| > I have learned everything from flawed sources. We all have. Sure, some historical books in the US South are wildly inaccurate! This is why we deal with multiple sources, as well as defining, validating, and exploring our sources. ChatGPT does not do this. It does not pull from definable sources, it is an LLM that outputs text based on statistical word sequences from an almost bottomless source. You can learn from it, but the sources may be sketchy, and it will mix source biases. I tried to ask it for sources on the fall of ancient Rome: >> As an AI language model, I don't have the ability to consult sources or remember my previous responses. However, the information I provided about the fall of Rome is widely accepted in historical scholarship and can be found in many reputable sources such as encyclopedias, textbooks, and academic articles. Personally, I think we need to tune the LLMs according to use. Creative? Then it one way. More factual, the other way. This does mean we would need to know sources for LLMs. |