@datarama @samir it’s similar to what happened in Chile. Most of the people that studied to become teachers after the dictatorship didn’t want to be teachers in the first place, but couldn’t get into other disciplines. A lot of the people that wanted to become teachers studied something else to have better working conditions and higher salaries.

The similarity with automation is the goal of making workers as cheap as possible

@datarama and we have a paper that explains why! (Although that’s not the point of the paper, but it sheds light on that direction) @samir have you read Ironies of automation?

@samir yeah, it’s a trend that has been going on for too many decades to count already. And it has many different shapes. Chile is a cautionary tale of your example, so it’s hard for me not to look at these things from a sociological and economical perspective.

This is why the «people that embraced LLMs for programming don’t like it» narrative makes sense to me. Keeping around the people that’s actually interested in a discipline is expensive, as it should

@samir exactly, just not in the way I expected

But I came to realise its value. I’m pretty sure we will have a similar shock here, unless we keep pushing back. In particular I’m worried about delegating intellectual activities

@samir we will see, it wouldn’t be the first time I’m wrong…

When I Eason colleague and blogs were becoming popular in Chile, I predicted the web would kill journalism because people would be writing blogs… I expected though the blogs to be reliable sources of information

@samir oh, yeah, they are impressive in how flexible they are, but I try to focus on what I know. I make no predictions on how these things could or couldn’t be used in disciplines I’m barely familiar.

So going back to programming, what I’m saying is that I believe they will plateau. People will get burned out (specially considering the burden of reading so much code as opposed to writing it yourself, as we have discussed before). And they will be yet another tool in your toolbox. Smaller models running locally may help to navigate code semantically («show me the function were I implemented this feature»). Prototyping will still be a thing. I’m in love with my technique of replacing TODOs with mostly functional boilerplate code. And so on an so forth. But I’m very sceptic of «you won’t need to read code» and «most code will be written by LLMs»

@samir

My current prediction is that after a while (and today I’m so optimistic that I think it will be sooner than later) we will be discussing these things just as we discuss debuggers. I’m primarily a pint gal, but some nasty bugs make me tolerate the slowness of the Xcode debugger

RE: https://mastodon.social/@samaaron/116193249451356736

Basically what I’ve been doing since the very beginning except instead of full prototypes I go feature by feature, and only if I’m struggling, otherwise I don’t even look at an LLM.

And only because I’m pressured to do so instead of asking a human