JustinCEO Topic

A post was merged into an existing topic: Deplatforming and Censorship Examples

Continuing the discussion from Deplatforming and Censorship Examples - #75 by JustinCEO

Sorry. Don’t feel obliged to response to this either. I was a bit torn cuz I didn’t want to create unnecessary moderation work. It also appeared that I could post in the thread. Was trying to be respectful. I was hesitant to ask because I didn’t want to be demanding/expecting a reply. I should have just followed explicit instructions about where I am allowed to post until told otherwise, though. So I’ll go back to just posting here unless expressly told I can post somewhere else (which only applies to Lonsdale thread at the moment).

Continuing the discussion from Moved, Off-Topic Posts - #7 by Elliot

I noticed that MetaCreation stopped posting shortly after I wrote a few critical reply posts to them about grammar and their manner of engaging with the forum. I hope they did not leave because of that. I think I was engaging with them in an honest and non-hostile manner, though, so I don’t think I did anything wrong (if anyone disagrees, let me know).

Atlas Shrugged, Rearden/Francisco scene:

Rearden chuckled. “If you understand that much, we have at least a sensible basis for conversation. Proceed on that. If you don’t have some fancy investment in mind, what did you want to meet me for?”

“In order to become acquainted with you.”

“That’s not an answer. It’s just another way of saying the same thing.”

“Not quite, Mr. Rearden.”

“Unless you mean—in order to gain my confidence?”

“No. I don’t like people who speak or think in terms of gaining anybody’s confidence. If one’s actions are honest, one does not need the predated confidence of others, only their rational perception. The person who craves a moral blank check of that kind, has dishonest intentions, whether he admits it to himself or not.”

I think Francisco’s last paragraph is important and has reach way beyond the issue of just talking about the confidence of others.

It’s a common thing for people to want to kind of coast on stale judgments from the past. (It’s something I think I’ve struggled with btw, not excluding myself here.) People want to be a member of a “group” after a certain point and not have to earn the positive appraisal of others anew with each action. Most people’s conception of “community” is basically a group of people who will let you coast and not challenge you much once you’re a member. If you’re not being treated in a “friendly”/biased way by people but are being constantly challenged, then that’s not “really” a community along this definition. I think this is related to why people have complained about an alleged lack of community on CF/FI (because it’s a no-bias/no-coasting zone). I think it’s also related to broader issues of tribalism and why it’s so pervasive (because people want to coast, be biased, not have to constantly earn the positive appraisals of others, so our society is organized accordingly).

Continuing the discussion from Capitalism Means Policing Big Companies - #18 by Elliot

I received what I thought was a highly deceptive mailing from an energy company that, from past experience, I already knew has a practice of high pressure/dubious sales tactics, and I actually tweeted at various people who I thought might care with a picture of it, but nobody cared or got back to me or anything (not even to tell me to report the issue via a more official channel - just literally no reply). I think that’s common, which means that even if you’re an observant consumer who sees something going wrong and tries to bring people’s attention to it, you’ll often just get silence or a shrug, which doesn’t really encourage marginally motivated people to come forward with complaints. The highly motivated people who will follow up are those people who have already been screwed, but by that point the fraud has already taken place and the fraudsters might have already absconded with their gains, dissipated the money, or moved onto their next scam. Not a great system.

I asked ChatGPT for a list of keto suitable foods organized by amount of potassium per net carb. This seems like something for which ChatGPT could actually be useful and save some time.


Incidentally I’m exploring a hunch about low potassium causing me problems based on 1) some symptoms I’ve had 2) the observation that my potassium intake is much lower than recommended on my nutrition tracker, 3) a change in perspective on the potential importance of vitamins and nutrients as a result of reading Why I Left Orthodox Medicine and discussing the ideas in it a bit.

I’m not sure why you think ChatGPT would be good at this. At first glance, this is something I would expect ChatGPT to be bad at - I would expect it to be making up at least some of the numbers, or using the wrong numbers in some places, and would want to double check all of its numbers myself.

Did you double check the numbers?

Interesting. I think my understanding of ChatGPT’s capacity for “hallucinating” stuff is different than yours. I thought it was more of a narrow issue that was particularly problematic for certain things like legal citations or tax law issues. I figured that with something like nutritional info, which is widely available from a variety of sources, ChatGPT would be more safe. This seemed more akin to running a query on database and then manipulating some info a bit in excel, so I thought it’d be right up ChatGPT’s alley. Might need to re-examine my ideas there.

No, I only checked the avocado figures. Since seeing your message, I checked the numbers for chard, spinach and zucchini as well. They seem about right (a little variation here or there, but then different sources give slightly different numbers. Nothing seemed way off).

The tax law stuff is widely available too, and pretty clearly written out. I don’t see why that would be a special case. (I also have more to say about the tax stuff from earlier, but I will reply to that separately.) My experience with ChatGPT is that it gets details wrong, but it does so confidently, in a way that would make any table of data it provides unreliable. If it can’t find an answer to something, it will just make one up. And there is no way to know looking at a table which things it got from a proper source, and which things it made up. It doesn’t give sources and it doesn’t tell you which things it made it. So, imo, that makes any data from ChatGPT unreliable. You would have to check every single piece of data yourself if you actually wanted to use the data.

ChatGPT also does things like give the wrong authors for books, or make up books entirely. That information is also widely available. I see that as a general issue with being unreliable, not a narrow issue.

1 Like

I don’t think that’s actually what happened. I actually asked it more than one question to see if it would give inaccurate extension info for other things too.

E.g., I asked it about HSA deadlines and got this:

Me:

When do I have to make my HSA contribution by?

ChatGPT:

The deadline for making contributions to a Health Savings Account (HSA) for a given tax year is the tax filing deadline for that year, which is typically April 15th of the following year.

For example, if you want to make a contribution for the 2022 tax year, you have until April 15th, 2023, to make your contribution. However, it’s important to note that if you file for an extension to file your tax return, your deadline for making contributions to your HSA will also be extended until the deadline for filing your tax return, which is typically October 15th.

Though I also got the correct answer when I asked the same question with slightly different wording. It wasn’t consistent.

I also asked it about the deadline to pay taxes and got the wrong answer:

Me:

What is the deadline for paying your taxes?

ChatGPT:

The deadline for paying your federal income taxes in the United States is April 15th of each year, unless that date falls on a weekend or holiday, in which case the deadline is moved to the next business day. However, if you file for a tax extension, the deadline to pay your taxes is extended to October 15th.

1 Like

Ooof, that’s a really bad one given how many people have that particular misunderstanding.

I asked it that question more than one time in more than one way. It usually gave the correct answer, but occasionally gave the wrong answer. There isn’t any way to tell the difference between when it is getting info wrong, besides just looking it up yourself (or already knowing the answer). It doesn’t give any “tells” (that I could notice) when it is just making stuff up.

I think this makes ChatGPT is unreliable for anything where it is giving you data.

1 Like

You’ve convinced me :+1:

Actually @qwerty , were you using 3.5 or 4? I’ve noticed a huge quality difference between the two

I am using 3.5. Are you using the paid version to get 4?

Yeah I’m using the paid version.

So do you think that 4 doesn’t make those same kinds of mistakes?

Also, I haven’t used 4, so I’m wondering: is it able to provide sources and links for the information it gives? I know that 3.5 will provide inaccurate links, but not sure if 4 can do better since I think it can actually search the current internet.

Way way fewer of them, at least. I had some egregious examples of mistakes but a pro-ChatGPT friend pointed out they were all from 3.5, and I re-ran them in 4 and got the right answers.

I’ve still noticed the occasional odd error from 4 (like it telling me that -8 was greater than -4) but overall it seems waay more accurate and useful.

I just asked it to give me a link and it wouldn’t, but I tried one of the google search suggestions it provided and found a source.

Here’s a short conversation I just had with 4. Let me know if there’s anything specific you want me to try and I can share it here HSA Contribution Deadline - A ShareGPT conversation