www.lesswrong.com/posts/yaz8nx4ogZmiqHzt7/what-just-happened-pragmatism-and-pess...

The paper repeatedly says it is working with existing large language models and trying to make them more helpful, honest, and harmless.

In the abstract, it begins: "Given the broad capabilities of large language models, it should be possible to work towards a general-purpose, text-based assistant that is aligned with human values" and says the authors study "simple baseline techniques and evaluations, such as prompting." That is a description of improving the behavior of existing models, not merely creating a testbed.

The introduction is even more explicit: the authors say their goal is to "align general-purpose AI systems with human preferences and values" and that their efforts aim to "measure and address this general problem with large language models." They add that the societal impacts of current AI models should be taken seriously, and that training a large language model to be helpful, honest, and harmless would alleviate those impacts.

So while the paper also treats language assistants as a useful research setting, it is incorrect to say they were not trying to make existing AIs more HHH. That is exactly what the paper says it is doing.