This is a pretty amazing video that demonstrates the state of the art of the leading LLM.
Wyatt Cheng is a Game Director at Blizzard Entertainment, so his results are based on what is visible in the video, as well as years of experience. Still, he was able to produce a working game without writing any code himself, relying on the Unity toolchain, ChatGPT-generated code, and his own experience and understanding.
As he says in the video, we're still not quite at the point were someone with no ability or experience can create a game using Chat-GPT, but we're very close. More to the point, I think this opens news doors for someone who "thinks like a programmer," but maybe lacks C# experience and has therefore felt intimidated by Unity.
Perhaps it's unsurprising that this seems to support my priors, but I continue to see these amazing LLMs more as productivity enhancers than as job destroyers. Not to say some jobs won't be destroyed! Enhancing the productivity of developers means that any large enough team now needs fewer developers. It's also more clear than ever to me that those who can communicate clearly and completely have an advantage in dealing with LLMs.
I think it's reasonable to be concerned about the developer pipeline when junior-level work can be accomplished almost as easily as it can be described. How will junior developers ever become senior developers if they can't get jobs doing junior development and building experience? I'm interested in how we will solve this challenge as an industry.
In the meantime, I'm enjoying having ChapGPT write shell scripts and simple Go functions for me.
@LouisIngenthron You've made a series of statements that I'm not sure are related to each other, and I agree with some of them.
Writing code *is* the easy part! I believe both Wyatt Cheng and I have said as much, both in this linked video and in my previous posts. I don't even think debugging is as hard as design, but okay, both debugging and design are harder than coding. So little of most developer jobs is actually writing code! The video demonstrates using ChatGPT (and Cheng's experience and understanding) for debugging, and I've used it for debugging as well. This gets back to what I said, which is those who communicate clearly and well can make better use of these tools than those who cannot. ChatGPT is faster than I am at spotting a certain class of coding mistakes, so I can give it a function and say "when I give it X, I expect Y, but instead I get Z," and it will tell me why I'm getting Z, and what I need to change to get Y instead.
You always opined on the quality of the jobs created, which... I didn't actually talk about creating jobs, so I'm not sure what you're talking about. I'm not even sure what it would look like for LLMs to create new developer jobs! I said I think fewer jobs are going to be destroyed than some are predicting, but less-severe destruction is not creation.
Finally, you said this is all very short-sighted. I'm assuming you mean "the entire concept of using LLMs to write code," but if that's now what you meant, let me know. It's definitely short-sighted in the Luddite sense, and I think the Luddites were absolutely correct! Machinery put skilled craftspeople out of work _en masse_, and ended an entire way of life for many. I mentioned my concerns about junior developer positions.
That said, right or not, Luddites are generally mocked today, as they are perceived as standing in the way of Progress™. In the moment, the Luddites were focused on what seemed like the longer view, while short-sighted factory owners were chasing profits and destroying livelihoods. But it turned out there was an even longer-term view, in which you and I can be dressed like royalty in ways unimaginable more than 200 years ago, and so can everybody else.
I don't know what the long term of skilled designers, directors, and developers having access to LLMs is going to look like, and I don't think anybody else does, either. I know that we'll all find out, because the advantages to those using LLMs today already outweigh the disadvantages both personally and as a group.
I suggest that there are short-term gains from LLMs, and medium-term losses, but almost certainly also long-term gains that dwarf both.