Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Cergorach" data-source="post: 9855673" data-attributes="member: 725">Do you do these things on a regular basis? If you do, you realize that you don't just write a summary for target audience xyz, it requires many, many revisions to get a concise but accurate approximation that's understandable for the target audience.Depending on the consistency of the output quality and the capabilities of the LLM, there might be a LOT of time saved.There are of course challenges, research in subject matter abc often can take weeks, months or years, chances are that not all the facts are present in your head (depending on the complexity of the material). Heck, I often need to check facts when I write technical stuff. This of course assumes that the model can actually work with your results without being trained on it (which costs a lot of time and money), and that you have written extensive documentation that it can reference, but not too extensive or it will also get confused.If all the requirements are met, it can save a LOT of time. The problem is meeting the requirements, I don't use LLMs in my work because a.) the customer doesn't have a LLM policy in place, b.) because I'm working on cutting edge (IT) stuff for application x and the LLM hasn't been properly trained on it, c.) it's an edge case that doesn't have much information available in the first place and the only way to get information is actual experimentation. Another thing is I often work on proof of concepts where there is no extensively written documentation, the point of a condensed report that's understandable for non-specialists is the whole point of the exercise that will get approved for further development, at which point extensive documentation will get written. In such a case you're essentially spanning the cart before the horse when you want to use LLM.Another issue is not the LLM, it's human nature. Not bothering to check the facts, either because they aren't qualified or  just can't be bothered to do so (or it looks believable). A good example of that is the Deloitte case, but there are, many, many such cases. And this is why I'm against many uses for LLM, the humans don't do their job...</blockquote>

Verification