Discussion about this post

User's avatar
Tig Tillinghast's avatar

Of the two major costs to producing "AI" engines, only one has been getting much attention. The compute power, and associated energy costs, is well-documented. But real humans need to judge and code responses that the computer produces to "train" the AI to be, well, more human. This involves thousands of people, typically in remote teams working from computers at home. And it involves many millions of dollars in labor costs.

Neuroscientists at Dartmouth hope to automate this too. They can't just ask people "A penny for your thoughts?" and try to get the computer to match what a person says. Science has long known that people’s interpretations of normative behavior (what should or shouldn’t happen) are signaled via m-waves almost instantly, even before the people themselves know their own thoughts.

The team found that developed a brain signal aerial that harvests cerebellum signals before they develop into cogent thoughts. These are then used to code AI engine responses at a speed faster than thought.

“Instead of asking them to think and code their responses, we have something better and faster” said project director Slim Vitae: “Antennae for your oughts."

Expand full comment
8 more comments...

No posts