3 ms·
> Engineers are employed to translate human needs into tangible system designs, I mean, isn't this with achievable prompt refinement? One could argue that it
by spoiler 4y ago
> Engineers are employed to translate human needs into tangible system designs,
I mean, isn't this with achievable prompt refinement?
One could argue that it wouldn't satisfy some $arbitrary_dimension (scalability, accessibility, whatever), but you can imagine it wouldn't be hard to get there with prompt refinement eventually, especially on an AI trained to satisfy SWE requirement (and not talking about GH's Copilot,which is an early application of a fairly young technology)
- spoiler 4y agoThen again, it also seems we perform worse when using AI tools, so maybe I'm wrong https://arxiv.org/abs/2211.03622 https://arxiv.org/abs/2211.03622
- polygamous_bat 4y agoRight now, how DALL-E works is that you enter a prompt, it gives you a few images, you fine tune your prompt, and after a few rounds you may pick an image and maybe polish it before you can use it. What would the co-pilot equivalent be? You give it a prompt, it generates a few repos, but then you still have to understand the whole project to judge that it respects the system needs and restrictions. After a few rounds, maybe you will have to change up the code a little, but 90% code will be written by the machine. I, for one, welcome this future -- because once again, this "prompt refinement" iteration is where our human creativity is used most effectively; I think it's a significant improvement over writing boilerplate code every time.