Em Adespoton

Not relevant to the article which isn’t about LLMs… but it IS relevant to “this isn’t about LLMs, don’t downvote!” Which implicitly implies LLM discussion is intrinsically worth downvoting.

Also, LLMs doesn’t equal generalist trained chatbots.

Personally I’ve found the way to use it is to get it to do the work of processing documents I could never be bothered to process myself, and to transform large volumes of information into an indexed and cross referenced dynamic index. It does both of these things well, because that’s what language models are good at.

The end result is that I am more informed. The downside is that I spend less time going off into the weeds, which does mean picking up less esoteric information.

It’s also more mentally exhausting to have to be constantly second guessing your sources of information.

But I’ve found well prompted LLMs to be a force multiplier if used in the background to do the things you previously just didn’t bother doing at all.

What it lacks in defensiveness it more than makes up for in offensiveness.

We live in time zones. “Standard” time is just as much a fiction as DST. I’ve never lived somewhere where the sun is directly overhead at noon. I suppose such places must exist; some of them might even be on standard time when it does. In my area, it’s closer to directly overhead during DST.

That must have been very N; sunset is around 9:20 on the summer solstice in Seattle.

It’s called selection bias.

These days we are flooded with information. We’re constantly having to pay attention to multiple streams of information at once.

But our brains aren’t developed to work at this volume; they tend to latch on to and enhance around one thing an hour.

Think about how many different topics you encounter in an hour these days.

This means that there’s almost guaranteed to be coincidences happening on a regular basis due to the sheer volume of things being observed.

If you’re focused on a specific topic and the other information isn’t related, you probably won’t even notice. But if you’re focused on how often coincidences are happening, you’ll notice every single one and forget the rest. Over time it will seem like you’re the involuntary subject of first person syndrome.

Other places this crops up: you’re considering having a baby and suddenly babies are everywhere. Or you’re thinking of traveling somewhere and then suddenly notice you’re getting bombarded with travel vacation ads. Or you’re trying to quit smoking and suddenly see cigarette ads and people enjoying a cigarette everywhere.

In all these circumstances, you’re suddenly noticing the stuff that’s always there, but you normally block out because it’s not important.

Bwa ha ha… make me!

So, nothing will change?

I like to think that socialists would try to give the election system a makeover; get rid of FPTP and accelerate the removal of the electoral college. This of course would require them to gain control at the state level of traditionally red states.

Something I’ve been noticing recently is that while the cost per token on specific models hasn’t gone up, the provided interfaces for using those models are starting to chew up significantly larger numbers of tokens for the same tasks that used fewer tokens with older versions of the interface software just a few months ago. Likely the interfaces are applying more expensive guardrail prompts and charging the end user for those tokens — but the end result is that it costs 4x as much to get the same work done.

Not transferable; just shareable.