So it eventually destroys the company.

So it is just as good as the typical CEO

It is a different substrate for reasoning, emergent, statistical, and language-based, and it can still yield coherent, goal-directed outcomes.

That’s some buzzword bingo there… A very long winded way of saying it isn’t human-like reasoning but you want to call it that anyway.

If you went accept that reasoning often fails to show continuity, well then there’s also the lying.

Examining a reasoning chain around generating code for an embedded control scenario. At one point it says the code may effect the behavior of how a motor is controlled, and so it will test if the motor operates.

Now the truth of the matter is that the model has no access to perform such a test, but the reasoning chain is just a fiction, so it described a result, asserting that it performed the test and it passed, or failed. Not based on a test, but by text prediction. So sometimes it says it failed, then carries on as if it passed, sometimes it decides to redo some code to address the error, but leaves it broken in real life. Of course it can claim it works when it didn’t at all. It can show how “reasoning” can help though. If the code is generated based on one application, but when applied to a motor control scenario, people had issues and so generating the extra text caused it to zero in on some stack overflow thread where someone made a similar mistake.

Oh phone trees are terrible, I refer exclusively to online self service. I suppose an LLM might be able to help a caller connect to the correct set of humans better than phone trees…

If I’m resorting to phone, it’s because I really really need a human. I know there still exist some very old people stuck calling… But if they can’t work your online portal, they won’t be able to work a phone tree either…

The robo-bullshit is great, if the thing has no nuance. Self checkout, paying bills, buying stuff online.

The things is those things are great because they are so predictable. LLM takes the predictability out. It’s also generally not allowed to do anything that the self service portal was not allowed to do, so you get stuck with a more imprecise interface instead of the nice, precise interface of a traditional portal, and no access to more nuanced help. It’s the worst of both worlds.

Yeah, have a new executive who managed a vaguely segment appropriate “hello world” with code gen and so regularly rants about why we should be paying human developers.

The biggest improvement on the user side was to stop trying to weigh the bagging area to prevent loss.

The newer machine vision based systems are less likely to screw up. “Unexpected item in bagging area” was an almost universal experience, nowadays I have only been flagged for human review once.

Also, one store I was at just lets you put your items under a camera without finding barcodes, and you just confirm the identified products.

Think the issue is either a self service portal that works in very predictable way (like the self checkout) or a human to deal with nuance.

To the extent an LLM might be useful, it’s likely blocked from doing so because the operator doesn’t trust it either.

The biggest annoyance is that the LLM support tends to more aggressively refuse to bring a human in.

And we didn’t have CFC deniers with huge social media platform amplifying fringe conspiracy theories into big political platforms.

Of course, the dangerous CFCs weren’t as critical, if anything new formulations were easy business opportunities for established players. There’s no easy pivot from being a big fossil fuel company to a replacement, any attempt to do so comes with huge risk of being distributed by an unexpected competitor.

The “reasoning” models aren’t really reasoning, they are generating text that resembles “train of thought”. If you examine some of the reasoning chains with errors, you can see some errors are often completely isolated, with no lead up and then the chain carries on as if the mistake never happened. Errors that when they happen in an actual human reasoning chain propagate.

LLM reasoning chains are generating essentially fanfics of what reasoning would look like. It turns out that expending tokens to generate more text and discarding it does make the retained text more more likely to be consistent with desired output, but “reasoning” is more a marketing term than describing what is really happening.

The things to remember is that these CEOs have made a whole living out of not knowing what they are doing, but being insufferably confident in whatever vomit of words they spew, whether they know anything or not, while ultimately just saying the most milquetoast blatantly obvious stuff and pretending it’s very insightful. All this while they believe and the money proves that are the most important people in the world.

So naturally it’s easy for them to believe LLM can take all the jobs, because it can easily take theirs.

I just don’t get how so many people just start by it. Every time I set my expectations lower for what it can be useful at, it proceeds to prove itself likely to fail at that when I actually have a use case that I think one of the LLMs could tackle. Every step of the way. Being told by people that the LLMs are amazing, and that I only had a bad experience because I hadn’t used the very specific model and version they love, and every time I try to verify their feedback (my work is so die-hard they pay for access to every popular model and tool), it does roughly the same stuff, ever so slightly shuffling what they get right and wrong.

I feel gaslit as it keeps on being uselessly unreliable for any task that I would conceivably find it theoretically useful for.

5% is a pretty high wealth tax, so I actually find this credible.

As rough as property tax can be, 5% would be crazy. Imagine being told you owe 25,000 dollars because Zillow shows houses near you selling for 500k…

I prefer the concept of stock secured loans counting as income.

Considering a lot of the full YouTube videos are full of padding, taking over ten minutes to get to the point, I can understand why the shorts would have appeal.

Problem is one way or another people are being incentivized to target a specific runtime regardless of whether they have the material to fit.

Well yeah, I would assume Steam would be a big priority for this scenario…

It is one reason why they go into small towns, they need a sucker municipal leadership that isn’t aware how universally badly deals for datacenters work out for communities.

Those are getting harder to find, thanks to enhanced media attention…

Had a relative with a toddler that almost died due to his GCM overreporting his levels.

My mom had one and learned immediately not to trust it.

I’m shocked that both people I know personally had those devices turn out to be uselessly inaccurate…

Nah, AI code gen bugs are weird. As a person used to doing human review even from wildly incompetent people, AI messes up things that my mind never even thought needed to be double checked.

Well, the document is real, so the question would shift to of it could be a false allegation. I suppose for each document we can’t know, so all we can do is keep in mind that any one document could possibly be a false allegation. If someone can connect this to other documents, that could help. Particularly connecting it to the claimed direct agent contact or followups.

On the one hand, it seems believable, Trump is a trash human known specifically for creepy regard for sexiness of underage girls. So undeniable that the reality of the Epstein release induced some die hard maga to pivot to a “technically not a pedophile if they started their period” bullshit. Openly ogling his own teenage daughter, peeping on underage beauty pageants, and after a casual interaction with a young girl the very first words out of his mouth were about when he could start dating her…

On the other I suppose we have to acknowledge that this was a tip submitted to the FBI website at the height of his re election campaign right when the death of Epstein and his Trump connection were national news, and all the stuff about trump was already common knowledge. Out of the hundreds of millions of invested parties, I wouldn’t be surprised to find out some submitted bogus tips to a website, either not caring about risks or not taking the risks of false information over a web form seriously.

Though that one had the most critical information unredacted. Certainly worth highlighting that particular document regardless, but technically not an answer to the question about redactions.

The school doesn’t even need to do that to effectively squash suspected behavior in the short term.

Maybe they can’t dole out a substantive punishment, but when I was growing up they absolutely would lean on kids for even being suspected of doing something, or even if they hadn’t done it yet, but the administration could see it coming. Sure they might of wasted some time on kids that truly weren’t up to anything, but there generally weren’t actual punishments of consequence on those cases. I’m pretty sure that a few things were prevented entirely, just by the kids being told that the administration sees it coming.

So they should have at least been able to effectively suppress the student body behavior while they worked out the truth.