The Wrong Approach to Context Learning

The way to make transformers, you know, large scale transformers, actually, current is by efficient retrieval and in context learning. So why do you think that's the wrong approach? I don't think it's a wrong approach, right? It's a small, like, which one is more efficient? Like, you should want a small performance, right? But if we can lift certain constraints such as cost, for example, then we would be able to do more. Have you looked at something called vector additions where you sort of inject vectors into the middle of the feed forward flow)? And what do you think of it? Chip: "I've seen several approaches deal with like