Speaker unverified: belief

Transcript context

…It feels like there’s a fundamental difference between raw LLM-generated data and LLM-generated data with a human in the loop that does some kind of verification, even if that verification is a small percentage- … of the lines of code. I think this goes with anything where people think, “Oh, yeah. I can just use an LLM to learn about XYZ,” which is true. You can, but there might be a person who is an expert who might have used an LLM to write specific code. There is this human work that went into it to make it nice and throwing out the not-so-nice parts to pre-digest it for you, and that saves you time. And I think that’s the value-add where you have someone filtering things or even using the LLMs correctly. I think this is still labor that you get for free. For example, when you read a Substack article. I could maybe ask an LLM to give me opinions on that, but I wouldn’t even know what to ask. And I think there is still value in reading that article compared to me going to the LLM because you are the expert. You select what knowledge is actually spot on and should be included, and you give me this executive summary. This is a huge value-add because now I don’t have to waste three to five hours to go through this myself and maybe get some incorrect information. And so I think that’s also where the future still is for writers, even though there are LLMs that can save you time. It’s kind of fascinating to actually watch—and I’m sure you guys do this, but for me to look at the difference between a summary and the original content. Even if it’s a page-long summary of page-long content, it’s interesting to see how the LLM-based summary takes the edge off. What is the signal it removes from the thing?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.