My guess? The main limit of their current approach compared with gpt4 is getting enough high quality fine-tuning data, and they announced basically as soon as they had results that looked good…

Posters: xlr8harder. xlr8harder

2 likes recorded by the source corpus.