Raw LLM Responses

Oddly, Meta's been releasing tons of open source models that have performed quite well. They're under the name LLaMa. The most recent Code LLaMa 70B has outperformed gpt4 on benchmarks. It seems like they're making the models open source to undercut proprietary models and are hoping that they can make up for with having tons of personalized data that makes the technology have value to each person they have data on, rather than the people have to try and figure out how to use the models to make it valuable to themselves. Google has some data, too. OpenAI has none. Microsoft has data, but it's largely business data, and I'm not sure how much they're actually sharing with OpenAI.