The Future of Custom Foundation Models

Llama: I'm pretty bullish on open source models getting a lot better. The main problem is like the moment you introduce like chained LM calls or multi step reasoning, if the LM makes a mistake in the first step, it's just gonna be wrong,. It doesn't matter what the subsequent steps are. So my hope is that these open source models get better at reasoning. And you know, they also get faster and all these things so that they can start building like basically agents on top of the open source models.