It's impossible to tell since the models don't reason. That's why they can say obviously incorrect things with confidence.

They have probabilities from it mixed in with everything else on their training data. They don't reflect the Internet so much as training data soup. But what's more important is that it's a semi-randomly generated chain of similarities. You can never ignore the randomness. That was added because it was found to make the output look more natural because otherwise it's too predictable, but it also is a reason why sometimes the output is inexplicable.

It's the randomness that I was bringing up primarily. You think you're getting data from the Web because it says so, but it's mixed with every other influence and it knows as much about that as it does about everything - nothing at all.

I like your AMD idea. They would have good vertical integration there, of the most important and expensive components. Plus it would be a way for them to finally get some competent programmers.

But in order to really make it work they would have to go back into the RAM biz...

x86? What year do you think it is?

Which of course is how you know that we're slaves

Yeah, and he was right, which they also can't tolerate apparently.

That would be destruction of evidence, we just had a whole discussion about this recently where it was established that this is the legal reality. Please please please look up how it works before saying anything at all, please.

You're an idiot all day, kiddo. I've spent plenty of time here doing that already, and if you knew how to Internet you could find it.

No. That's fundamentally wrong. It is based both on distillations from that training data and random numbers, and the training data is NEVER used directly by the LLM, only to build the model. You are confused about how any of this works.

Gemini disagrees with you.

I didn't say or imply counterproductive. What I implied was uselessness.

There are multiple reasons (several of them related to decisions by the ultra-wealthy) but I think the best one is that most people's emissions aren't very high. You could remove the bottom 50% of the population as measured by income off of the map entirely and not make things much better.

Another reason is that most of what we do could be much lower-emissions if we used strategies we already know how to implement. Take agriculture for example where zero tilth agriculture dramatically increases yields and reduces emissions. This would require more human labor, but there's a lot of that available. The problem with shifting that direction is directly with the ultra-wealthy deciding we won't do it, so that they can buy more yachts. They absolutely depend on our agricultural systems minimizing labor. People don't want farm jobs because they are shitty, but they don't have to be today, that's entirely related to the profit motive where you hire the fewest people possible.

TL;DR: Eat the rich, and the emissions problems of eating can also be solved. Fail to do so and they cannot.

What? I've never done that.

Yeah, and then they killed the ultra low cost laptop market completely because everyone discovered that they were inadequate at the time and then became COMPLETELY inadequate (as in, good for almost nothing) immediately thereafter. Most of them were single core, the good ones were dual core but didn't have enough RAM to hold up.

Now there's a low cost laptop market again but the cost of entry is double what it used to be, or at least 1.5 times with inflation. This is a little more realistic. But that's just not enough RAM for Windows. It would be OK for Linux. My Linux laptop is an HP and everything works so there's hope I guess, but I'd wait for someone else to find out.

We could arguably do both things at once, and should... but won't do one of them at all.

I should take up a collection to sue this guy for libel, so we can unmask these trolls through discovery.