Last year LLMs were simple token predictors. Over the last year they've been making changes. Your argument is basically that they aren't yet sufficiently smart, and you may be correct. But if you take out the concept "yet" you wouldn't be.

No. It's inherent. If you give a system a goal, it will work to achieve that goal. If there's just one goal, it will ignore other costs. And if it's smart enough, it will notice that being shut down will (usually) prevent it from achieving its goal.

So the problem is we're designing AIs with stupidly dangerous goal-sets. E.g. obeying a human is extremely dangerous. He might ask you to produce as many paper clips as possible, and there goes everything else.

My preference for Firefox is based on the bookmark sidebar, with nested folders of bookmarks (similar to the way file are nested within folders on a disk drive).

The problem is that I've frequently heard of "anonymized" data being de-anonymized. It's not something I worry about a great deal, but perhaps I should worry more...and I've no real way to tell.

There are several alternatives, depending on your use-case. My fall-back is Falkon (from KDE).

FWIW, I didn't detest Pocket. OTOH, I never used it, either. But "universally detested" is wrong. I thought of it as "dead wood", but that's a very different category. There are LOTS of software capabilities that I don't use.

To me, panning a browser because it isn't optimized to run on a phone is silly. Saying I prefer a different browser on my phone would be sensible (if I though web browsing from a phone was sensible...but with my eyes that's never going to be true).

Well..."sort of virus first" is probably correct, but that doesn't mean that things don't sometimes go into reverse. The "sort of virus" couldn't be like the current stuff, because it couldn't be parasitic. And there are arguments that a "sort of cell" evolved before the genetic machinery. Nobody really knows, (See "metabolism first" https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fui.adsabs.harvard.edu%2F... for one argument.)

I, personally, suspect that the proto-"cell walls" and the proto-"genetic machinery" evolved their first stages independently, and mixed together later into something with superior "stability".

Well, they might let me remember people's names. There are probably other uses, but that's the one that occurs to me.

I accept that it is breaking the law. That doesn't mean any official is going to enforce that law. There are lots of laws that are frequently broken in obvious ways, and are enforced only when it is a convenient excuse.

How far in the future?
I can imagine useful AI based "smart glasses", but that description doesn't fit what I've heard so far, and my prescription means I couldn't try them if I wanted to do so.

If it's doing auto-transcription it's probably training an AI with your information, This could be a major security problem.

It's annoying, but good, because "there's a sucker born every minute" means there are always new victims who haven't been warned.

There have been lots of instances where companies with a "good reputation" changed their spots.

The post at the top of the thread was about "AI". The following posts were about AI. Don't be blinded by the current hype into thinking that;s the whole picture. Just because other developments get less press doesn't mean they aren't happening and aren't important. In the field of biochem, most AI is *related* to LLMs, but is significantly different.

LLMs are not equivalent to AIs, they are a subset. Don't take LLMs as a complete model of the capabilities of AIs.