Stack Exchange
log in

A. R.

 Discussion on question by Duzii2: Man…

Imported from a comment discussion on workplace.stackexchange....
Andrew Ray
Sep 17, 2023 22:21
@ThorbjørnRavnAndersen No Outlook for Linux. And a lot of companies don't particularly like open source alternatives because they can't get a support contract for them, which means that if you have a problem with whatever email client, it's got to be dealt with in-house and they mightn't have the staff for that. You might say "Oh, that's okay, I know what I'm doing," but the company might say "We're paying you to do XYZ, not fiddle around with your laptop settings trying to get things working. That's what we get service contracts for." On the other hand, banning testing is an outrage.
 

 Discussion on answer by Neil Slater: …

Imported from a comment discussion on ai.stackexchange.com/que...
Andrew Ray
Aug 31, 2023 14:50
@user21820 That's not how probability works. If a source is right 50% of the time answering "yes or no" questions, that would be true. But even when picking between three options, a source that is right 50% of the time has value. And in the real world, most questions are open-ended, with thousands of possibilities. In that environment, 50% accuracy is quite good. BUT then there's also the question of evaluating accuracy in the first place, especially if reasonable people can disagree. How can this be done?
 

 Discussion on answer by Franck Dernon…

Imported from a comment discussion on travel.stackexchange.com...
Andrew Ray
Aug 30, 2023 22:49
I'm pretty sure I read somewhere that Waze contributes to traffic by trying to route people around congestion, which means that any spot of bad traffic anywhere will cause a ripple effect of traffic on nearby routes.
 

 Discussion on question by Lunartist: …

Imported from a comment discussion on workplace.stackexchange....
Andrew Ray
Jul 22, 2023 17:24
@A.bakker Your company uses your unique ID number for email addresses? That sounds extremely annoying.
 

 Discussion on question by IamJojo45: …

Imported from a comment discussion on travel.stackexchange.com...
Andrew Ray
May 31, 2023 03:26
How long will you be in Tallinn? If you will be there less than 7 days and it is not the primary purpose of your trip, then you should instead get your Schengen visa from the Hungarian or French embassy.
 

 Discussion on question by Hans-Peter …

Imported from a comment discussion on ai.stackexchange.com/que...
Andrew Ray
May 26, 2023 16:38
If I may be so bold, you're not very good at giving Jeopardy! answers. You'd never flip a blue tile on Jeopardy! and the other side just says "Utilitarianism." That's more like the Technical Difficulties' Reverse Trivia Podcast, which is a podcast and not a quiz show.
 

 Discussion on answer by cheery.beach7…

Imported from a comment discussion on academia.stackexchange.c...
Andrew Ray
May 19, 2023 16:07
@cheery.beach7701 I'm confused. Are you a professor of computer science and a professional carpenter? In your answer you say "I'm a professor" who asks people to bring laptops to class "all the time," but in the comments, you say "I happen to be a carpenter."
 

 Discussion between Andrew Ray and Jon…

Imported from a comment discussion on academia.stackexchange.c...
Andrew Ray
Dec 29, 2022 19:47
Perhaps "acceptable" was not the best choice of word. Of course, the ultimate authority on acceptability will be the journal themselves. You could certainly ask for an opinion from the editors, and perhaps should, but that is separate to this discussion. I meant to say that in the specific scenario you describe, it seems to me that you and the student from Japan are being sufficiently careful to avoid the issues I discuss in my post, but I am concerned that others would not be.
Andrew Ray
Dec 29, 2022 19:44
I'm sorry, I don't mean to imply that the original author is incompetent. I'm trying to imagine if I would miss similar things if I were writing in Spanish, in which I am competent but not proficient, and I must conclude that I probably would.
Andrew Ray
Dec 29, 2022 19:37
I mentioned earlier that I was skeptical that the author who is not proficient in English and therefore needed to use GPT to clean up the paper would recognize any subtle changes in meaning introduced by GPT. I maintain that concern. In your specific case, it may be acceptable to have a main author write the paper, then clean it up and have a second proficient English speaker review the result, assuming you are also familiar with the subject matter.
Andrew Ray
Dec 29, 2022 19:28
Your original question does not make any mention of a human reviewer, and in that context, I believe that anything other than a categorical "no" would perpetuate the blind and unquestioning reliance on GPT by which I have lately been so alarmed.
Andrew Ray
Dec 29, 2022 19:26
As a programmer, I see a lot of people who don't understand what GPT is and is not. GPT is very tricky, because it is designed exactly to be very good at pretending to know anything when it in fact knows nothing. Many people of late appear to have been misled by this (admittedly very convincing) pretense, and as a result don't treat its output with sufficient caution.
Andrew Ray
Dec 29, 2022 19:16
I said before that "a human reviewer would also need to look carefully for any information added or changed by GPT and find an appropriate source for that information or discard it if no source is found."
Andrew Ray
Dec 29, 2022 19:12
Common knowledge is knowledge that any reader with suitable education in a particular field can reasonably be expected to know and accept in advance. For example, the claim that the earth is round should not need a citation in a paper on geology. What passes for common knowledge depends on the target audience of the paper.
Andrew Ray
Dec 29, 2022 19:10
But you seem to have a misunderstanding here. Reading a few books and then using the knowledge to write parts of a paper is not common knowledge.
Andrew Ray
Dec 29, 2022 19:09
Epistemology is the philosophy of the nature of knowledge. To discuss the difference between human knowledge and GPT's knowledge would require delving quite deeply on the one hand into Markov models, which I could explain given several hours, and on the other into epistemology.
Andrew Ray
Dec 29, 2022 19:06
Sorry, I'm not really interested in arguing epistemology today.
Andrew Ray
Dec 29, 2022 19:05
@JonathanReez Sure, if I wrote that very same sentence in a paper about gamification, I would want either to provide a source, or claim it as my own "common knowledge." GPT doesn't have common knowledge--- everything it knows comes from source material, so it needs a citation. Unfortunately, GPT also can't tell you where it got that information from; it can't cite its own sources. So a human reviewer would also need to look carefully for any information added or changed by GPT and find an appropriate source for that information or discard it if no source is found.
Andrew Ray
Dec 29, 2022 19:05
GPT doesn't just rewrite your own sentences. GPT can't just rewrite your own sentences. GPT doesn't understand the concept of rewriting sentences. In your own example, GPT correctly writes in its first sentence that gamification is about feedback to the user when the input text incorrectly says it's about feedback from the user. It's pulling that correct information from somewhere, and that somewhere is not the input text. In an academic paper, if you can't cite a source for that, it's plagiarism, whether you meant it to be or not.
Andrew Ray
Dec 29, 2022 19:05
@JonathanReez I will allow that, combined with the judgement of a fluent English speaker, GPT may produce better texts. The non-fluent speaker will not likely catch these errors. Even still, the potential for accidental plagiarism remains, so for academic writing in particular, my answer is still no.
Andrew Ray
Dec 29, 2022 19:05
@JonathanReez My answer is not based on this sample alone. These issues are inherent in the technology GPT uses, and GPT-4 will have the same flaws in this regard. At a very high level, GPT only answers the question "what text is most likely to come after another given text?" GPT will never understand the input it is given because its underlying technology is not capable of understanding, and it will always be prone to incorporating ideas from other sources because that's what it does. I draw examples from your given sample, but the conclusions are general.
 

 Discussion on answer by Weather Vane:…

Imported from a comment discussion on law.stackexchange.com/qu...
Andrew Ray
Aug 16, 2022 20:46
@Davislor True, but their "not guilty" verdict must stand regardless because of the prohibition of double jeopardy. The jury could be sanctioned if (in the world of the film) it came to be known that they knowingly broke the rules, but there's no possibility of a conviction for the defendant.
Andrew Ray
Aug 16, 2022 20:46
@RomanOdaisky Apart from the fact that Twelve Angry Men is a work of fiction and nobody was actually killed at all, that the jury acquitted the defendant suggests that there was, in fact, reasonable doubt. The reasonable doubt standard is very high. You may believe that the jurors were wrong, but if you make a habit of calling people guilty who were acquitted, you should expect someday to find yourself defending against charges of slander.
 

 Discussion on answer by littleadv: Ca…

Imported from a comment discussion on travel.stackexchange.com...
Andrew Ray
Aug 16, 2022 16:08
@Doc Matter of Hira is about "an alien who, on behalf of his employer[...] travels to various cities in the United States to take orders from, and measurements of, prospective customers whom he does not solicit[...] and who receives only expense money while in this country, his monthly salary being sent to his parents[...]" which is so far off the situation described here, where the traveler a) travels for personal reasons, not on behalf of the employer, b) engages in business not requiring presence in the US, and c) continues to receive a salary while in the US, as to be irrelevant here.
Andrew Ray
Aug 16, 2022 16:08
@DarrelHoffman That sounds like an interesting question, but probably doesn't belong on this stack.
 

 Discussion on answer by davidbak: Wha…

Imported from a comment discussion on retrocomputing.stackexch...
Andrew Ray
Aug 11, 2022 08:52
@Bobson Probably the most obscure character in Unicode is ⍼ RIGHT ANGLE WITH DOWNWARD ZIGZAG ARROW, affectionately called "angzarr;" there seems to be nobody on earth who knows what that character is meant to be used for.
 

 Discussion on answer by nvoigt: Shoul…

Imported from a comment discussion on workplace.stackexchange....
Andrew Ray
Jul 15, 2022 05:50
@FreeMan Technically HR can legally say anything that isn't defamatory or subject to confidentiality; that is to say, anything that is complementary, subjective, or true. But defamation is a notoriously fraught bit of law and most employers are not interested in spending time in court defending the contents (or omissions) of their references when you sue them, so they stick to stating only things that are neutral, unambiguous, and easily proven fact. Such as "John Doe worked in the basketweaving department at ACME Company, Inc. from July 3, 2014 through April 19, 2020."
 

 Discussion on answer by DJClayworth: …

Imported from a comment discussion on workplace.stackexchange....
Andrew Ray
Mar 2, 2021 07:29
As a US software tester, I can confirm that it is "good code overall, but there's just a small mistake here."