@repligate i've read speculation that o1's training method comes at the expense of in context learning performance, which could explain some of this.

Posters: xlr8harder. xlr8harder

2 likes recorded by the source corpus.