Meta’s Open Source Llama Upsets the AI Horse Race

Disappointing to see Wired uncritically repeat Meta's deceptive marketing. Llama 2 is not open source because it does not allow unrestricted use. In addition to the "Acceptable Use Policy" and the "700 million users" restriction mentioned in the article, Meta also does not allow any user to use Llama 2 to improve other LLMs, which is another usage restriction that prevents Llama 2 from being open source:

> v. You will not use the Llama Materials or any output or results of the Llama Materials to improve any other large language model (excluding Llama 2 or derivative works thereof).

https://ai.meta.com/llama/license/

On the other hand, Ars Technica correctly saw through Meta's deceptive marketing to discern that, despite Meta branding Llama 2 as "open source", Llama 2 is only source-available: https://arstechnica.com/information-technology/2023/07/meta-...

I don't know if they mention Code Llama but that has been a huge milestone. But from what I've heard it's still at least 15% behind GPT-4 for code. But open models are gaining ground.

Although ChatGPT Enterprise kind of changes the benchmark again in away.

I think this article is overstating things a bit. Chatgpt already lets you finetune the AI for your own use, so it's not like releasing LLaMa-2 as open source is going to seriously change things by a whole lot.