We have compute to train Llama 3 and 4. The plan is for 3 to be as good as GPT-4twitter.com2 points·modeless··0 commentsOpen articleSaveView on HN