter.

@ekmiller

I saw a diagram you posted on twitter on attention. ( I no longer post on twitter because Musk is a racist). what do you think the inferior frontal junction is (IFJ in the diagram)? Is it part of the FEF, the dorsal part of area 44, or what?

@ekmiller

The demonstration of retrospective and prospective coding in WM in humans does not show that this is true of 9/46 and 46. In humans the activations are only in the FEF/8Ad. They are told the rule.

In monkeys there is also WM activity in 9/46 as well as 8 for remembering a single location (Riley and Constantinidis). But the monkeys have to learn the task rule, and this activity may reflect that.

You are never going to get rid of me. I will, for ever, be a millstone round your neck.