Like some Australian teens are now successfully (!) using to sign up to social media.

Lets face it, you cannot keep kids out of any mainstream social activity humans do. As soon as they are interested, they will find a way in. Trying to prevent them will only cause harm and have zero benefits.

While this is a very fuzzy statement and requires qualification, it probably close to what is happening. Whether it will actually work out that way remains to be seen. As soon as that task requires some mental flexibility or high reliability, LLM-type AI is already out, even if people may not realize that.

I do agree, there are other effects at work. LLMs are far too incapable to have caused these effects all by themselves. They are clearly being used as pretext in many cases.

That is not quite true. Quite a bit of the world does run at least in part on compassion. The uncivilized part certainly does not.

I do agree on that one. I have, so far, found two instances where Internet-trained LLMs are useful: Search when you only can describe the issue or describe a concept and looking for cooking recipes by description. Both instances of "better search". For the effort invested, those results are utterly pathetic.

If it were AGI, it clearly had enough information. Since it is not AGI and cannot think at all, it did not have enough information and it would have needed the actual solution as input.

Indeed. But, you know, I am beginning to think that most people (outside of the about 10-15% independent thinkers and the additional 5% or so that can be convinced using rational arguments) are actually incapable of rational thinking or actively chose not to do it. People that do not understand the difference between an implication and a correlation. People think that people that do some thing makes them responsible for something entirely different with no causation chain present, but they have this fuzzy association. MAGAs that are keyword-operated and cannot do anything beyond reacting to simple keyword-triggers. And all the people that can only do yes/no and do not understand that most things are in degrees and shades of grey.

For these people, who can essentially only do unreliable statistical correlation instead of actual reasoning, an LLM may indeed look like it has insight, because they do not understand what insight actually is. And that LLM has a far larger "knowledge"-base.

Well, we can argue whether useful idiots share the blame or not. It is a complex question with no easy answer.

I fully agree that the "I got mine and fuck you" people are complicit and malicious.

Indeed. But there are tons of natural idiots that may be even dumber than "AI" of the LLM-variant and they are in awe because they do not understand what is going on. And many of them are aggressive.

Well, Musk is a half-educated idiot that got lucky and now thinks, like so many idiots with some luck, that his skills are superior.

Just not AI slop. Bad-quality things can often be replaced by cheaper to make other bad quality things. And nothing of value is lost.

Very temporary in a strategic sense. Obviously.

They may surprise you, but these are not insights the LLM had. These are insights other people had and the LLM essentially only found them by insightless statistical pattern matching.

To be fair, that strategy works on tons of people. Some will even aggressively oppose anybody that points out facts. Was nicely observable here as well.

I don't think so. LLM-type AI with cheap mainstream availability may well be a very temporary thing. First, they still do not have any business model that would even remotely justify the expenses. Second, LLMs cannot perform on professional level and cannot perform any task that requires insight. Third, the training-data piracy mau well kill the whole thing. Hence some niche applications for small special-purpose LLMs may remain, but that will likely be it.

On the other hand, some types of jobs may still get "gutted", even with that limited usefulness.

For criminal organizations and terrorists to hire tech expertise. Will be interesting to see how much damage this "optimization" does in the end and how much it will cost.