Later on, out of curiosity, I did try a 3/32″ pilot hole, on a separate scrap piece. I did see the phenomenon that ChatGPT and Claude were warning about, where the surface of the material mushroomed up and ruined the surface finish. A 1/8″ pilot hole seemed to work fine, but I didn’t conduct a rigorous stress test to determine whether it was significantly weaker than the 7/64″ hole. It’s entirely possible that I could have used a 1/8″ bit and it would have worked fine.
Yes, my usage of 6 LLMs for this was gratuitous. It was less about getting practical advice and more about seeing how different LLMs would handle a question for which there wasn’t a single “correct” answer based on internet sources, and how they’d respond to feedback from the user about other approaches. I learned that ChatGPT and Gemini tend to fixate more and push back against the user, while Claude is less committed to its answer. ChatGPT, Gemini and Claude all seemed to have a similar level of detail while the quality of the answers provided by Meta AI, Deepseek and Kimi was a step below.
Later on, out of curiosity, I did try a 3/32″ pilot hole, on a separate scrap piece. I did see the phenomenon that ChatGPT and Claude were warning about, where the surface of the material mushroomed up and ruined the surface finish. A 1/8″ pilot hole seemed to work fine, but I didn’t conduct a rigorous stress test to determine whether it was significantly weaker than the 7/64″ hole. It’s entirely possible that I could have used a 1/8″ bit and it would have worked fine.
Yes, my usage of 6 LLMs for this was gratuitous. It was less about getting practical advice and more about seeing how different LLMs would handle a question for which there wasn’t a single “correct” answer based on internet sources, and how they’d respond to feedback from the user about other approaches. I learned that ChatGPT and Gemini tend to fixate more and push back against the user, while Claude is less committed to its answer. ChatGPT, Gemini and Claude all seemed to have a similar level of detail while the quality of the answers provided by Meta AI, Deepseek and Kimi was a step below.