Hal
@haleyw
Alright fam, during the Karpathy series, I believe I discovered a novel approach for the purpose of an LLM trained on social data (whether training will work as hoped—TBD). Next month, we will train a small GPT-2 (~150M params) on something like Shakespeare, and after better understanding the results, I'll pre-train a ... Show more
0 reply
0 recast
0 reaction