4 ms·
> I also just don't find asking a LLM in the "correct" way and having it spit out code trained on millions of other programmers interesting in the slightest. T
by zamfi 4y ago
> I also just don't find asking a LLM in the "correct" way and having it spit out code trained on millions of other programmers interesting in the slightest.
Think broader.
Right now our only interface to these powerful models is prompts, but this is an artifact of how we discovered them.
Future you will not be (paraphrasing) "finding the right prompt to have an LLM spit out the right code" nor "translating client requirements to prompts".
Instead you'll be letting an LLM fill in the basic details of code, syntax, systems, and propose algorithmic approaches that you pick and choose from, while you think bigger and write programs with with substantially more complexity and capability.
How this could become a less-remunerative activity I have no idea!
- radlad 4y ago> Instead [...] you think bigger and write programs with with substantially more complexity and capability. I fear that this is true, but that the complexity will be accidental rather than incidental. Humans are capable of understanding architecture and design patterns - as far as I've seen so far, Copilot and the like do not.
- zamfi 4y agoI think...that's my point? You program at the level of architecture and design patterns, and Copilot and the like fill in the details. Do you need to keep a watchful eye on the details? Of course. The complexity will either be necessary for the great things we build, or unnecessary but time-saving, or just bad all around -- but that's true as software grows independent of LLMs.
- ShouldIPivot 4y agoYeah, I use copilot quite extensively to help my RSI (though I always know what the expected output is, it doesn't solve problems for me) and that doesn't ruin my enjoyment. If what you describe is how programming ends up, I think I'd enjoy it. (And my wrists will appreciate it :) )