Like Pascalâ(TM)s wager (somewhat odd, given Pascalâ(TM)s expertise) itâ(TM)s alsoâ¦exceptionally dubiousâ¦in its handling of probability. Not just the question of âwill hyper-computronium actually make that sort of simulation viable; and will there be enough of your state stored to simulate?â(TM); but the decision matrix itself.

Pascal just flattened it into âoebe theistâ/âoebe atheistâ; despite the fact that he was well aware of at least a handful of mutually incompatible ways to do theism, some of which have explicitly jealous gods; and itâ(TM)s trivial to construct as-plausible adversarial cases(eg. a god who rewards sincere believers, regardless of belief; while punishing those who are trying to work the system regardless of whether that happens to lead them to correct beliefs; which is essentially the worst possible god to wager with).

The Basilisk bros go on at greater length; but similarly flatten the decision matrix until it flatters their desired conclusion. Apparently no enthusiasts of I Have No Mouth and I Must Scream, because the possibility of the AI torturing you specifically out of displeasure with your culpability for its existence isnâ(TM)t a concern; nor are more indiscriminate Skynet types; or powerful but disengaged universal paperclips maximizing expert systems.

The wager was shoddy work, if at least provocative, even back when it was written; and it has aged poorly.

Alternatively, require companies to pay into an annuity fund backed by insurance.

Since a pension is in effect deferred payroll, going before bonds should be a matter of law.

That's just plain untrue. You generally can't unilaterally alter the terms of a loan; but (potentially subject to certain penalties) you can just repay the loan on your preferred schedule if you can find someone offering a more favorable one you can use to do so; and, if the creditor agrees, you can just outright alter the terms of your previous agreement. They wouldn't really have any incentive to accept something worse than what they think the court will offer during a chapter 11 proceeding; but unless they really, really, want to prove a point it's entirely possible that they'd take terms worse than the original ones but better than the bankruptcy is likely to provide.

Obviously it costs a certain amount of money just to provide coffee and toner, plus the hourly cost of whoever will be discussing the matter, so you probably aren't going to 'renegotiate' some teeny little loan; especially if it's some sort of unsecured consumer credit, or situation where the finances of the debtor aren't entirely transparent or fixed(Kodak's ability to just stop buying materials from their suppliers without folding even faster is likely very limited and they were already presumably trying to get favorable prices; while Visa might suspect that you could spend less on groceries by going store-brand if they put the fear of collections hell into you); so a lot of consumer-tier 'refinancing' is strictly of the "pay off high interest credit card debt or usurious dealer vehicle loan with mortgage"/"pay off mortgage from higher interest period by taking out mortgage during low interest period); but a $440 million publicly traded company isn't really home economics territory.

Probably most who make it to the hospital are found by someone else.

They said in right in the summary. Given by the emergency department (ER) or used by first responders (paramedics).

Since clearing the CO is extremely urgent, I assume IV push would be preferred. Since it's a protein, I doubt inhaling it would be effective.

I doubt it would be over the counter just because of needing to give it by IV.

Since the employees worked there in consideration of the eventual pension, slashing it should be considered outright financial fraud. In cases of bankruptcy, the pension funding should be first in line for whatever remains.

This is the Prius hypermiler thing again.

In the same way that it's possible that the next time I trip, the wind will suddenly align just right and stand me back up unharmed. It's possible. But I'm still going to make sure to tie my shoes.

And yet, they still can't factor a 2 digit number without cheating. Note that with the cheats employed, a Commodore Vic-20 can do it faster and costs significantly less to buy or operate.

If silicon valley were so concerned about the 'open' model gap isn't there something fairly obvious they could do?

It's not like the proprietary as-a-service-so-Sam-can-protect-you-from-terminators-and-stuff models remain as tightly gated SaaS stuff by accident. I'll believe that team VC is peevish about the speed with which their unbelievably massive volume of dubiously justified capex is being commodified; but don't even try to tell me that the state of the 'open' offerings is some kind of perturbing surprise to the people who could change it tomorrow by making their not-open offerings open if they felt like it.

what kind of behavior would demonstrate that LLMs did have understanding?

An LLM would need to act like an understander -- the essence of the Turing Test. Exactly what that means is a complex question. And it's a necessary but not sufficient condition. But we can easily provide counterexamples where the LLM is clearly not an understander. Like this from the paper:

When prompted with the CoT prefix, the modern LLM Gemini responded: âoeThe United States was established in 1776. 1776 is divisible by 4, but itâ(TM)s not a century year, so itâ(TM)s a leap year. Therefore, the day the US was established was in a normal year.â This response exemplifies a concerning pattern: the model correctly recites the leap year rule and articulates intermediate reasoning steps, yet produces a logically inconsistent conclusion (i.e., asserting 1776 is both a leap year and a normal year).

Jellyfish leave nuke plant in a jam?

Just to see, I asked Google as well and it said 2-8%.

That would make 18% significant, but with a high false positive rate.

Can anyone comment on why TFA describes the litigation as 'stalled'? You can debate whether or not it's good policy; but Section 230 is relatively clear cut about "you pretty much can't pin liability on platforms for this class of activity", which seems like it would just end the case; rather than leave things open for an indefinite amount of appeals and pondering.

It would be somewhat surprising if there wasn't at least a bit of 'AI' involved in churning out that many slightly-off-looking extensions and websites; if there's anything even the pessimists will tell you it's good for it's to churn out dubious quality in volume for applications where quality doesn't really matter; but the use of a single piece of C2 infrastructure and a bunch of random pirated attack toolkits does not suggest that Glorious Age of AI has contributed much to that side of the attack.