Had a comment listing specific reasons why that post is wrong, but I just realized a generic reason. I see two broad spectra of opinion on AI. One is "LLMs are the greatest thing ever!!!11!1!eleventy! They will usher in the post-scarcity economy!" The other is "LLMs will eat the world! We're doomed! Doooooooooooooomed!!!" In my opinion, both are wrong, because both assume unbounded potential for LLMs that just isn't justified by the facts. Look at how much trouble you have getting good code out of an LLM, for example. The fact that you simply cannot rely on LLM output but have to carefully scrutinize it for mistakes isn't going to go away, because of the fundamental reality of what an LLM is. So why would you assume that LLMs would be reliable in other areas? This is a new-and-improved version of Gell-Mann Amnesia, replacing newspapers with LLMs: people know LLMs aren't reliable in the field they personally specialize in (I've seen, on a lawyer's blog, examples of other lawyers who submitted LLM-written court papers with hallucinated citations; Courts take a, shall we say,
dim view of such things). But then those same people go and assume that the LLMs are going to take over the world in other fields.
It's not always the case that given an opinion, both extremes are wrong. But in this specific case, it certainly is. Neither the "LLMs will usher in the post-scarcity economy" view nor the "LLMs will doom everyone to unemployment" view (which are remarkably closer to each other than it would appear at first glance) are correct. LLMs are a useful tool with inherent, fundamental limitations that mean that they will never be able to do everything. AGI is currently a pipe dream, and LLMs are not going to be the technology that achieves it.