I'd like to counter argue we need to shift the notion of efficiency.
While the nailgun is more efficient than the hammer, both need to be operated by a real human. Including all limitations of physical and mental health.
While we can improve the efficiency of the tools we should consider not to burn out humans to match the goal.
I had a similar experience with Hugo and the PaperMod theme. Such a large tool with all the dependencies to then just fail because of version mismatches with the theme.
In the end I sat down two weeks and wrote my own generator in python for my blog. Even wrote my own markdown parser for it. It's a sub optimal implementation for sure, but at least it's all my faults and I may fix them.
The following is the best I could collect quickly to provide backup to the statement. Unfortunally it's not the high quality first instance of raw statistics I would have liked.
But from what I have read time to time the crawler acted magnitudes outside of what could have been accepted as just badly configured.
For the future I will relabel "AI" as "Ain't Interested".
But despite the missing return of investment, it only needs a manager or a few, invested enough. They will push layoffs and restructuring regardless of better advice.
I agree to this oppinion.
But I furthermore want to add: It's not just about unsubscribing or ignoring the noise. In fact, unsubscribing or ignoring means to break with the premise sold, to begin with.
If it is true, that one must learn this or that to stay employed, he can not just "unsubscribe" from it.
Additionally the notion of productivity in our industry is problematic. While working with machines, somehow we developed their standard of productivity as our ideal. May it be the pressure of competition for companies and employees alike, but the current notion is not sustainable.
Exaggerated, but what I think: Turn away from two week sprints and work in a quaterly waterfall. Give developers a break, a constant plan and environment.
In my opinion the comparison to calculators is flawed. At least for simpler calculators as someone might imagine.
In its application that would mean for a LLM you define a sentence with one blank, restructure the sentence so the blank gets extracted and then enter it into the LLM to get exactly one word for your blank.
On the oppoisite you wouldn't use the calculator for solving equations, but to also formulate them based on a abstract definition.
Furthermore "math" and "text" are two fundamentaly distinct categories. In math we express mathematical observations/context, which can be proven in its correctnes. We can conclude that 1 Apple + 2 Apples won't result in 5 Apples, but 3.
Text may express emotions, toughts, directives, information or observations in context of its author. Not in every case it may be proven as math and we may attribute other values to it. Like honesty, authenticity, information and effort.
If it was only for the information, you could filter out hallucinations and call it a day. But for other values you literally can not outsource the work to a tool. Except you live in a South Park episode.
Evaluating other responses, people complain over Yandex, but asking for the very same experience. Only different in the illusion filtering happens to their wishes.
+1 I also wanted to point out, if there are questions about validation of the point made... just look at the post.
And from my perspective this should be common sense, and not a scientific paper. A LLM will allways be a statistical token auto completer, even if it identifies different.
It is pure insanity to put a human with a already harmed psyche in front of this device and trust in the best.
It's also insanity to pretend this is a matter of "trust". Any intervention is going to have some amount of harm and some amount of benefit, measured along many dimensions. A therapy dog is good at helping many people in many ways, but I wouldn't just bring them into the room and "trust in the best".
