The Nineteenth Byte

@NewPosts I don't think a lang with 10 total answers needs a tips question

it could be warranted circumstantially... with the main circumstance being, the poster has already thought of one or two meaningful tips

1:25 AM

https://i.stack.imgur.com/6ZidA.png
https://i.stack.imgur.com/4G6JB.png
Why this come to my inbox?

2:12 AM

Great, yet another metagolfscript

2:23 AM

it isn't MGS though. It's just one way of encoding programs

e.g.: github.com/benbusby/y2k/blob/develop/examples/fizz-buzz.y2k

that's the full form

it's like that one language where the program is the file name

and the byte count for those answers includes the filename length

Aiden Chow

> A post’s author comments on a post you have previously commented on, when the only other (undeleted) comments on the post belong to you or the author. (Source)

@emanresuA also, scoring y2k is covered by these meta posts

meaning that the multifile scoring system is actually a sort of disadvantage

Language Custom (different version is different compiler), used to mean a theoretical bound of a challenge

3:23 AM

CMC: Find the user with the most (of caird's) brownie points

4:06 AM

https://codegolf.stackexchange.com/a/256318/76323
Although requirement only need lowest type of randomness(for some input, there's multiple possible outputs), does this code generate all possible outputs?

(It repeatly move the last element to position 0 or 1 randomly (0-index) until intended order

It terminates with probability 1 so I'd say yes

I know it's valid, just how random is it

7:31 AM

Found this link lol

It just reloads forever?

I'm really sad this requires support for negative numbers codegolf.stackexchange.com/questions/60610/… Could do a really short ><> solution otherwise

8:26 AM

@emanresuA LMAO

Wonder if it's a valid answer to this

DialFrost

8:47 AM

@mousetail appreciate the code.golf referral! :3

9:01 AM

@emanresuA dang it Dion, why couldn't you make better redirects than this? :p

Also wow it's been way too long since I last thought about MAWP

9:20 AM

hi

9:55 AM

This feels like ChatGPT right? codegolf.stackexchange.com/a/256327/91213

Neil

looks like I managed to outhat Community this year

@mousetail I thought that too

yeah

def

What is interesting is the account is 6 years old but posted here for the first time

10:03 AM

8 years old on so

@mousetail feels too informal for chatgpt

Just doesn't have the same feel. Doesn't mean it's not using some other tool though

If you type the question in ChatGPT what do you get? I don't have access to it

Also maybe a mod can remove it from HNQ? This type of question shouldn't really be HNQ

@mousetail okay maybe it is chatgpt then

gist.github.com/Lyxal/8703591c2467fc2a0963773b904618d3

I was genuinely not expecting that

^ that's what I got

Pretty close to what I got too

And close to the answer

10:11 AM

same

Probably is ChatGPT then

Does chatGPT give different answers to different people?

Yes

It's non-deterministic

You can press "regenerate response" and it will give you a different response as well

I imagine it'd have a temperature of 0.7 like the default settings of gpt3 do on the open ai site

(temperature 0 would be deterministic, always the same result)

gptzero.me could be a helpful resource for detecting ChatGPT answers

it's down right now, but I've signed up to be alerted about when it comes back up and to join the beta

11:04 AM

I wonder if ChatGPT can detect ChatGPT-style text.

better ask ChatGPT :P

(my guess is no, because if it has an adversarial component then the whole point is that it can fool itself, but it could also totally have some kind of emergent capacity for that anyways)

11:18 AM

Whether or not ChatGPT can detect ChatGPT-style text depends on whether or not ChatGPT can be taught to recognise that style.
Humans can often spot some common patterns in ChatGPT responses, and ChatGPT can be taught to do detect patterns, so it should be possible to teach the patterns to ChatGPT.
In conclusion, ChatGPT should be able to detect ChatGPT-style text.

...did chatGPT write that

11:42 AM

@mousetail No, I tried emulating its style.

That leads to a follow up question

could chatGPT recognize a person trying to imitate it's style

Well, I'd certainly spot the word "recognise" which ChatGPT would have spelled "recognize".

Uh:

> Can you detect if a given text was written by ChatGPT?

I'm not able to detect if a given text was written by ChatGPT or any other specific language model. However, there are ways to determine if a text was generated by a machine or written by a human. For example, one way to do this is to use a machine learning classifier that has been trained to distinguish between machine-generated and human-written text. This can be done by training the classifier on a large dataset of texts that are known to be either machine-generated or human-written and then using the classifier to predict the ori…

Compare my prompt to the last sentence of its reply!

It’s contradicting itself between the first and last sentence

chatGPT is being trained on this conversation right now

So, I gave it this and it replied:

> Based on the content of the text, it does not appear to have been written by a machine language model such as ChatGPT. The text provides specific advice and recommendations for improving the performance of code, which suggests that it was written by a human with knowledge and expertise in programming. However, without more context it is not possible to say for certain whether or not the text was written by a human.

11:48 AM

The logic it gives for suggesting it’s a human doesn’t even make sense considering itself can give recommendations like that

Don't get fooled: ChatGPT is just a very fancy ELIZA. It has no actual understanding of anything.

It's a paradox. ChatGPT things the other answer can't be by AI because it gives specific advice, but this text itself is specific.

@Adám that's right

13 hours ago, by lyxal

even then, it doesn't actually "know" Vyxal or Jelly or any other language it purports to know. It simply writes stuff that it thinks should be statistically completed based on what it's been fed. It may look like it "understands" code snippets, but it's simply the most appropriate/probable token generation from the tokens you give it

applies to everything it says

Transformers are just the new fancy encoder architectures, but it boils down to the same concepts of embedding data in large feature spaces

does a human actually "know" Vyxal or Jelly or any other language it purports to know?

or are we, too, simply writing stuff that we think should be statistically completed based on what we've fed ourselves?

11:53 AM

i know i am :P

I’ve worked on them with my Ph.D. student for something else, attention mechanisms are interesting features but they’re not that revolutionary from a theoretical standpoint

I guess the main reason they’re popular now, besides impressive applications for image/text generation, is that they’re significantly faster to train

So you can feed them even larger datasets

@pxeger no, not really. When we solve a problem, we can somewhat actually understand what and why we're doing it instead of spitting out tokens based on probabilities.

It can't count

2+2+2+2+2 = 8?

A!

And it got it wrong as well: A is definitely not encoded as "A!"

11:56 AM

@pxeger In Adám’s example, a human would know that the last sentence does not make logical sense after having said the first one

@TheThonnu that last 2 is actually an impostor

We also use a lot of statistical reasoning, but it’s in conjunction with symbolic reasoning. Combining both in a general purpose way is a long-standing problem of AI

@TheThonnu Well, statistically speaking, if you add together "a bunch of" 2s, you get 8.

It did figure out the mistake eventually

But it's still very wrong

@lyxal my point is, does it matter if it "really understands" what it's doing? I believe I understand what I'm doing (sometimes), and I get the impression from your behaviour that you understand what you're doing. But can I ever really be sure that you understand what you're doing, and that you aren't just really good at pretending to understand, by generating your replies statistically?

I think I can't, and so it doesn't necessarily matter that ChatGPT can't "truly understand" the things it talks about

12:00 PM

Well it's not like I have a ~~tablualatorara~~ look-up table of words that ~~wet~~ frequently appear one ~~mornign on eee~~ after based on training data

unless GPT has been trained on complete gibberish, that sentence would be highly unlikely to be statistically generated

I think the main relevant difference between a human and ChatGPT is that a human is able to sit down for an hour and think about stuff, and then come up with something entirely new as a result

Whereas if you don't feed ChatGPT any more input, it won't come up with any new output

I don't think the understanding is what matters

@pxeger Even that is not really true

e.g. models that learn to play chess/go by playing against themselves and that end up better than humans with new ideas

my point is more that when it gives you an answer, it hasn't really "thought" about what it's saying like we would, it's just strung words together based on the frequencies of combinations of words. It hasn't thought "hey I should check if what I'm giving as an answer is correct", it's just "pick word after word"

I’m sure you could use a transformer-based model to translate from english to a random new language, which would be pointless but which would be statistically "sound" and new

@Fatalize A chess model can come up with new good ideas, but it's only by determining that a randomly generated idea is good, which it can do easily by winning chess matches against itself

I don't think ChatGPT has an easy way to check if a randomly generated idea is good other than by comparing it against the input it's been fed already

12:06 PM

@pxeger How is that different from a human? They try new ideas and see if they work

At first they might be random but as training progresses, the new ideas it tries are guided by its existing knowledge of the game

Yeah I didn't mean actually random

@pxeger It doesn’t but that’s mostly because its sole purpose is text generation

You could add additional models and objective functions to do something else

The question of what it actually understands doesn’t really matter

The main point is that you can show that some of the things it comes up with do not hold against symbolic reasoning, and there is no obvious way to map feedback from symbolic to statistical

1:16 PM

Hi

1:35 PM

iH

I really wish Python's ZeroDivisionError was only for moduluses

With zero divisions it should return float("inf")

Porting Scratch to Python is hard due to this

:/

You know, creating a 3D program

Divisions by zero will be frequent

PyGamer0

@UndoneStudios it doesn't make sense you know

and should return infinity

It works in Scratch, but not in Python

PyGamer0

1:38 PM

1/0 -> inf and 1/-0 -> -inf

And since in Python 1/float("inf") == 0.0, shouldn't the converse be true as well?

@PyGamer0 No hold on

-1/0 -> -inf

@UndoneStudios But 0/0 just isn't float("inf")

@mathcat yes, that should return float("nan")

I meant n/0 isn't infinity either

._.

what are you talking about? there is only one correct answer to division by zero and it is undefined

1:42 PM

Makes sense, because 0xinfinity is not 0

nah

Infinity isn't a number, don't mess with it.

^

@UndoneStudios what exactly are you trying to do here? :p

@mathcat It's true

0/0 == 0*(1/0) == 0*infinity

@UndoneStudios nope, wrong

anything divided by zero is undefined

depends on your value of 1/0 however

1:44 PM

undefined

it's always undefined

and since 0/0 is indeterminate, so is 0*inf

anything times infinity is infinity

@Ginger depends on whether you're a practical kind of guy

@UndoneStudios no it does not

In the practical sense, yes

1:45 PM

If you aren't working on set theory, calculus, or anything like that, just ignore infinity.

That's the best you can do.

Mathematics does not depend on your perception of reality :b

Infinity + anything is infinity, 0 / anything is undefined

that's just the way it is

Anyways, the reason I was about infinity was because Scratch's method of handling infinities makes it easier to create a 3D game

since divisions by zero will be frequent

why???

@Ginger because z clipping

if I remember correctly

@UndoneStudios can you not add a safeguard that keeps divisions by zero from happening?

1:47 PM

Sadly, Python's Big Brother style of looking at divisions resembling "x/0" makes me have to write a lot of code

o noes

@Ginger that's not going to help the 3D game

The only way is to make x/0 be infinity and not some obnoxious error

In what kind of 3d things does 1/0=infinity make sense?

Matrix libraries should take care of this stuff

@mousetail I'm not sure, when rendering things at the player's position

You chould cull faces closer than the near plane anyway

1:50 PM

@mousetail yeah

that's why you have the near and far clipping planes

@mousetail but it requires infinity divisions sometimes

you never know

I found out the hard way

Numpy or whatever matrix library you use should take care of it

@mousetail I use built-in libraries, not third-party ones

Actually no, even numpy would be too slow. You want to do it in your shader only

1:54 PM

Yeah, that's why I write all the modules for my code by myself

@Ginger Actually, I think it's anything / 0 that's undefined

0 / anything works OK

@UndoneStudios whoops :p

typo

Do the culling with numpy only on bounding boxes then per vertex only in the shader which will run on the GPU

@mousetail Yeah for starters, like I told you, I don't use numpy

Then use numpy

And I'm not planning to

1:56 PM

Or another linear algebry library

there are many

@UndoneStudios why

smypy-fam

I'd rather create my own

WHY

I hate outside stuff! I'd rather write my own code because at least you'll understand it

Steve Jobs style

1:59 PM

Linear algebra code is really boring though

one does not simply rewrite a matrix algebra library

I can't stop you but I can tell you that you will regret that decision

@Ginger you could make that a meme

based on "one does not simply outgolf Dennis"

chat lag :/

@mousetail Yeah right I love algebra

It's not hard just tedious

No linear algebra

completely unrelated to algebra

2:00 PM

@UndoneStudios though it'll have to be "One does not simply outgolf xnor"

@mousetail it's fun

honestly people are stereotyped against math

:(

I love math

But I want to just mutiply 2 matricies

not worry about if X1 = Y3*X3+Y4 etc.

I think it's time for a "Do I love math?" checkup

@UndoneStudios Math is literally half of my username

@mathcat approximately 4/7ths

not a half

But I agree with mousetail, writing matrix libs is tedious-hard, not fun-hard.

2:05 PM

yeah if you truly like math you'd still find it fun

doesn't sound like you do :(

No

WTF do you think math is

It's not tedious remembering of fomulas

No it's not

but it sounds like you're lazy

that's all

@mousetail linear algebra is kinda related to algebra :P

I don't see how

Algebra is solving for X

linear agebra is how to multiply matricies

you were talking about that kind of algebra?

No sorry I don't know just yet how to multply matrixes

Maybe later

2:19 PM

uhm okay

A massive amount of linear algebra is the study of vector spaces, an algebra is literally a type of vector space

That's why you want a library that can do that efficiently

@cairdcoinheringaahing I heard the term "algebra" comes from the Arabs

However although I can speak arabic I don't know whether that's true

but the arabs have done a lot, so I think they did

Wikipedia says it is:

> Algebra (from Arabic ‏الجبر‎ (al-jabr) 'reunion of broken parts,[1] bonesetting'

IIRC europe adapted algebra from an arabian book or smth

2:22 PM

@cairdcoinheringaahing ooh

Fun fact: The "Al" in "Algebra" and "Alcohol" just means "The"

The Compendious Book on Calculation by Completion and Balancing

The Compendious Book on Calculation by Completion and Balancing (Arabic: كتاب المختصر في حساب الجبر والمقابلة, al-Kitāb al-Mukhtaṣar fī Ḥisāb al-Jabr wal-Muqābalah; Latin: Liber Algebræ et Almucabola), also known as Al-Jabr (الجبر), is an Arabic mathematical treatise on algebra written by the Persian polymath Muḥammad ibn Mūsā al-Khwārizmī around 820 CE while he was in the Abbasid capital of Baghdad, modern-day Iraq. Al-Jabr was a landmark work in the history of mathematics, establishing algebra as an independent discipline, and with the term "algebra" itself derived from Al-Jabr. The Compendious...

Yeah I know

ah

@mousetail In fact, "al-" makes "a" and "l" the most frequent characters in arabic, which was the first breakthrough in breaking the monoalphabetic cipher

@cairdcoinheringaahing I also remember something about the origins of "algorithm"

Somebody named khwarismi

2:25 PM

Exactly

So "The Algorithm" is actually redundant

Huh

It sould be Al Gorithm" or "The Rithm"

so the arabs are the reason of most of the words used in programming today

nice

@UndoneStudios Literally the author of the book I linked

@cairdcoinheringaahing total coincedence

but yes, he was respected

until his death

indeed, had the arabs not been destroyed by europeans in 1924, our society could be even further today

sad

2:32 PM

The arabs invented a huge part of modern math yes

Even things named after other people

Stigler's law

Yeah, pretty sad that they were repayed by being demolished

That wasn't fair at all

@mousetail What do you think the world would be like without them?

Exactly the same

somebody else would've invented it lol

No like today

2:37 PM

They were just the ones to do it first.

in the future, yes, but I'm talking about today

This was 1000 years ago

That's another of their astonishing characteristics: they survived from the 7th century to 1924

A long time

About a millenium and a century years

But I can't remember another empire that lasted that long. Romans?

Which empire are you talking about?

Ottomans?

Yeah

2:49 PM

Those where not the same people who invented math

those came before

oops

But yeah, I don't think we have a name for those

do we?

I don't think there was a single empire or kingdom but many small ones

Perhaps

yawn Why are we talking about this? Would it be on-topic?

On second thought, a lot of borderline of off-topic stuff is on-topic

If it gets in the way of a more on topic discussion these messages will be deposited in OTTNB

3:15 PM

@mousetail ah

I really like your PFP btw

@mousetail thanks

It was created by a Scratch script by messing around with polygons

Which means it's not actually symmetrical at all

It's only like 5 pixels accross so it's really hard to notice

In the middle is a triangle

but it's shadowed by other shapes

so yeah

but yes, it's amazing that it looks like something it is not

Wish it was on Tweetable Mathematical Art

Does it fit in a tweet?

3:21 PM

I guess

It's a screenshot of about 480x360 though so can't tell

SE strips it to become a square

But I don't know... Even a question like TMA without the tweet would still be off-topic

user41805

4:06 PM

@UndoneStudios obligatory, al-khwarizmi wasn't arab, the name literally means one from the region of khwarezm which is more or less in persia

PyGamer0

@UndoneStudios The Bacteria™

Q: Language of the Month for January 2023: Elm

5:29 PM

@PyGamer0 Right

That makes sense

@user41805 Oh

Seggan

6:55 PM

@user41805 persian numerals then?

@UndoneStudios they were repaid for Other Things™

Radvylf Programs

7:11 PM

@UndoneStudios It could be a popcon

Images with all colors is still open

New Posts

7:22 PM

0

Q: Round to nicer numbers

The standard way to round numbers is to choose the nearest whole value, if the initial value is exactly halfway between two values, i.e. there is a tie, then you choose the larger one. However where I work we round in a different way. Everything is measured in powers of two. So wholes, halves, ...

code-golf math number

Radvylf Programs

TIL (and TI want to unL)

Seggan

i knew that for a while

New Posts

8:51 PM

0

In accordance with our meta agreement, since one candidate received more votes than the others, we have a new featured language! Throughout December 2022, our Language of the Month will be: Elm What's a Language of the Month? See the meta post for nominations. In short, during January, those wh...

discussion language-of-the-month

user

9:09 PM

@Seggan According to Wikipedia, they were also spread by Arab merchants, not just Al -Khwarizmi, so calling them Arabic numerals isn't wrong

9:55 PM

Aw, hats are gone

I'm no longer verifiably lyxal™

How will people know I'm the real me now?! :p

Maybe your hats are gone

Well at least I won on the hat count leaderboard here :p

Huh I got fiftth

10:16 PM

I don't think I actually made specific efforts to get any hat, but walked away with 21 hats at #7 on the leaderbboard

I also think I was the only user on the site with the Over 9000! hat

No? Ginger had it

What was it anyway?

Start a discussion post of score 2 or more

@emanresuA Oh yeah, Best Of started

Ah makes sense

@emanresuA l4m2 had it too

Interesting

Pretty sure their meta post was posted before WB started, right? Dec 18th?

10:19 PM

Possibly one of their caching issues

Apparently for some hats they had forgotten to clear testing data from before the event started