I think what we do have evidence for is that you didn't read the paper, but I did, because it was interesting. From the paper:

Godel does no such thing. The incompleteness theorem says that some things can't be proven, and aren't computable, but every example of that *includes humans*. It's not a case that you can't build a computer and program in an axiomatic system that is consistent and can prove every statement with godel numbers, but that a human can prove a statement in that system that that computer can't prove. The human can't either. It's a statement about the limits of axiomatic mathematical systems.

There's no evidence anything in human thought falls under the realm of uncomputability. In fact, given that the brain is made up of neurons that are guaranteed to fire or not fire given specific conditions, electrical and chemical, there's plenty of evidence that it *must* be computable and algorithmic.

That is a terrible system, I'm surprised they do it that way. Banks are usually better about that. The only times I got a call from my bank that required me to prove who I was, it was either a returned call, and they mentioned the subject and that I had called, before they started verifying my identity, so I knew it was legit. Or the fraud alert people, and they could easily verify that they were who they said they were, because they asked about specific purchase attempts with the amount and location before they tried to verify my identity.

I did get one *actual* phishing call decades ago that made me absolutely crack up. The person on the other line said they were from "the bank." They didn't say which bank, just "the bank." Usually I immediately hang up on phishing, but that one made me want to engage a bit: I asked "which bank" and he answered, "your bank." At that point I just burst out laughing and the gig was up, so I hung up.

I had an instance of a work e-mail years ago, that was sent from a third-party contractor, that had so many red flags for very obvious phishing (including coming from outside the organization, wtf).

Where I work, we have a place to forward phishing emails so that IT can review it. I forwarded it there, and apparently so many other people did that a follow-up email had to be sent out that said, "we thank everyone for pointing out this e-mail as phishing, but we can confirm it's actually legit."

I think they learned the lesson from that, because it has not happened since that we got such a terrible email. I think my point is that overtraining may not work, but having a place to report phishing is a great idea. It only takes one person to report it, and then the IT department sends out a massive e-mail to warn everyone else about it, so it doesn't rely on them recognizing it (and anyone that already fell victim to it can report that they have, so action can be taken to minimize the damage). And in cases like you and I experienced, they can also do the opposite and confirm that it's real.

I agree, but go the other route for the comparison to humans and thought: people need to stop thinking that what we do when we "think" isn't algorithmic. Of course it is. We're not that special.

The models are trained on the same data, and they create their output based on the connections they made with all the previous data. When we ask it to generate "random" numbers, they're not any more random than when a human is asked to generate a random list of numbers. It's not purposefully encoding the information in the numbers because transmitting its love for owls is important to it, but the favorite animal information tokens are part of the seed made when it's generating those numbers.

Invariably, the second LLM that has been trained on the same data as the first will have the same data connections to those numbers. It's similar to how, when I was dating, I was filtering out anyone that added the information in the app that they had not been vaccinated for COVID. There's a *lot* of information associated with the type of person who was not only not vaccinated, but felt that they needed to state it. The information isn't contained in that assertion alone, but combined with the information already in my brain, it tells me a lot about their belief structure in things completely unrelated to vaccines and COVID. The LLM is doing that.

The fact that ChatGPT gives bad answers is a plus when using it to debug your knowledge. I often bring it down from GPT 4 to GPT 3 when Iâ(TM)m using it to help with my reasoning, because when it gives me something that doesnâ(TM)t sound right, if I can then reason out why it doesnâ(TM)t sound right, it means Iâ(TM)ve reached the understanding Iâ(TM)m looking for. Itâ(TM)s not about using it as a source. Itâ(TM)s a better version of explaining something to a rubber duck to force yourself to reason it through. Better because it gives you feedback. As long as youâ(TM)re engaged, incorrect feedback is just as useful as correct one.

Because people with high clearance sometimes want attention too, and that shouldn't be surprising. We have a lot of people with high clearance, you really expect 0 of them to be crackpots?

Of course there is. Denying funding for a program that doesn't exist changes nothing, so there are no actual consequences to the provision, so you do no damage by voting yes. On the other hand, voting no will get you accused of being part of the conspiracy hiding aliens, which is a political can of worms.

We're absolutely dealing with a cabal of conspiracy theorists, but our elected officials are too cowardly to go against them. It's the safer path to just make a show of it, it even gets you some media attention which is great name-recognition for re-election time.

I've been on reddit for two long, I apologize for the giant formatting mess above. Here's the properly formatted version:

In addition to that, it's important to keep in mind that they want to be the ones to decide what the regulations are: "Murati said the company is constantly talking with governments and regulators and other organizations to agree on some level of standards."

They don't really want to be regulated, they want the ability to tell the government what the regulation should be. If the government were to regulate in a way that says, for instance, "all AI research must be open, and you must publish the methods and source" I'm pretty sure they'd be suddenly very anti-regulation considering their newer stances.

>Because they're the first on the market and benefitted from zero regulation to get where they are, and any regulation enacted now will put barriers on the growth of future competitors. In addition to that, it's important to keep in mind that they want to be the ones to decide what the regulations are: "Murati said the company is constantly talking with governments and regulators and other organizations to agree on some level of standards." They don't really want to be regulated, they want the ability to tell the government what the regulation should be. If the government were to regulate in a way that says, for instance, "all AI research must be open, and you must publish the methods and source" I'm pretty sure they'd be suddenly very anti-regulation considering their newer stances.

I agree with you in principle, but don't you find it problematic if we say that any one individual can decide which part of the classified information should not be classified because it violates the laws of the land according to their interpretation? Everyone has different interpretations, the system in place is precisely set up to make that determination.

And yes, the system can fail us, and as far as US government spying on American citizens, I think it has. However, I think that if you decide you know better than the system, you have to be brave enough to face the consequences, and face that 10+ year in federal lockup as part of your civil disobedience. If it's not worth that, it's not worth sharing.

That's a fair point. I made an implicit assumption that it's an entry-level position for someone just graduating. In which case, they don't have that much professional experience, so you ask about stuff like that.

There's time spent on both sides, and as someone who has had to interview potential hires, I'm also thinking of the time on that end. It's not just important to fill the position, it's important to not fill the position with dead weight, and I find that time spent in the interview saves a lot of time you'd waste later if you hired the wrong person.

Yeah, they are. And it doesn't help that most applicants don't have it. Anything you can do to filter them out as quickly as possible is helping, not hurting. That includes filtering out people who think they're too good to answer a few questions.

In the time it takes me to write up a 40-page response to all these questions, I could apply for jobs at 20 other companies.

It's not a 40-page response, and if you replied to that questionnaire in 40 pages, that would probably immediately filter you out. Most of those questions are a couple of sentences, some of them are a paragraph. Furthermore, why would you apply to 20 companies? Like you said, it's a seller's market. You pick two or three companies you want to work at because what they do is interesting to you, and you apply to them. That's plenty enough to get them to fight each other when negotiating for pay.

I didn't expect that they wouldn't. What I'm saying is that they can get that information in a written interview, and use the in-person interview for questions that are more suitable for it.

I didn't really see anything in there that was irrelevant to the job. Essentially trying to get information on your aptitudes, interests, and experience. The in-person interview will do a technical white-boarding and culture-fit. I like it, seems like a solid process.

What's wrong with that? I find it rather refreshing in that it gives you a pretty good idea of the skillsets they are interested in, and lets you provide a thoughtful response in context instead of depending on you coming up with answers on the spot during an in-person interview.

Well, the only financial problem that he created is that by laying people off, they can't make their debt obligations as a result of no longer getting a salary. So that's why I was using the average debt.

So you can argue solving their financial problems is one of two things: in the first case, it's paying all of their debt, which this wouldn't be enough. In the second case is that it would allow them to make payments until they find another job, which $20k would help with...but that's the purpose of unemployment, so I would argue that by paying unemployment insurance the employer has already done their part.