Ben Waber

Me: Look at all these fundamental issues with LLMs. Here's an example of someone losing their job from using them where the truth matters

AI Bro: I was worried about that, but I checked the answer one time and it was right and I made sure in the prompt to tell the LLM to only tell me the truth

Feb 16 at 4:46pmWeb
Show threadJonathan YuFeb 16
@bwaber I think it would be useful in general for more people to learn the intuition of how models work, both training and inference, because it is helpful to understand their limitations. They are very useful for a great many tasks, but as with other technologies, including the humble checklist, they are not an invitation to outsource thinking...
Show threadBen WaberFeb 16
@jawnsy I think with the way commercial tools are currently constructed the vast majority of people can't use them appropriately. LLMs could be more useful if incorporated into novel interfaces that weren't just text boxes. The small number of people who correctly use them today in areas where truth matters are highly technical specialists (scientists for data labeling, programmers to make mock ups or small, unit testable functions)