I’ve been thinking a lot about the gulf between people who think LLMs can cut 90% of the cost of software development or whatever and people who think that’s a pile of catastrophes balanced on top of a nonsense wish.
I think a lot of that chasm boils down to the fact that software is already •terrible•, and people have very different feelings about that. I don't just mean “Liquid Glass is ugly” terrible; I mean “it’s broken and insecure and unreliable and just doesn't flipping WORK” terrible, like Arbiter Sports.
We are right now in a war between people who think software should treat humans humanely and people who are perfectly willing to offload their development costs onto captive users in the form of wasted time, anger, anxiety, and even more severe forms of personal harm. That war was raging well before LLMs entered the picture, but LLMs drove it to fever pitch. I’m extremely skeptical of the idea that LLMs make any significant positive contribution to the long-term speed or cost of writing •good• software, but they clearly have the potential to lower the floor for •bad• software to unheard-of levels at unheard-of speeds. How you feel about that fact probably predicts a lot about how you feel about the “LLMs can 10x software dev!” hype.
3/3