I honestly don't care if it's really a leak from inside Google or not: I think the analysis stands on its own. It's a genuinely insightful summary of the last few months of activity in open source models, and makes a very compelling argument as to the strategic impact those will have on the incumbent LLM providers.
I don't think it's a leak though, purely because I have trouble imagining anyone writing something this good and deciding NOT to take credit for the analysis themselves.
Yes this is an amazing summary! Just for its summary alone, it is probably one of the top five writings I saw on LLMs and I read every one!
> because I have trouble imagining anyone writing something this good and deciding NOT to take credit for the analysis themselves.
Not everyone has a substack that they spam onto hacker news every time a thought enter their head. Or imagine that INTP exist. In my opinion the best take ever on LLMs is the simulators essay https://generative.ink/posts/simulators/ and the author is so shy they went pseudonymous and put their twitter as private and don't even want their name in conferences.
For example the simulators paper was written before ChatGPT was even released, based on research with the GPT-3 base model that only had text completion and no instruction tuning or any kind of RLHF or lobotomization. In another example, in the interviews with the people who had access to the base model of GPT-4 like the red teamers and the ones at microsoft who integrated it with bing, they consistently explain that the raw base pretrained model has the most raw intelligence which is deadened as they put RLHF and guardrails onto it.
1: https://jblive.wufoo.com/cabinet/af096271-d358-4a25-aedf-e56...