Re: AI sceptic/conservative
> No-one can read and digest the sum total of all human knowledge, but LLMs more or less already have.
Well, everything that has been easily and cheaply (well, freely, in all senses) available online, which leaves many gaps. However, skipping over that:
> And yet, if you don't know the topic you're asking an LLM about, it will be very convincing.
> I usually ask them about ... because ... I'm sufficiently fluent in both to spot when the answers are wrong. You have to check everything they write before you rely on it's accuracy.
So you can *only* sensibly use them in areas in which you *already* have the expertise; so if you want anything useful about any other topic, you first have to teach yourself enough about that new topic to reach the point you can answer the question yourself. OR you go the other way, take the answer they provide and do all the learning required to validate (or reject) that answer. Same result - you are now learned in that topic - although perhaps only in a tiny slice of it (which then means you missed the errors of omission and accepted a wrong result after all).
So then:
> No-one can read and digest the sum total of all human knowledge
True (although some act as if they believe they have)
> but LLMs more or less already have.
Which means nothing in practical terms as, from you own analysis, in order to have accurate results you are still restricted to working in the areas you have read and inwardly digested.
> Avail yourself of this new tool or you'll become like the 19th century knocker-ups.
Knockers-up (note where the 's' goes, btw) were replaced only when timepieces became demonstrably reliable and provably economically sound enough that alarm clocks could be found in a significantly large enough number of households that they could be relied upon to get enough workers into the yard for the company to operate. Including taking account that your knocker was intelligent enough to bang again if you dozed off on the stairs, unlike the alarm clock (if the wife paid the extra penny to make it worth the knocker's time to hang around).
SO, IF (or when) you can demonstrate that the LLMs have become as reliable as an alarm clock (which your own statements indicate they are not) *and* that they are economically sound (so won't just have the plug pulled when the VC money runs out and the banks call in the deficit loans) THEN you can draw that comparison.
BTW, knockers also had their job opportunities reduced when the big yards and factories themselves went away and the street no longer had every man working to the same shifts; if don't have enough clients it isn't worth walking down the whole street.
I'd like to equate that to the way that use of LLMs is being pulled into job descriptions by the force of FOMO more than realism, which *IS* a reason that individuals ought to avail themselves of the LLMs, not because the AI is a tool, but because the HR is a tool. However, my analogy has run aground on that last piece of wordplay.
