It seems to me that just maybe by not lobotomizing the training in an attempt to keep it pg, they’ve unlocked a lot of performance. This probably isnt the only

It seems to me that just maybe by not lobotomizing the training in an attempt to keep it pg, they’ve unlocked a lot of performance. This probably isnt the only thing of course but it is a major difference between deepseek and other models