English Language & Usage: Multi-Layered Discourse Room

jlliagre

00:00

Daily Octordle #1121
🕚9️⃣
4️⃣🕛
🔟6️⃣
3️⃣7️⃣
Score: 62

Daily Sequence Octordle #1121
4️⃣5️⃣
6️⃣7️⃣
8️⃣9️⃣
🔟🕚
Score: 60

Daily Extreme Octordle #1121
3️⃣9️⃣
5️⃣🔟
6️⃣🕛
4️⃣7️⃣
Score: 56

Xanne

Wordle 1,341 6/6

🟨🟨⬜⬜⬜
⬜⬜⬜⬜⬜
⬜⬜🟨🟨🟩
🟨🟩🟨⬜🟩
🟩🟩🟨⬜🟩
🟩🟩🟩🟩🟩

CowperKettle

00:58

CowperKettle

03:24

The model, now published in Nature Human Behaviour, demonstrates that people focus not by concentrating extra hard on a subject, but by ignoring inputs that were distracting in the past.

medicalxpress.com/news/…

2

Robusto

05:53

> Ultimately, AI-powered surveillance technology could result in the creation of total surveillance regimes that monitor citizens around the clock and facilitate new kinds of ubiquitous and automated totalitarian repression. A case in point: Iran’s hijab laws.

Harari, Yuval Noah. Nexus: A Brief History of Information Networks from the Stone Age to AI (p. 245). Random House Publishing Group. Kindle Edition.

> After Iran became an Islamic theocracy in 1979, the new regime made it compulsory for women to wear the hijab. But the Iranian morality police found it difficult to enforce this rule. They couldn’t place a police officer on every street corner, and public confrontations with women who went unveiled occasionally aroused resistance and resentment. In 2022, Iran relegated much of the job of enforcing the hijab laws to a countrywide system of facial recognition algorithms that relentlessly monitor both physical spaces and online environments.[30] A top Iranian official explained that the syst…

> Shortly afterward, on September 16, 2022, the 22-year-old Mahsa Amini died in the custody of Iran’s morality police, after being arrested for not wearing her hijab properly.[32] A wave of protests erupted, known as the “Woman, Life, Freedom” movement. Hundreds of thousands of women and girls removed their headscarves, and some publicly burned their hijabs, and danced around the bonfires. To clamp down on the protests, Iranian authorities once again turned to their AI surveillance system, which relies on facial recognition software, geolocation, analysis of web traffic, and preexisting dat…

> In September 2023, on the anniversary of Mahsa Amini’s death, Iran’s parliament passed a new and stricter hijab bill. According to the new law, women who fail to wear the hijab can be punished by heavy fines and up to ten years in prison.

Harari, Yuval Noah. Nexus: A Brief History of Information Networks from the Stone Age to AI (p. 247). Random House Publishing Group. Kindle Edition.

@M.A.R. ^ I wasn't aware your country had that kind of Orwellian surveillance.

Michael Rybkin

08:05

@Lambie Thank you.

M.A.R.

08:55

@Robusto the reformist (less extreme) and principalist (totally nutty hyperreligious freaks) parties don't act all that different, but the perception is when reformists win, people want less totalitarian zealotry. This is their version of filibustering and blocking what reformists want to achieve. They would want to focus the government on the economy, which would eventually entail that we need a change in foreign policy or we're screwed.

Unfortunately, zealotry is approved by the Supreme leader and cemented in the constitution, so they turn to passing bills like this where they're unbeatable. No one, not the people nor the opposing party can object because the narrative is shaped in a way that objection means rejection of Islamic virtues.

IOW it's the religious freaks taking their revenge.

Yesterday I was applying for a "fuel card" where it'd allow me to buy cheaper subsidized gasoline and there was someone there who had to impound their car only because of "kashfe hejab", meaning the cameras caught a woman without hijab in his car.

sigh

Well it's a tug of war and the idiots don't realize the rope can snap too.

GratefulDisciple

10:47

@M.A.R. and @Robusto When I was living in Jakarta (not Aceh, the only province where Islam rules as a theocracy), my perception was that the women who wore hijab (jilbab in Indonesian, and it didn't cover the face, like this) were signaling virtue out of their free will, similar to Catholics wearing a cross hanging from her necklace. Turns out I lived during the relaxation period from Suharto hijab-prohibition era from 1965 to 1985.

But after I left, it appears that Iran-style rationale (as morality requirement imposed from above) infiltrated the more "fundamentalish" movements (I borrowed the term from Roger Olson).

This 2022 article How Indonesian Islamic politics shape mandatory hijab rules and uniform policies in schools has a link to the exact Suharto era document and the analysis rings true. So, thanks, Iran!

In Canada I often see many Sikh men wearing headdress, like the NDP party current leader. Now that makes me wonder whether that's compulsory or not.

SmokeDetector

[ SmokeDetector | MS ] Potentially bad keyword in answer (1): What is the name of the truck shown in this picture?‭ by Mr.Siripong Siripin‭ on english.SE

GratefulDisciple

@Cerberus Today's Connection is one of the easiest. I wonder whether you find the words to be American slangs or not.

Connections
Puzzle #619
🟨🟨🟨🟨
🟩🟩🟩🟩
🟦🟦🟦🟦
🟪🟪🟪🟪

SmokeDetector

[ SmokeDetector | MS ] Potentially bad keyword in answer, blacklisted user (77): What is the name of the truck shown in this picture?‭ by Mr.Siripong Siripin‭ on english.SE

GratefulDisciple

@CowperKettle Rings true for me. I have a very hard time ignoring background music which either engrosses me (classical music) or irritates me (like techno), so I prefer to do my work (especially when coding) in silence. For me, there's no such thing as ignorable "background music".

GratefulDisciple

11:39

@Mitch I was more hung up with OS features than CPU improvements, although I was too young to care about API in Windows 3.1, but I remembered being super excited of the better FAT32 file system that came out with Windows 95 OSR2 and when I started using Windows 2000 at work that comes with more reliable NTFS, there's no going back to non-preemptive multitasking OS despite the growing pains of using NT before Windows XP came out that solved all compatibility issues.

@Robusto Only now do I feel computer ethics to be necessary. But it's now even harder to predict whether the technology I'm working on (should I work in the AI field) will be used for reducing human freedom. I wonder how ethicists navigate through proper R&D for dual-purpose technology (which also includes GPS, satellite, nuclear, or CRISPR).

Vikas

12:09

I have two questions

1. If your uncle is of similar age to you, would you still call him "uncle" or use his real name only instead?
2. If your uncle is younger than you, would you still call him "uncle" or use his real name only instead?

Bonus question: What do you call your father's cousin?

Bonus question 2: If G1 is your paternal grandfather and C1 is his cousin. What would you call C1's son?

4 questions actually.

Note that I'm more interested in casual language (the terms you'd actually use in real life) rather than formal terms.

GratefulDisciple

12:37

@Vikas I don't think this is a language question, but a cultural question. Among Chinese who adhere to some Confucian conventions (like my extended family), we resolve it biologically. So yes, one relative of a similar age with me calls me "uncle".

But if we both speak Chinese, we would use the much more elaborate terminologies that even distinguish whether I'm his uncle from his mom side or his dad side. I myself am vague on this as, my parents were Westernized and didn't pass this tradition to me. My more traditional Chinese Indonesian friends scolded me harshly from not knowing this elaborate scheme of calling a relative appropriately.

HippoSawrUs

13:28

I'm getting the World Book Encyclodedia 2025 set. I want to say I've read it all, before the world ends, and be superficially topically brilliant and too busy to notice the latter. Questions: 1) Why is it the 2025 ed. if published December 1, 2024? It's not a Ford Explorer. 2) What if the world had evaporated on December 20, 2024, and only one volume (3, C-Ch) survived. Would aliens assume we were obsessed with car or church redirects, in heaven, or what exactly?

Robusto

14:05

@GratefulDisciple It's compulsory for Siihs.

@Vikas 1 & 2: Matter of choice or family tradition. As for the rest, that would be a second cousin, twice removed.

Robusto

14:41

#WhenTaken #358 (19.02.2025)

I scored 768/1000🏅

1️⃣📍12.5K km - 🗓️1 yrs - 🥉99/200
2️⃣📍5.9K km - 🗓️12 yrs - 🥉95/200
3️⃣📍937 m - 🗓️4 yrs - 🥇196/200
4️⃣📍733 m - 🗓️2 yrs - 🥇198/200
5️⃣📍332 km - 🗓️7 yrs - 🥇180/200

https://whentaken.com

Wordle 1,341 4/6

⬛⬛⬛⬛⬛
🟨⬛⬛⬛🟨
🟨⬛🟨⬛🟩
🟩🟩🟩🟩🟩

alphabet

@Lambie You're absolutely right on that "oblique accusation" answer. Upvoted. The fact that you can't find tons of uses of some adjective + noun pair doesn't make it unidiomatic or wrong.

The only answer you need is just: "yep, the dictionary confirms that those words mean what you think."

Sure, oblique is a high-register word and I'd guess that plenty of native speakers don't know it, but it's not so obscure that one would have any reason to avoid it in formal writing.

jlliagre

15:04

#WhenTaken #358 (19.02.2025)

I scored 787/1000🏅

1️⃣📍679 km - 🗓️8 yrs - 🥈169/200
2️⃣📍2.2K km - 🗓️8 yrs - 🥈137/200
3️⃣📍8.0K km - 🗓️4 yrs - 🥉104/200
4️⃣📍451 km - 🗓️0 yrs - 🥇186/200
5️⃣📍496 m - 🗓️7 yrs - 🥇191/200

https://whentaken.com

Wordle 1,341 4/6

⬛⬛⬛⬛⬛
⬛⬛⬛🟨🟨
🟨⬛⬛⬛🟩
🟩🟩🟩🟩🟩

Daily Octordle #1122
8️⃣7️⃣
6️⃣4️⃣
🔟9️⃣
2️⃣5️⃣
Score: 51

Daily Sequence Octordle #1122
5️⃣7️⃣
8️⃣9️⃣
🔟🕚
🕛🕐
Score: 75

Daily Extreme Octordle #1122
5️⃣3️⃣
🟥6️⃣
🔟🕛
7️⃣9️⃣
Score: 65

Mitch

15:28

@HippoSawrUs The World Book Encyclopedia? It's pretty conventional up until I, but the J and K it really turns around and becomes a whole nother thing. No spoiler but S is not what it seems.

@GratefulDisciple it's a sociolinguistic question. The phenomena are contingent on both language and cultural realities.

The culture encouraged words to exist for the social concepts, but nothing is necessary (ie words may exist for non-existent cat gories, and concepts may be thought of without single words or multiword phrases to capture them)

@alphabet I still haven't been able to find a link to a question that talks about 'oblique accusation'. Where does d this?

@Vikas in US and English, you might tell others that this person is an uncle of yours, but in conversing with him, you'd usually only ever say 'Bob' and not 'Uncle Bob'

#2 same

And frankly it feels like even for someone aged your parents age, you might get away with not using 'Uncle'

@Vikas any male that is your father's age and related somehow (even just a good friend) can be and is often called 'Uncle ---'

Lambie

@alphabet Ah, alphabet, you provide relief to me. :) I get tired of fighting uphill battles.

Mitch

Also these may slowly fade away as you get older (the callee becoming a child no longer)

@Vikas as Rob said, something like first cousin once removed. You might call them 'Uncle ---' to their face.

Casually, Uncle --- or just their name depending. And again that's for kids. As you get older just their name.

In US English, kinship names are very sparse and aren't very strict.st

There are some families where the parents are called by their children by their first names... this feels very very intellectual to me and sounds really wrong. But some people do it.

Robusto

16:00

@Mitch Well, now, SPOILER

Mitch

@Robusto To be honest, if honesty is indeed the relevant sentiment, it always seemed strange to read through the dictionary or encyclopedia linearly. If you get tired of the story by B, you're missing quite a lot.

I mean aardvark followed by aardwolf is not a compelling start.

Robusto

@Mitch Honestly, this is something I'd never do. I'm annoyed by so many things in this world that I don't have time to do that to myself.

Mitch

Really, start with a car chase.

@Robusto You should organize and curate your annoyances so that you can apply your energies more efficiently.

Some people think that starting a dictionary or encyclopedia with a car chase is a facile mixing of genres.

Robusto

@Mitch Doesn't AI do that yet?

Mitch

But those people don't have a mansion on Mulholland Dr and a Swiss chalet and a Highland hunting lodge.

@Robusto You don't know what my opinion is of Swiss chalets?

@Robusto scribbles on back of napkin

sends photo to VC

puts bid on Swiss chalet

My personal assessment of Swiss chalets is that...

Better steak in a chalet than gristle in a a castle.

Robusto

16:14

@Mitch That's a lot of climbing on a bike.

Mitch

@Robusto Pfft...take a bus to the top and coast all the way down.

Robusto

@Mitch Where does chateaubriand come in?

GratefulDisciple

@HippoSawrUs Why not see whether your local public library offer access to the online version? I just checked my local library and compare the online version to the sample page on Venezuela. Every element is available in the online version, plus the latter have links to other articles and to an audio of Venezuela national anthem.

Mitch

@Robusto I think he is mentioned several times before he actually appears in Act IV.

GratefulDisciple

@Robusto That's too bad. Thankfully I was not born a Sikh.

Mitch

16:16

@GratefulDisciple Sikhs probably thank themselves that they weren't born otherwise.

Or their mom.

Robusto

@Mitch But then he suddenly appears out of nowhere.

With a knife.

And a fork.

Mitch

I mean you should always thank your mom what ever the situation.

GratefulDisciple

@Mitch That's true.

Mitch

@Robusto To be honest, if that is indeed the sentiment I am going for, I don't know what steak Chateaubriand is.

Robusto

@Mitch I could google that for you, but I'm too lazy atm.

GratefulDisciple

16:18

@Mitch When I hear Swiss chalet my mind goes here.

Mitch

I suspect it is, like almost the entirety of French cuisine, assuming by the spelling that it is indeed of French origin, taking some old gross half spoiled food cooking it with a lot of weird spices, and throwing on a garnish to make it look fancy.

@Robusto Same.

GratefulDisciple

@Mitch I like the British practice of calling a student (high school and above) by their last name. Teaches them dignity.

Mitch

Chateaubriand (dish)

Chateaubriand (French pronunciation: [ʃatobʁijɑ̃] ; sometimes called chateaubriand steak) is a dish that traditionally consists of a large front cut fillet of tenderloin grilled between two lesser pieces of meat that are discarded after cooking. While the term originally referred to the preparation of the dish, Auguste Escoffier named the specific front cut of the tenderloin the Chateaubriand. In gastronomy of the 19th century, the steak for Chateaubriand could be cut from the sirloin, and served with a reduced sauce named Chateaubriand sauce or a similar, that was prepared with white wine and...

There. The temptation overrode my sense of energy savings.

Robusto

It's winter. You need to save every erg of energy you can.

Mitch

I've read it, but I still don't know how it is different from other steaks.

It's all just steak right?

Some poor cow is a couple pounds lighter and a little anemic.

What? The whole cow, you say?

That doesn't seem right.

I mean

I mean...

GratefulDisciple

16:24

@Mitch Yes, for some reason being unthankful to mom feels a lot worse than being the same to dad, though both my parents worked equally hard.

Mitch

I mean what happens to the cow?

@GratefulDisciple The whole mom-dad thing is weird.

I mean your dad is just ... there, but your mom is totally involved with you. -But- when you dad dies, everybody is wailing and crying and the world has fallen apart but when you mom does it's like it was totally expected and well that's the end of a generation, what next?

And here you all came on a Wednesday to get all cheered up about the state of the world.

Always remember, no matter how bad things are, everything is falling apart around you, people are losing their minds, no one is acting like an adult, bad luck after bad luck happens, always remember it could get worse.

I mean it could be raining.

And if it is already raining, it could rain harder.

Robusto

@Mitch Apparently they cook it between two lesser cuts and then throw those away after cooking. Sounds kinda wasteful to me. Also disrespectful. Are there cattle that only have "lesser cuts"? Think of the shame they must endure walking around the feed lot waiting for "harvesting".

Mitch

And if it is raining harder, you could realize that as you went outside in the extra hard driving rain in your pajamas to take out the trash, as the door slowly swings shut to a click, that nobody is home and your keys are in your other pajamas.

and no one else is at home.

@Robusto You get it.

The cattle don't however. I think the cattle owners don't tell the cattle about their impending weight loss. It's kind of a shock to the cattle frankly.

Literally. That's how they off the cattle in prep for slaughter. A shock gun to the forehead.

GratefulDisciple

@Mitch Spoken like a true Stoic. I "fact check" this with ChatGPT prompting it this way:

> Which philosophy / religion would most agree with "Be cheerful because it could have been worse"? For example, if it is already raining, it could rain harder.

Robusto

@Mitch The shock doesn't last long. Remember Anton Chigur and his piston-driven cattle snuffer from No Country For Old Men?

Mitch

16:34

hmm...I may be mixing things up here. A cattle prod is simply for 'crowd control' and is a literal shock mechanism. But they off the animal with a single direct shot to the head.

GratefulDisciple

It then listed 6 possible answers and concluded:

CowperKettle

Physics of the day (new to me): Majorana fermion - "Microsoft isn’t using electrons for the compute in this new chip; it’s using the Majorana particle that theoretical physicist Ettore Majorana described in 1937."

Majorana 1 Explained: The Path to a Million Qubits

►

GratefulDisciple

> In summary, Stoicism is perhaps the most directly aligned with the "it could have been worse" mentality, but Buddhism, Taoism, Christianity, Sufism, and Hinduism all have elements of gratitude, acceptance, and resilience that fit the idea.

Mitch

@Robusto ugh yes. If I were a cow, I'd be glad not to know about that.

@GratefulDisciple It seems a little spoiled to go the other way, to exclaim that it could have been better.

GratefulDisciple

What's scary is that when I fact check the "fact check", essentially ChatGPT is correct.

Mitch

16:37

like 'this toilet is pretty nice with its automatic flushing and programmable music, but if only it were made of gold'

Robusto

I think gold must be cold.

Vikas

@Mitch I haven't heard this phrase until today: first cousin once removed

Sounds very technical lol

GratefulDisciple

@Mitch A Korean / Japanese would say that it's nice it's made of gold, etc. but I wish there is a seat warmer.

Mitch

@GratefulDisciple LLMs aren't the internet or the entirety of knowledge (despite what we expect of it) but often the output is not bad.

@Robusto Are you saying it could better?

@Vikas It is very technical

Robusto

@Mitch I'm saying it cold be better.

Mitch

16:40

In normal conversation you'd say "This is my cousin Gerald. Or really the son of my cousin. Technically my first cousin once removed"

@Robusto +1

GratefulDisciple

@Mitch So I prompted ChatGPT: What about a philosophy / religion that exlaims "it could have been better"? (BTW, I didn't realize I had a misspelling but ChatGPT didn't bat an eyelid). The answer gave me 6 options (each with a sample phrase) prefaced with "A philosophy or religion that emphasizes "It could have been better" often has an element of dissatisfaction, striving for improvement, or lamenting missed opportunities." and ends with this summary:

> - If the focus is on missed opportunities, then Existentialism fits.
- If it's about social or structural improvements, Confucianism applies.
- If it's about personal ambition, Western capitalism & self-help make sense.
- If it’s about idealizing a more beautiful reality, Romanticism is a match.
- If it's about deep cosmic dissatisfaction, Gnosticism is relevant.
- If it's about desire and suffering, Buddhism acknowledges this mindset but seeks to move past it.

Again, scarily quite accurate.

Mitch

@GratefulDisciple There's a difference in feeling between "Mr. Johnson, fifty pushups please" and "Johnson, fifty pushups please"

There was a trend when I was a kid for kids to call each other by last name and it always seemed very derogatory.

GratefulDisciple

@Mitch Yes, it only works when the teacher does it. Sometimes I imagine what it feels like to go to a high-brow British private school like Eton.

Mitch

@GratefulDisciple I think kids are jerks everywhere, entitled rich kids or impoverished thugs.

There are nice kids of course, but they are beaten into submission by the jerks.

GratefulDisciple

@Mitch Or a military drill sergeant yelling "Johnson! Drop and give me fifty—NOW" (courtesy of ChatGPT).

Mitch

16:50

@GratefulDisciple It's pretty amazing that probabilistically predicting the next most likely word can produce sequences that can be interpreted as making sense. and even moreso, making cogent arguments or descriptions that have never been explicitly made before.

@GratefulDisciple Which is probably what that boring history teacher is trying to evoke when he yells at you to turn in your exam 30 seconds past the deadline.

GratefulDisciple

@Mitch Yes, which really piqued my interest to learn how generative LLM work and to know theoretically how educated human can beat them. Because if we cannot, then I would have to start over in my philosophizing, like @Robusto was saying: Sisyphus starts over when the boulder goes back all the way down.

@Mitch Yeah, I can relate, being in the middle of furiously writing down your thoughts after spending 15 minutes brainstorming and constructing your wonderful history essay, wishing you take the exam digitally (I can write using keyboard 2-5 times faster than handwriting).

Mitch

@GratefulDisciple random words scattered on a page aren't likely to make sense, but sometimes, because we are so attuned to making patterns out of very little data, we see Bayesian Ghosts (thank you @CowperKettle), patterns that just aren't there.

and with a little statistics, the random words can be made to be a little less random, but we are still reading a lot between the lines, bridging very large gaps to make sense.

GratefulDisciple

@Mitch If you're a Platonist you wouldn't be amazed since it's an act of recollecting. If you're an Aristotelian, you would thank your active intellect acting upon your store of phantasms from decades of experience. And if you're ChatGPT it's because of the Internet-full of verbal representations of billions of humans typing their mental contents.

Mitch

less and less randomness, means it can mostly make sense of it, with very small leaps, but the structure that is imposed by the statistics is not human-like intelligence.

"billions of humans typing their mental contents" - in a manner of speaking yes, but there's a lot of qualification going on.

Language is an attempt at ESP, transferring thoughts. But if all you have is the language part (just the text) there is no grounding in reality or in, what the philosophers like to say, qualia (inner senses/feeling).

The statistics might capture some of that, might model some of the grounding (there are all sorts of papers about how an LLM might 'have' a model of spatial structure or color similarities, whatever 'have' means), but 'capture some' is doing a lot of work.

GratefulDisciple

(sorry on a phone call now, will reply later)

Mitch

17:04

More importantly, I think there should be a concept called "'Buy Us' bias".

I don't know what the bias is but I'm looking forward to people suggesting I have a stutter.

GratefulDisciple

17:18

@Mitch I'm pretty sure ChatGPT is fed with 1) human constructed "output templates" paired with many "types of queries" as well as 2) "ontological categories" and "essential connections in reality" onto which words are being mapped and maybe even 3) books like Jordan Peterson's Maps of Meaning and WorldBook Encylopedia to structure reality as humans see it. And the statistics are second level stuff.

@Mitch So that's how AI programmers remedy the lack of LLM's possessing the faculty for sensing / intuiting qualia.

So I would see that an AI programmer genius is probably similar to Mr. Frankenstein engineering his monster from simulated parts of the human mind (not brain).

CowperKettle

I tried reading up on the Majorana fermion, and understood totally nothing.

If it's a kind of neutrino, how on earth can it be measured and manipulated in a small chip.

Mitch

@GratefulDisciple as to 1) LLMs (to avoid thinking of ChatGPT as the only one) are not trained like that. They just take gobs and gobs of text and compute interword probabilities (with some sequence properties thrown in). I'm not sure what you mean by 'templating' but if you mean anything with variable slots in them, they are definitely not doing that.

GratefulDisciple

@Mitch (sorry, didn't quite follow you there on "Buy Us' bias" and stuttering)

@Mitch Templating is just the structure of the output. It's quite consistent: first, summarize the question, then provide the body of the answer, then summarizing, then ask how it can help further. For questions like I just posed, it would be itemized list. For others it's a generative text in a particular style. For some others, it would be compare / contrast. For yet another, it would be a math proof or a computer code.

Mitch

as to 2) there may well be, in their training data, such books that contain in narrative form (or in outline) some kind of upper level concept ontology (kind of like Roget's original table of contents), but that is not part of the architecture of the system, it is only training data. The architecture is not calculating using some kinds of formal definitions of words, it is only using nearby word correlations.

as to 3) yeah that's not at all how LLMs work. LLMs do not operate on logical semantic representations of the sentences it scans over. It's all interword correlations.

GratefulDisciple

@Mitch What I'm proposing is that the training data is on the second level after the ontology is input manually. Of course this is just conjecture. I'll have to see ChatGPT model training code (isn't it open source)?

Mitch

17:28

@GratefulDisciple The way I pronounce 'buy us' and 'bias' they're almost the same thing.

GratefulDisciple

@Mitch oh okay. But isn't the manually-input ontology is precisely what's need to avoid confusing "buy us" and "bias"?

@Mitch You maybe right (which then makes me all the more amazed) but my gut instinct tells me it's a both-and.

Mitch

@GratefulDisciple Oh... that kind of 'templating' is caused by two things 1) when you make a request on an intellectual topic, that sequence of words is often followed (presumably) in the training data by essay-like answers (like a student or encyclopedia writer) may write. and they tend to have that kind of form you describe. That's why you see that kind of common pattern.

2) RLHF (Reinforcement Learning with Human Feedback) is a method to 'fix up' or rather to 'pick the best of multiple alternative responses' that most of the LLMs use (this involve having humans rate many possible responses and rank them and uses these rankings to train a much small model.

GratefulDisciple

@Mitch I have a feeling they have multiple LLMs. One model is used for selecting the kind of query, that is then paired with a standard "essay" template as its most probably desired output format. Another model(s) are then consulted for the meat of the essay itself. Those models are trained on the genre of essays.

Mitch

There's a third thing 3) most LLM products nowadays tack on to your prompt some extra instructions that help steer the response towards more wikipedia like answers, probably even with words like 'first summarize...'

GratefulDisciple

Now I wonder WHY and HOW my own mind can auto-generate essays when a teacher prompts me for one 😃. Can I BE a better AI programmer by introspection?

Mitch

17:37

@GratefulDisciple There is no first level/ontology or manual input of some 'concept organization'. It's all just training data (gobs and gobs of text).

GratefulDisciple

@Mitch You keep saying that and you maybe right. Now I'm more pumped up than ever to study machine learning. But this week I need to finish a project. Really nice chat, human Mitch (despite your cat avatar), hope you have a good rest of the week as a human.

Mitch

@GratefulDisciple The whole "buy us bias" is not serious and is entirely distinct from the LLM discussion.

@GratefulDisciple all the current LLMs follow or are derived from the "Attention Is All You Need" paper.

You may conceptually feel like it's 'both-and' (whatever that means), but there's no ontology first step.

Of course, there is initial step to all LLMs, the word2vec step that takes the text sequences and converts the token into word vectors, and the word vectors are created to be 'nearby' semantically.

That might be considered to be sort of kind of like a very very poor ontology.

and in some sense may be a sufficient replacement for the formal ontology you're thinking of.

and is a very important part of an LLM.

But is not usually considered a defining characteristic of these LLMs.

@GratefulDisciple The big commercial LLMs (ChatGPT, Claude, Gemini, Mistral, DeepSeek, etc etc etc) are not done like that at all. It is just interword correlations. The training data just has lots of examples that tend to follow the kind of format that you expect.

But surely many smaller companies are trying to create new LLMs from scratch (or using pieces of other LLMs) that try to do what you are suggesting, one LLM to do named entity recognition, another one to do 'analysis' (not actually do analysis but to construct text that looks like analysis), etc

BU I don't think that is a profitable company to do that for essay writing for academic use. No one is spending money to do that. If you come across a company that claims to be doing that, then they are just using ChatGPT or Claude as a backend to process queries that have their own special context appended. I wouldn't be surprised if Grammarly does this already.

@GratefulDisciple "I wonder WHY and HOW my own mind can auto-generate essays when a teacher prompts me for one": That is a common and reasonable analogy to make, because the results seem hard to distinguish. The two processes are very different though.

"Can I BE a better AI programmer by introspection?" 1) all programming is introspection. If by AI programming you don't mean coding a neural network but instead writing a prompt better, then sure introspection helps a lot with writing any kind of text, but how -your- mind works is not how the LLM works. It's just interword correlations - you'll get likely following words/sentences/passages based on the training data.

Aug 30, 2018 at 20:48, by Mitch

I am not a cat

Also I am not a cat with a cat in a box in a box.

GratefulDisciple

18:27

@Mitch Thank you for your feedback on how products like ChatGPT work. Maybe by next year after reading AI journal articles I can report back to you on how the best generative-AI products really work under the hood (I'm sure there are some trade secrets), and how exactly they created and trained the LLMs associated with those products.

@Mitch For the first time I notice the math formula on the box. Quite cool.

Robusto

@Mitch This is supposedly what someone said was happening with Google's quantum computer, Willow.

alphabet

@Mitch It's on ELL: ell.stackexchange.com/questions/362243/…

@GratefulDisciple No! Bad ChatGPT! The Stoic response (to the "rain" example) would be "rain isn't actually bad; if you think it is, you're a slave to rain-aversion."

Robusto

18:48

Connections
Puzzle #619
🟨🟨🟨🟨
🟩🟩🟩🟩
🟦🟦🟦🟦
🟪🟪🟪🟪
.

MetaEd

@GratefulDisciple - also -- scholars are considering the idea that Taoism is Buddhism, historically speaking. Also, Stoicism is an answer to Pyrrhonism, and Pyrrhonism is Buddhism.

Mitch

@alphabet whew...thank you. now I know what the brouhaha was.

kerfuffle

donnybrook

melee

ado

thing

Robusto

@Mitch I think it was more of an apocalypse. Nobody will survive it.

@MetaEd And the whole thing is a Viconian cycle?

Mitch

@GratefulDisciple In a suitable logical language (namely modal logic) it is the statement of Lob's Theorem:

Löb's theorem

In mathematical logic, Löb's theorem states that in Peano arithmetic (PA) (or any formal system including PA), for any formula P, if it is provable in PA that "if P is provable in PA then P is true", then P is provable in PA. If Prov(P) means that the formula P is provable, we may express this more formally as If P A ⊢ P r o v ( P ) → P {\displaystyle {\mathit...

or rather it is the antecedent of Löb's theorem, where the consequent is box(cat)

or a cat in a box

TRiG

imgur.com/gallery/white-whale-of-sales-FBcZu96 This post writes kind of as kind've, which is a beautiful example of hypercorrection.

Mitch

18:54

all I'm saying is that cat's love boxes.

and sitting in them.

@TRiG Who'd've thunk it!

TRiG

@Mitch My brain skipped straight over it till I saw a comment pointing it out.

alphabet

@MetaEd The claim that Buddhism influenced Pyrrho is very controversial. (I'd say any influence was probably minimal and this is mostly Diogenes Laertius' usual BS.)

Robusto

@Mitch I only do Piano arithmetic.

Mitch

@Robusto foofaraw

furor

furore

furor é

ballyhoo

MetaEd

19:23

@alphabet Pyrrho was literally there in Gandhara only 40 to 150 years after the Buddha's death, and brought back with him several clearly identifiable Buddhist concepts, notably the three marks of existence.

alphabet

19:37

@MetaEd The Pyrrhonian doctrine (which, granted, we only know from a quote of a quote of a quote) bears at most a superficial resemblance to the "three marks of existence" and he applies that doctrine completely differently, to defend a philosophical project that has no clear relationship to Buddhism.

It's impossible to rule out any influence of Buddhism on Pyrrho, but there simply isn't good evidence for it.

Xanne

@Mitch If I had a huge amount of digitized text on an historical figure and the history of the times, could I then generate a story of several chapters that illustrated the main themes of the person’s life and accomplishments?

@Mitch I assume the answer is no, but I’d like to know why :-)

MetaEd

19:56

Wordle 1,341 4/6

🟨⬛⬛⬛⬛
🟨🟩⬛⬛⬛
⬛🟩🟨🟨🟩
🟩🟩🟩🟩🟩

@alphabet what philosophical project is that?

Robusto

Daily Octordle #1122
3️⃣🕚
🔟6️⃣
7️⃣4️⃣
8️⃣9️⃣
Score: 58

Daily Sequence Octordle #1122
4️⃣5️⃣
7️⃣8️⃣
9️⃣🔟
🕚🕛
Score: 66

Daily Extreme Octordle #1122
3️⃣4️⃣
9️⃣7️⃣
6️⃣🔟
5️⃣🕚
Score: 55

MetaEd

Strands #353
“Pick your own prefix”
🔵🔵🔵🔵
🟡🔵🔵🔵
🔵

@Robusto I hope not. I wasn't familiar with this, but it seems this guy Vico thinks numbers are real

Robusto

20:18

@MetaEd It's a rookie move.

Mitch

@Xanne There's a lot of uncertainty in the qualifiers in that question and in the current reality. 'huge amount...on' 'historical', 'history of the times', 'story', 'several', 'main'. But reading between the lines, I think yes it is reasonably possible with current technology.

Caveats:

MetaEd

Connections
Puzzle #619
🟪🟪🟪🟪
🟨🟨🟨🟨
🟦🟦🟦🟦
🟩🟩🟩🟩

Xanne

@Mitch awaiting caveats

MetaEd

Dick Cavett

Richard Alva Cavett (; born November 19, 1936) is an American television personality and former talk show host. He appeared regularly on nationally broadcast television in the United States from the 1960s through the 2000s. In later years, Cavett has written an online column for The New York Times, promoted DVDs of his former shows as well as a book of his Times columns, and hosted replays of his TV interviews with Bette Davis, Lucille Ball, Salvador Dalí, Lee Marvin, Groucho Marx, Katharine Hepburn, Judy Garland, Marlon Brando, Orson Welles, Woody Allen, Ingmar Bergman, Jean-Luc Godard, Robert...

Mitch

@Xanne sorry, looking up values

ok that's a rabbit hole for another day

Xanne

20:35

@Mitch I would think the main points would be extracted frm previous writings, or specific ones.

Mitch

- huge amount of text plus relevant contextual history: LLMs work better the more data they have trained on and that means in a sense as much as possible outside of your question. so if you want any kind of biography on George Washington, you probably want to train on all history and frankly non-history too. it helps with getting the language right, how to organize paragraphs and longer patterns. so I'll basically answer with 'let's just use the latest ChatGPT/Claude etc'

Xanne

@Mitch So, a master’s essay on, say, the Aztec Empire might just extract from published works, if they’re included in the data base.

Mitch

also I'm guessing you don't want a fictional type story, but rather something closer to factual/biographical.

Xanne

@Mitch Yes. not fiction. But book texts may not be included in the training data—or do they use Google Books?

Mitch

@Xanne yes, and by 'extract' not select a bunch of quotes verbatim, but really throw all that stuff in a blender, and probabilistically out put a narrative.

As to book length, that would seem difficult at this point, but a wikipedia style article is well within the scale of possibility at this moment.

It would contain -lots- of factual errors, but for the most part be pretty good.

Xanne

20:42

@Mitch So if you wanted copyrighted books, you’d have to feed them in yourself from a scanner?

Maybe not. I guess Google Books includes material under copyright.

Mitch

The more obscure the topic, the more likely there will be factual errors (lack of consistency, law of large numbers says variance will be higher on smaller n), and -also more likely for there to be data leakage, actual verbatim copies of original works (because there's very little text to get probabilities from)

@Xanne All the main LLM manufacturers have been not forthcoming about the contents of their training data, and one of the reasons -might- be because the trained on copyrighted data and that -might- incur them some legal action.

So you're imagining someone trying to do this (write a biography) who wants to create an LLM from scratch -and- has their own data?

If so...

Xanne

@Mitch I know someone who thinks he would like to try it.

Mitch

There are techniques that allow you to do this without having to do the $50million training cost that takes 2 1/2 months... you can use a technique called RAG (Retrieval Augmented Generation) where you submit into prompt your request "Write me a bibliography of Montezuma" appended with all the very specific documents that you care about, and the LLM will use all that data to spit out something that looks biographical on (mostly) just that topic.

The current limitations for Claude are I think 10k tokens for the whole prompt...is ChatGPT already up to 32k?

My google search was telling me all sorts of different numbers that weren't consistent (because of dates?).

1k is roughly 1 book page.

So at this moment it is totally possible to spit out a wikipedia like text (if you give it a list of facts that add up to about 10-30 pages worth).

Xanne

@Mitch That’s for the prompt, not the added info you provide to the already-existing LLM?

@Mitch Or you can only add about 30 book pages?

Mitch

@Xanne To clarify, if you use ChatGPT or Claude there's no 'adding training data' to them, you can only make a prompt. But you're adding to the prompt all that extra text.

@Xanne Roughly? Also, using the current interface of ChatGPT/Claude I'm pretty sure they both consider the output response part of the context window (this 32k size).

Xanne

20:58

@Mitch So RAG takes a prompt of only 32 pages?

Mitch

So "Write me a wikipedia like page about Montezuma" would be a prompt that just uses the big statistical stew of neural network coefficients to write it...

Xanne

No added data, e.g., 20 scanned books?

Mitch

and "Write me a wikipedia page about montezuma from the following text: 'Monetuma was born a poor Aztec kid in a poor Aztec village in the mountains of ...." (of all the text you want). The LLM will continue this prompt and in so doing will most likely use some of that text (but may well use some text created from the neural network coefficients (that are not part of the prompt).

Xanne

I don’t think this is going to put current historians out of business any time soon.

Mitch

One book is likely ~250 pages or 250k tokens... which is kind of a lot and I don't think research LLMs are at that large a content width yet.

@Xanne I mean you can sorta get around these scale limitations by doing things piece by piece.

Xanne

21:07

@Mitch Thank you very much. I think I’ve got it.

Mitch

of course all these kinds of compositions will suck,even if you could ingest 20 book in a prompt.

too many errors to check, the style will be nauseatingly bland and sycophantic, and no self-respecting historian would want to read it.

Xanne

Or college professor.

Mitch

and the latter is probably going to be the case for all such 'autoregressive' LLMs forever, no matter how much processing or training data is collected.

HippoSawrUs

@GratefulDisciple I really just want to smell them, regularly. The last Venezuelan element I saw was Font at Rick's Lounge, circa 1989, and I gotta tell you: It didn't smell like a World Book Encyclopedia.

But I digress…

Mitch

-every- college/high school student who is asked to write a paper will want to be able to compose like this.

I suppose professional academics are being squeezed so hard right now for producing things, that they might feel compelled to use LLMs even just to help out.

Xanne

21:12

@Mitch Copilot is a true sycophant.

Mitch

There's been a glut of submissions to academic journals that have been shown to be ...how shall we say... produced by automation.

@Xanne I haven't used Copilot... I've only tried small things right in ChatGPT/Claude and they worked OK.

Xanne

@Mitch There were all those Alzheimer's papers that reused slides. But that was just fraud.

Mitch

It doesn't feel too different from 'coding with StackExchange'

HippoSawrUs

@Mitch I remember flags (Mexico and Italy), dogs (Bedlington terrier), and anatomy (the see-through pages, whatever that's called).

Robusto

Copilot is not a great resource for writing advice. What it told you is nonsense. Bracket "strictly soaking" with commas for better clarity. — Robusto yesterday

Mitch

21:15

@Xanne Is that where somebody found that the pathology images had pieces repeated from other papers? or is this something different?

@HippoSawrUs Oh yeah! The layers of the human body. Awesome!

@Robusto 🎶🎵🎶

That's proof!

Xanne

@Mitch Yes.

Robusto

@Mitch The actual proof has key signature, time signature, and clef marks.

Mitch

@Robusto So it's a proof sketch. Details left to the reader.

Robusto

It's maybe like 22 proof?

Certainly not enough to get you drunk.

Mitch

Yeah that's for kids at their first sleep over.

We want a drink for men to get drunk with.

I think I'll have ... a flaming rum punch!

Robusto

21:23

You look more like a brandy alexander kinda guy.

HippoSawrUs

Once upon a time, I wrote about my grandma's gobs of costume jewelry and shades of silk stockings. Like 10 shades of gray in the '30s. Who knew. Scored the highest, only 90. 1) Know your audience. 2) Don't bore kinky little professors.

That's all I got.

Mitch

@Robusto weird though that someone would think that an LLM created primarily for dealing with code would be able to dispense reliable copy-editing advice.

(even though copilot was probably ChatGPT style trained on everything and then a little extra training on some more code.)

@Robusto snifts

Robusto

@Mitch Since when are programmers fussy about writing?

They're better than designers, I'll give you that.

But that's a very low bar.

Mitch

@Robusto They sure better be fussy about writing code. But yeah, not expected to be wordcels or Scrabble snoots.

@Robusto Ouch.

Designers, man.

They're visionaries.

They don't need no ol fashioned rules about spelling and punctuation to get in their way.

Robusto

Don't get me wrong, I respect what they do. But I think you either lean toward words or pictures.

Mitch

21:28

There's only so much time in the world.

Robusto

@Mitch I long ago learned that designers think of text merely as a design element. When I was a young copywriter, one actually asked me if I could make the text a certain size "so it wouldn't get in the way."

True story.

HippoSawrUs

Every time I ask my husband to answer an IT-word question, he scoffs like somebody should get fired somewhere because they're taking up too much space and time.

And possibly air.

JS, I tried.

I can't bend the space-time continuum…

I'll learn all I need to know about that in World Book Encyclopedia 2025.

GratefulDisciple

21:58

@HippoSawrUs I don't blame you. It's nice to have encyclopedias on paper. I remember handling Encyclopaedia Britannica at the library before the Internet era, which probably also had that see-through layers of human anatomy.

Mitch

@HippoSawrUs Well actually you can bend the space-time continuum. You're doing it right now. Your body is actually curving the space-time manifold in proportion to its mass.

@Robusto Oh got it. Yeah designers tend to come from an art background.

But the museum curators who write the little blurbs on the art pieces seem to have come from a creative writing background.

And the project managers seem to have come from a hall monitor background.

And the sysadmins seem to have come from a Rubik's cube background.

alphabet

@MetaEd Why Pyrrhonism, of course, aka "Pyrrhonian Skepticism." Unfortunately our sources for Pyrrho's own views are kinda bad, but we do have the excellent work of Sextus Empiricus from a couple of centuries later; he claims to be following Pyrrho's own views and I think he's not lying.

Mitch

And the CFOs seem to have come from a selling candy in eighth grade but it kinda turned into to pills background.

And the CEOs seem to have come from a smiling while you're gladly handing over your lunch money to them background.

alphabet

@MetaEd He has an SEP article and the IEP has some coverage of the school as a whole.

HippoSawrUs

23:08

@Mitch Sure, but is it in the S volume or what?

@GratefulDisciple Yes, the World Book set had gold embossing, nice. Now it has a binding mural, or the term for that

@Mitch Is it exponentially if I have an old-lady spine curvature?

Just asking for a friend.

@Mitch I think there's another word for the blurbs. Or term. I can't remember…

Mitch

23:27

@HippoSawrUs I don't know the math. I get all my science knowledge from the captions to diagrams in 50 year old Scientific American.

HippoSawrUs

I feel like I've lost more than I've retained at this point, so people who didn't even try could've just been more space conscious.

Somebody bent my space conscious.

Mitch

@HippoSawrUs that's not necessarily a bad thing for the universe. You're becoming a sink for excess entropy in the universe. Or is it you're creating entropy to help out the universe? Something something quantum chromo dynamics something navigable wormhole!

HippoSawrUs

Old ladies go crazy and delete things and then don't remember what they knew that nobody else did.

Right now, I know I knew how to make better monkey bread than the other ladies…

But I didn't write down all the brilliant things :-(

@Mitch In 7th grade, my friend's parents gave her a bag of weed to sell, instead of allowance, and now she's in prison. See there; sometimes things work out just like you think they will.

The school wouldn't let her be a cheerleader in 8th grade because she was a playground drug dealer, even though she could do a handstand with her feet on her head.

But they didn't have any problem with the actual weed on the playground part.

Mitch

23:45

@HippoSawrUs huh

Sounds like a lot of adults not adulting

HippoSawrUs

All the really weird stuff…is actually true.

One time she was doing flips on the monkey bars and her baggie fell out of her pocket. Uh-oh.

It's like they didn't know how to deal with it.

Like the little man robber in the baby carriage…

Can't arrest a baby gangster.

I forgot about that gag. 3 Stooges? IDK.

Anyway, kids in the '70s were grown ups. Scary.

Mitch

@HippoSawrUs I survived

Mostly

Not too many hospital stays .

Transcript for

English Language & Usage: Multi-Layer…