9,258 karma · joined August 18, 2007
I find LLM learning to be mixed. I can ask questions, seek clarification, and that helps me get to a specific answer quickly, or helps me to get past misconceptions quickly. But it seems to fall somewhere in between reading books and watching videos for me -- I still feel like I learn best through books, even if it takes longer. Specifically, it feels like actually being a little bit harder forces me to think deeper and/or retain more.
I do not wish for LLM learning to go away, but nor do I wish for it to replace books. I hope that many people continue to write in traditional formats.
I wonder if there’s any parallel to that in software?
So those musicians are no longer getting booked for that bit of music. Instead, one person produces it in their home studio. But, there’s now an industry for creating software tools that support that workflow, and there are a lot more opportunities for such music than there used to be. The amount of music used in background spots on television is astounding.
Things changed. Some jobs diminished (studio players?) or went away altogether (music copyists?). But new work came into existence.
The fact that an LLM works great for one user on one project does not mean it will work equally great for another user on a different project. It might! It might work better. It might work worse.
And both users might be using the tool equally well, with equal skill, insofar as their part goes.
If that is what he meant, I presume this remark was written well in the past, as TeX has lasted way more than 12 years.
And I find I still talk to people personally about such things, along with researching online sources.
https://news.ycombinator.com/item?id=46543516
presented a reasonable approach for using AI coding tools but still staying engaged with the code.
I am wary of hanging over the reigns entirely to AI, not just on a one-off project basis, but long-term. I don't think it's wise to rely so heavily on proprietary cloud-based applications as software development tools. Keeping one's hand on the wheel enough -- maybe like as that article describes -- should hopefully leave one still able to take over control entirely if needed.
Of course, who's to say what your employer might demand? It sounds like some people are insistent upon producing code at maximum possible speed, such that if you are still even reviewing code, much less writing some of it by hand, you are wasting time. I don't care for or agree with that notion, but it's not my place to tell someone else what to do.
This has been a while; perhaps the latest frameworks account for all of that better than they used to. But at that time, I could absolutely see budgeting several days to do what seems like a few hours of work, because of all of the testing and revision.
I imagine some folks would still say that the code itself doesn't matter, so you're wasting time even thinking about it, and you could ship faster if you didn't look at the code. I imagine this is true, and if your goal (or your manager's goal) is as much code as quickly as possible, then perhaps that's the path you have to go. I personally do not feel comfortable shipping code that I have not reviewed and understood, much less not even looked at.
I'm not entirely convinced that the legal questions can be so easily swept away, but I suppose we will find out more in time. If there are legitimate legal problems, they will be found out, I trust.
I also remain concerned about the proprietary SaaS nature of all of this. I am surprised that, given how much programming culture has historically valued having free / open source tools that could be completely controlled by the users, that handing over so much of the software development process to closed-source cloud computing is so agreeable. Hopefully local (and open source) models will progress enough that this ceases to be an issue, but I suspect that the proprietary industry leaders may remain the more popular choice. We'll see.
I think the best thing to do there is, don't forget how to do it yourself. If you want to benefit from the convenience, then wonderful, but should that convenience ever get shut off, you don't want to be stuck having no idea at all what you're doing. Which, leveraging AI coding tools the way this article recommends, you should stay in a good place there.
I don't want AI involved in my laundry machines. The only possible exception I could see would be some sort of emergency-off system, but I don't think that even needs to be "AI". But I don't want AI determining when my laundry is adequately washed or dried; I know what I'm doing, and I neither need nor want help from AI.
I don't want AI involved in my cooking. Admittedly, I have asked ChatGPT for some cooking information (sometimes easier than finding it on slop-and-ad-ridden Google), but I don't want AI in the oven or in the refrigerator or in the stove.
I don't want AI controlling my thermostat. I don't want AI controlling my water heater. I don't want AI controlling my garage door. I don't want AI balancing my checkbook.
I am totally fine with involving computers and technology in these things, but I don't want it to be "AI". I have way less trust in nondeterministic neural network systems than I do in basic well-tested sensors, microcontrollers, and tiny low-level C programs.
I think the new question here is, if the new status quo is to offload your creative thinking to LLMs, will now any engineers do anything truly novel?
If you're not engaging your mind to create and think on a day-to-day basis, will you be in position to have some new insight?
If you're wanting to, say, set up an online store website, and you don't know how to build a website at all, I think using something like Shopify is a better path than consulting an LLM and having it generate code for you.
On the other hand, if you do understand at least generally how to build what you want, you can leverage LLMs to build it faster and customized beyond what something like Shopify could offer.
So I reckon, for true non-developers, some no-code platforms will carry on. For developers, LLMs will probably replace some subset of such things.
You wrote the lyrics. There are professional songwriters with many hit songs who only write lyrics. Some can't even play an instrument, much less compose music. So what do they do? They work with a music composer. They hire a music arranger. They hire a band.
So in this case, you still did the foundational songwriting part yourself, but instead of hiring humans to help you finish it, you hired AI.
I don’t enjoy boilerplate. I don’t necessarily enjoy all of the error checking and polishing and minutia in turning algorithms into shippable products.
I find AI can be immensely helpful in making real things for people to use, but I still enjoy doing what I find fun by hand.
Serving customers who want non-MDF options can be a perfectly sustainable and profitable — even if smaller — business. There are plenty of small furniture shops doing just that.
Extrapolating these numbers, perhaps Employee B is so skillful that, if B did use $2000/month, B could get not 5 features done but 100 features.
That sounds great as an isolated metric, but does the user want 100 new features each month? Does marketing want to advertise 100 new features each month? That sounds like quite a bit.
Or what if B spent $4000/month and completed 200 features? Or spent $10,000/month and completed 500 features?
The software product pipeline has never really had to consider this question too much, because it has usually been the opposite problem: planning to ship 10 features in a month, and only completing 8. Well, if we can now complete 100 features in a month... is that good? Do we actually want to ship that?
Perhaps some discussions need to be had, but there probably should still be some reasonable limit on how many new features to put into a software product in a given timeframe. And then, sort out expected engineering productivity to match.
But to the initial core question: I would think a company / department / team might have some overall monthly budget for AI tool usage. If one person uses more than another, I don't know it really makes sense to give extra cash to those who used it less. Like, if some engineer needed a $2000 compiler tool for their project, and another used GCC for free for their project, does the GCC user get $2000? Not sure things traditionally worked that way.
If (say) the code generation technology of Anthropic is so good, why be in the business of selling access to AI systems? Why not instead conquer every other software industry overnight?
Have Claude churn out the best office application suite ever. Have Claude make the best operating system ever. Have Claude make the best photo editing software, music production software, 3D rendering software, DNA analysis software, banking software, etc.
Why be merely the best AI software company when you can be the best at all software everywhere for all time?
For example, it seems reasonably that using a good programming editor like Emacs or VI would offer a 2x (or more) productivity boost over using Notepad or Nano. Why hasn't Nano been banned, forbidden from professional use?