AI in July: OpenAI caves to developers, RAG in focus, agents pick up steam, and Llama 2
A review of some of the zeitgeists in the AI community for the past month.
So much for the summer months typically being “slow news” months.
We saw a lot happening in AI—some major news events, like the release of Llama 2, and some more subtle trends emerging.
Because of the pace of development, I wanted to start providing a kind of “review” at the end of the month in a similar fashion to what I did with the half-year mark at the anniversary of the release of the Transformers paper.
In this issue here are a couple topics that I wanted to cover that come up a lot more often in conversations with sources and companies I talk to. They are:
Context windows and their utility: We’re seeing some research that is showing stuffing a context window full of information might not be that valuable. So other techniques are gaining some interest.
Useful agents pop up on the radar: Agents earlier this year were a hypothetical do-everything-for-you AI tool, which turned out to be kind of a novelty. But the past month or so has seen some more interesting developments in single-purpose agents.
OpenAI caves to developers: OpenAI extended its deprecation window for older models after “feedback” from some developers over its short-notice deprecation of its existing models.
Llama 2: Yes.
This is going to be kind of a mix of some of the bigger news events along with a lot of emerging topics and trends I am picking up from conversations. We’ll start, first, with what’s coming up in conversations more often lately.