Correct. This is why I don't like the term "hallucinate". AIs don't experience hallucinations, because they don't experience anything. The problem they have would more correctly be called, in psychology terms "confabulation" -- they patch up holes in their knowledge by making up plausible sounding facts.

I have experimented with AI assistance for certain tasks, and find that generative AI absolutely passes the Turing test for short sessions -- if anything it's too good; too fast; too well-informed. But the longer the session goes, the more the illusion of intelligence evaporates.

This is because under the hood, what AI is doing is a bunch of linear algebra. The "model" is a set of matrices, and the "context" is a set of vectors representing your session up to the current point, augmented during each prompt response by results from Internet searches. The problem is, the "context" takes up lots of expensive high performance video RAM, and every user only gets so much of that. When you run out of space for your context, the older stuff drops out of the context. This is why credibility drops the longer a session runs. You start with a nice empty context, and you bring in some internet search results and run them through the model and it all makes sense. When you start throwing out parts of the context, the context turns into inconsistent mush.

Is this the same Klarna that has already been through one cycle of being super bullish about AI and then having to re-hire some actual customer support?

That's only very partially true. The uptick in unpaid mortgages gave the house of cards a little tap; but it was the giant pile of increasingly exotic leverage constructed on top of the relatively boring retail debt that actually gave the situation enough punch to be systemically dangerous; along with the elaborate securitizing, slicing, and trading making it comparatively cumbersome for people to just renegotiate a mortgage headed toward delinquency and take a relatively controlled writedown; rather than just triggering a repossession that left them with a bunch of real estate they weren't well equipped to sell.

His surname is one transposition away from "AI Mode".

Yeah, because all even/all odd is (from basic statistics) rare, and happens to be rarer than the percentage of people who play all-even or all-odd, so you'd be more likely to split any winnings.

It probably makes more sense given their scale for them to have their own power generation -- solar, wind, and battery storage, maybe gas turbines for extended periods of low renewable availability.

In fact, you could take it further. You could designate town-sized areas for multiple companies' data centers, served by an electricity source (possibly nuclear) and water reclamation and recycling centers providing zero carbon emissions and minimal environmental impact. It would be served by a compact, robust, and completely sepate electrical grid of its own, reducing costs for the data centers and isolating residential customers from the impact of their elecrical use. It would also economically concentrate data centers for businesses providing services they need,reducing costs and increasing profits all around.

It's certainly possible that some people do, sincerely, 'fear' that the onrushing machine god will speak chinese and that it would be just the worst if all humans were rendered obsolete by the wrong side's robot when that's supposed to be our job; but, especially with how tepid the results are for the money poured in, it seems much more the case that we are seeing a lot of nakedly cynical playing of the 'give us what we want, lest the chinese win' by people who are otherwise on deeply shaky ground in terms of things like massive copyright infringement, voracious data mining, and an endless hunger for capital without any signs of returns.

It's like a vastly hypertrophied case of the 'race to 5G' stuff; where, if we didn't give Verizon whatever they asked for, China would have a faster rollout of 5G and we would lose the 4th industrial revolution or something? It was never entirely clearly what losing the race was going to involve.

The existential tone of the claims seem especially curious given how meagre the leads people are pouring billions into seem to be; and how readily 'AI' models can be poked at via distillation attacks or good, old-fashioned, electronic intrusion. If The Singularity kicks off that presumably changes everything beyond the powers of meaningful prediction(though that holds for whoever develops it as well as everyone else; given the odds that it will slip the leash); but as long as you are in the realm of incrementally more or less flakey chatbots it seems a bit weird to even talk like there is some sort of victory condition that will trigger and cause one side to lose.

Modern models will take into account statistical biases in lotteries. For example, here's ChatGPT:

Pick good lottery numbers for the Virginia powerball lottery

Good question. First, a reality check: Powerball is pure chance — there’s no “good” numbers that are more likely to win. Every set of 5 white balls (1–69) + 1 Powerball (1–26) has exactly the same probability.
wwwuat.valottery.com
+1

That said, I can give you a few tips + a sample “random but balanced” pick, if you’d like to pick your own numbers rather than use Quick Pick.

Some tips when choosing your Powerball numbers

Spread your numbers
Try to pick a mix across the range (low, mid, high). According to lottery-strategy advice, not clustering all your numbers in one group is better.
samlotto.com

Avoid patterns
Many people pick birthdays, anniversaries, or sequential numbers. That’s fine—but if you win, you might have to share with more people because these are popular picks.
lotteryvalley.com

Use “due” or “hot” numbers carefully
Some sites (like lottery-analysis tools) track which numbers are “hot” (drawn recently) or “overdue” (haven’t appeared in a while). But remember: past draws don’t change future odds.
m.lottostrategies.com

Mix even and odd
According to some statistical analyses, it’s less likely to draw all odd or all even numbers.
samlotto.com

Play responsibly
Odds of winning the Powerball jackpot are extremely long: 1 in ~292 million.
lotterycritic.com

Sample “balanced” Powerball pick

Here’s a random-but-balanced combination you could use just for fun (or as a basis to tweak):

White balls: 8, 23, 41, 57, 65

Powerball: 12

If you like, I can generate 5 different “smart random” picks for the next Virginia Powerball draw (balanced by probability and number distribution). Do you want me to do that?

It's a damned shame '99942 Apothis' will miss us in a few years

It's not too late to fix that.

The thing is, it very much *won't* be random numbers. It's about as far from a random number generator as you can get.

In a really bad model, one would expect it to likely be "numbers it's seen the most in the context of a lottery". But in a good model, esp. a thinking model, one would expect it to think over which sorts of numbers are statistically over-chosen (birthdates, etc) and avoid them in giving its answers.

The funny thing is, think of how this woman will interact with ChatGPT from now on. ChatGPT could say "Abraham Lincoln was married to John Wilkes Booth", and she'd be thinking, "Okay, that sounds really, really wrong, but on the other hand, it picked the right lottery numbers, so..."

I'll go with NHTSA and NASA over the "Barr Group" ambulance chasers, thank you. Barr found that it's possible if you get like a cosmic ray to flip just the right bit you could stick the throttle on (but still not make it overpower the brakes). NHTSA and NASA investigated not just the software but the actual cases. In not a single actual case that they investigated did they find that it wasn't well explained by either stuck pedals or pedal misapplication (mainly the latter).

Oh hi, I remember chatting with you earlier :) Sorry to hear about your back. If you're ever planning on visiting Iceland, drop me a line, I can make recommendations. :)

There's some fascinating new work on "inverse-vaccines". In the same way that antigens can be flagged as "foreign", they can also be flagged as "non-foreign" by attaching N-acetylgalactosamine (pGal) to them. The liver recognizes that tag and uses it to suppress immune activation against that antigen.

That's not the goal of a vaccine against a dormant virus (destroying B-cells), it's about developing a more capable immune reaction against the virus itself. See for example the shingles vaccine (targets dormant VZV, aka shingles / chickenpox). With a strong immune recognition of the virus, as soon as it tries to reactivate, it's immediately targeted, preventing it from becoming problematic.

Dormant viruses use a combination of (A) techniques to suppress immune recognition of them, and (B) low / no reproduction until your body's immune recognition of them has weakened. Vaccines help deal with both issues.

(BTW, if you're getting up there in age and haven't gotten your shingles vaccine, do so. It's one of the "rougher" vaccines, IMHO (both on my initial and followup doses I had "flu symptoms" for a day, when I normally have no reaction at all to vaccines), but that's *way* better than getting shingles)

The funny thing is that as soon as I saw "[condition] may be linked to a common virus" I thought, "It's Epstein-Barr, isn't it?"

Seems it causes bloody everything under the sun :P

As soon as there's even a clinical trial I can sign up for to get vaccinated against it, I'm getting it. I had mono in my late teens, so I can be expected to have dormant Epstein-Barr in me. A horrible autoimmune condition that my mother has (which leads to among other things her skin regularly feeling like it's on fire) seems to be linked to Epstein-Barr reactivation.