For most of the issues that human evaluators catch, I can conceive of technical solutions, except for the problem of factual errors. Solving the problem of fact

For most of the issues that human evaluators catch, I can conceive of technical solutions, except for the problem of factual errors. Solving the problem of factuality requires a sufficient model of the world, which is possible only in very restricted domains. I'm afraid the end result of LLM development will be an extremely convincing purveyor of misinformation. There is some discussion of LLMs and models in another thread. (https://news.ycombinator.com/item?id=44625629 https://news.ycombinator.com/item?id=44625629)