LLMs Will Break Things
I have this suspicion that the biggest thing no-one in the tech industry really seems to take seriously or give a shit about is the implication of many second-order effects of LLMs, but especially about collapsing the pipeline of developers from junior- to mid- to senior-level skillsets. Silicon Valley has a long history of caring about the short term: what can we get done in the next 18 months, will it get us through to our next round of funding, etc. at the expense of the long term. We’ve recently seen things like Jack Dorsey axing almost half of the workforce at Block because of LLMs.
It really concerns me that we are going to end up in a situation where the junior developer role is so depleted, as far as a dearth of positions, that we get into a scenario where the developer pipeline collapses and 5, 10, 15 years down the road we end up with code written by the currently-trained models that has errors in it. The errors won’t be fixable by our current models; they won’t be fixable by the then-current senior devs who never developed the proper coding chops in a normal sense. It’s easy for the tech industry to do what it has always done and kick the can down the road by relying on future advancements to correct their errors of today.
It really seems though like our current models have been built by a small number of companies burning through investment capital at an astounding, never-before-seen rate. Sure, there is revenue to be had from users and companies paying to use the model, but that is a pittance compared to what it takes to train them (not to mention the environmental toll, but that’s another topic). There’s a decent chance that will lead to investors pulling back for a more sound investment in the future. That will leave us with models that created code with issues they can’t fix.
Oh yeah, I created this microblog of futilely vain caterwauling using an LLM. Can I fix it if it breaks? Maybe.