The premise was my dream for so many years. In fact, from my days at high school until the LLMs arrived in all their vulgarity. Now I wonder if necromancy would also be a terrible thing to really have ?
AI is useful for me in writing programs that I specfify in detail. However, it shows no curiosity or capabilty to grasp the general objective. Also, it really sucks at debugging and writing modularly structured code. Still, I appreciate this idiot savant and use it daily.
It sounds like you’re arguing the journey is always more interesting than the destination. Handmade shoes beat industrial ones. Cooking from scratch beats frozen meals. Sure, for some people, some of the time. It depends entirely on what you’re optimizing for.
The bigger tell is that you’ve decided you’re the final judge of what “proper” LLM use looks like, while admitting you only ever get slop. That’s not an insight about AI. That’s an insight about how you, yourself, use AI.
A well-prompted frontier model doesn’t sound like the charlatan you describe. It sounds like a very fast, very well-read collaborator who still needs a human brain in the loop. If yours doesn’t, the problem isn’t the model.
This reminds me of the following joke:
- « Ah, I found out yesterday my dog could play chess. »
The premise was my dream for so many years. In fact, from my days at high school until the LLMs arrived in all their vulgarity. Now I wonder if necromancy would also be a terrible thing to really have ?
AI is useful for me in writing programs that I specfify in detail. However, it shows no curiosity or capabilty to grasp the general objective. Also, it really sucks at debugging and writing modularly structured code. Still, I appreciate this idiot savant and use it daily.
It sounds like you’re arguing the journey is always more interesting than the destination. Handmade shoes beat industrial ones. Cooking from scratch beats frozen meals. Sure, for some people, some of the time. It depends entirely on what you’re optimizing for.
The bigger tell is that you’ve decided you’re the final judge of what “proper” LLM use looks like, while admitting you only ever get slop. That’s not an insight about AI. That’s an insight about how you, yourself, use AI.
A well-prompted frontier model doesn’t sound like the charlatan you describe. It sounds like a very fast, very well-read collaborator who still needs a human brain in the loop. If yours doesn’t, the problem isn’t the model.
This reminds me of the following joke:
- « Ah, I found out yesterday my dog could play chess. »
-« What?!? That’s amazing!!! »
-« Meh, not really, I beat it most of the time. »