Correct me if I misunderstand your implication, but it seems to be "I have seen poorly-written articles with obvious LLM tells and factual mistakes, therefore LLMs are incapable of writing good articles". It's the same logic people use to say things like "criminals are so stupid, just look at these prisoners", not realizing that their sample selects for incompetence.

There's no incentive whatsoever for any institution to credit an LLM for an excellent article, and in fact there are many incentives to avoid disclosure of any AI use. Typing "write me an article about $TOPIC, make no mistakes" into ChatGPT yields quite different results than the LLM-driven content generation pipelines major institutions have already been using for well over a year now.

The average engineer is... average. I have watched the frontier models raise the bar for hiring quite high over the last year, to the point where it becomes a real question whether most companies will need more than one strong employee per job function in the next few years.

The way processes are shifting, the companies with the most staff are the most likely to be outcompeted by smaller teams leveraging agentic development. I suspect we haven't seen anything close to the worst of layoffs at large corporations yet. The limiting factor is that it takes a while for a massive ship to turn, but inertia will only go so far.

As for what humans can do better than AI, there's one thing: assume liability. Lately, however, it seems that no one with the power to do so is particularly concerned about holding anyone accountable for AI-driven mistakes.

I wish I could justify your optimism, as I expect life to deteriorate materially for the typical member of the working class in countries without strong safety nets. But I just don't see it leading to more hiring when even the current frontier models are more capable across most dimensions than the average white-collar worker.

When a bug creates damage, the user is responsible. But like at most companies pre-AI, it is not prudent to focus too much on the individual. There should be a post-mortem that determines which processes were weak, and these days that can be fed directly into a harness for the LLM to improve your tooling or testing to avoid that class of problem.

Some of the most miserable people I've known have been the hardest workers. One example: a friend of mine tried to start a freelance career, but had little experience in the field and rarely reflected on what his clients' needs were. He would work very hard to satisfy his own curiosity about their problems, eschewing deadlines and leaving little time to review his work before delivery. He was among the hardest-working people I've met, yet one of the least accomplished. No one would keep working with him for more than a month.

Correctly identifying which work needs to be done and how is more important than actually working hard. Sisyphean work is not virtuous, no matter how much effort is put forth.

I agree that there are ethical problems around helping people twist facts for their own purposes. My point is that everyone in a courtroom except the judge and jury are there to do exactly that. They make the strongest possible case to get their desired outcome from the available facts.