33TB of text data for a 1T-parameter model

Re:33TB of text data for a 1T-parameter model by est31 on Sunday April 16, 2023 @12:27AM (#35587279)

There are a few innate reflexes but you lose them quickly and they are pretty basic. Our brains have way way more neurons/parameters than these models and are still not overfitting... the issue lies in the learning method.

[ Parent | on HN ]

Re:33TB of text data for a 1T-parameter model by og_kalu on Sunday April 16, 2023 @02:56AM (#35587880)

It's not just about reflexes though I would hardly classify reflexes as "lost quickly". That's just not true.

It's deeper than that. Our brains are optimized for a lot general human functions. Learning language is one of them. There's a whole section of the brain dedicated to it and other things.

There's also a lot of vital biological information encoded in DNA/RNA. We are not even close to starting from scratch.

[ Parent | on HN ]

Re:33TB of text data for a 1T-parameter model by est31 on Sunday April 16, 2023 @03:11AM (#35587941)

I was talking about this stuff: https://en.wikipedia.org/wiki/Primitive_reflexes

[ Parent | on HN ]

Re:33TB of text data for a 1T-parameter model by MVissers on Monday April 17, 2023 @07:56AM (#35599449)

We actually still have them as adults, but are suppressed by our brain. With brain damage or spinal cord injury, a lot of these reflexes return.

[ Parent | on HN ]