h3m3
@h3m3.bsky.social
READ:
>———————————————<
| https://www.quantamagazine.org |
| https://www.currentaffairs.org |
>———————————————<
>———————————————<
| https://www.quantamagazine.org |
| https://www.currentaffairs.org |
>———————————————<
September 7, 2025 at 6:47 PM
I don’t know where this comes from but Nathan Lambert (Allen Institute of AI) believes ~35T is only for the »frontier models« but in general they trained each version on roughly 2x the amount of tokens then 2.5
I guess we will have to wait for the report
open.substack.com/pub/robotic/...
I guess we will have to wait for the report
open.substack.com/pub/robotic/...
May 3, 2025 at 9:45 PM
I don’t know where this comes from but Nathan Lambert (Allen Institute of AI) believes ~35T is only for the »frontier models« but in general they trained each version on roughly 2x the amount of tokens then 2.5
I guess we will have to wait for the report
open.substack.com/pub/robotic/...
I guess we will have to wait for the report
open.substack.com/pub/robotic/...