Well, a lot of us find ChatGPT and other LLMs to be impressive, but the notion of them completely replacing human programmers is far-fetched. These tools can assist and augment our work, but I'm sure the need for skilled developers with deep domain expertise will remain. I've recently read about body leasing and team leasing https://seclgroup.com/body-leasing-and-team-leasing/ It seems these models can help address talent gaps. However, I guess they should not be viewed as a long-term solution.
Very true. And LLMs only build things that already exist out there, if there's something not easily searchable online, they tend to fail badly or produce code that is very slow.
>> The end of the programmer is near. Yeah, right. Like one of my current tasks is not even to introduce a new feature but "simply" refactor existing code by deleting an interface and porting the calls to another existing interface which does the same thing. Well it's a fucking nightmare. With 60+ implementations (think exchange gateways), each of them takes a subtly and not so subtly different approach, and changing that is anything but straightforward. A lot of understanding the context required which in no way can be inferred by non-generally intelligent entity and we're nowhere near that. And that's just maintenance of existing code. Don't get me started on implementing a new gateway based on exchange documentation. LLMs basically regurgitate existing implementations while every new development worth developing is just that ... new. Non-existent. One can only infer some vague properties from previous experience but overall the biological neuron gotta do it's work and figure things out.
LOL, looks like some guys "hired" AI to do their job. Things are changing so fast, you gotta wonder—are we going to start farming vegetables for a living now?
I agree! While ChatGPT and LLMs are great tools, they’re nowhere near replacing programmers. These AI models can assist with tasks like debugging or generating code snippets, but they still rely on human expertise to design, structure, and build complex systems. At best, they help automate repetitive tasks, allowing developers to focus on more creative and strategic work. It's about enhancing productivity, not replacing programmers. What are your thoughts on how AI is supporting dev work?
O1-preview is what really got to me. It wasn't released when you posted this. 10/22 Claude Sonnet already passed O1-preview. It is just preview though, the non-preview is coming. Then right now I can't imagine Sonnet 10/22 doing the same thing as what the o1 models are doing with reinforcement learning when they already beat o1-preview. I don't think it is a question of replacement. It is a question of can you leverage these technologies or not? I am using the Julia programming language now that I knew nothing about a month ago. That is already old though compared to the stuff I made with Cursor and Sonnet this week. Your either becoming smarter from LLMs or your not. If your not, it won't get better going forward. There is no such thing as "artificial" intelligence. It is just the same old intelligence. There is this huge opportunity though in understanding cybernetic relationships between human/machines and not falling into this science fiction AI trap. A good place to start is here https://en.wikipedia.org/wiki/Second-order_cybernetics Linus Torvalds gave a great talk on how LLMs are just another level of abstraction when it comes to programming.
Its going to replace programmers but it will not replace architects. Big difference. In many domains it is the same old story. AI beats human. Human plus AI beats AI. US employment of software developers. There is no reversal in sight.