4 ms·
That is a very thoughtful piece. Thank you for posting it. I especially like the idea of that new mode of programming (or problem solving, whatever you call it)
by antigremlin 11mo ago
That is a very thoughtful piece. Thank you for posting it. I especially like the idea of that new mode of programming (or problem solving, whatever you call it) when people very much enjoy getting an LLM to do what they want.
This is both new and old, because it's the same joy (or dopamine hit) of making a machine do your bidding. Honing your prompts is not that different to honing your shell scripts. I think many people overlook this aspect.
- int_19h 11mo agoI do a fair bit of "prompt honing" myself, but there remains a fundamental difference between that and traditional hacking of yore: the latter is predictable. The system may be very complex, but you can map it out step by step, and the more you do so the more advanced things you can cajole it into doing - and, importantly, those things don't suddenly stop working "just because". But with LLMs, your well-honed prompt might be an abject failure on a very similar task that just happens to have enough differences in context that the "magic" in the prompt no longer works. And debugging those kinds of issues basically amounts to throwing things at the wall and seeing what sticks. And then of course the providers switch models from under you and retire the old ones. So it's not quite the same joy of making a machine do your bidding. In the former case, the joy was from knowing that you can do this now and in the future - that you grok the thing enough to make it do this bidding at any point. With LLMs, it's more like the joy of successfully casting a spell, but you still have no idea why that particular spell worked better than a dozen others, nor can you rely on it working in the future.