Not today, but eventually: manton.org

@manton Good post. I work in a bit of a niche area writing software tools for the IBM i OS, working in languages that are specific to that OS – ILE RPG and CL.

I also write in Python, Ruby, Javascript, and Smalltalk. For those languages I regularly generate code using AI but I always inspect it. Much of the time the code sort of works. The more specific the request and the smaller the scope, the better the quality.

The code I see gen’d for RPG and CL is dicey, since the learning base is relatively small. But I’m under no illusions here. It will get better. Will I be out of a job? Probably not, since I still have to decide what to implement. Customer requests are always specific. I need to generalize them to make them applicable across a larger customer base. There is where my 45+ years of experience on this platform comes into play.

@ddykstal Thanks. Yep, there is so much of the work that requires humans in the loop.

@manton I found myself agreeing with a lot of what you write here. Despite what I have said about LLMs over the past few years, I agree they will be useful and will fundamentally change jobs like programming. I would even go further and claim that what is currently billed as AI fits on a continuum that spans decades and has been changing our relationship with computers all along.

But this line is laughably wrong:

> In a world where machines are smarter than we are, what should we work on?

AI is not smarter than we are. It is not smart. Not in the least. Useful, yes, but the smarts are coming from other humans who encoded those thoughts into text. Any intelligence detected in AI is an illusion generated by finding and repeating patterns it has seen in its training data. That’s just how it works. And it’s nothing like the intelligence embodied in living creatures of all species around the world.

Will LLMs change programming? Yes. It already has. Will LLMs change software engineering (which I define as using computers to solve problems)? Not as much as everyone thinks.

@fgtech Interesting! I agree it’s an illusion for now. But it’s so convincing, it feels real. There’s probably a better word than “smarter”… Maybe just better at its job than I am.

@manton The concept of “smart” is, in itself, very problematic and ill-defined. But that is a whole other discussion!

@manton I think “better at its job” is closer to what I’m thinking. Maybe it could be argued that we only made it the job of humans to program computers because we hadn’t yet figured out how to get them to use our language. LLMs are well suited to help us translate human language into machine code.

@fgtech Yes, this is one of the things I find fascinating, what might be possible when programming is accessible to many more people.

@manton I think the divisiveness you mention is driven by the people selling AI, representing it being able to do tasks it cannot, and forcing it into products it does not need to be in. LLMs can be wonderful at focused use cases (like code generation). But the people selling AI are not interested in limitations and guardrails, which is how you get Google telling students to use Chrome’s AI mode for homework, and OpenAI featuring a chatbot trained on incel forums. They are burying the utility of these products by throwing everything at the wall to see what sticks.

@pilch Yeah, AI is currently being stuffed into everything, and some of those things will fall away.

@manton I’m curious, how’d you add the narration to your post? I really like the idea behind this.

@SteveSawczyn Thanks! There’s a little more about it on this help page. Basically you have to (for now) record the audio yourself, then upload it and insert into the post.

@manton Wow this is fantastic. How does it know if the audio file appears to be a narrated version of the post, presumably by the transcript matching the post content? Really love this feature and hoping to use it soon.