The Death of the Software Developer

I cannot currently afford enough RAM to run a ~3 trillion parameter model like Kimi K3. However, I do expect for the current trend for more expensive RAM will reverse and return back to the long-term trend of it being really cheap. Five years from now, I expect today's SOTA will probably be totally affordable to run locally (this is of course a terrible future to anyone who has recently invested in data centres).

Many asked the same question when loom operators were on the way out. But in hindsight would you have rather seen it play out the other way and now have loom operating being the thing that you do? Freeing one from operating a loom allowed us to move into software development. I, for one, am thankful I don't have to operate a loom.

What you may have missed is that many software developers are actually engineers. Engineers don't care about software, they care about solving problems. Even if LLMs can somehow solve all problems that can be solved with software, there will still be an endless list of more problems to solve that cannot be solved with software. Software is just a tool in the toolbox, not one's identity. So what if one of those tools is no longer needed?