Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI makes the humanities more important, but also weirder

resobscura.substack.com

Post reply on HN
Print view View original

291–300 of 322 posts

Re: AI makes the humanities more important, but also weirder

#291

Post by lapcat »

School-Cotton wrote: post #290
lapcat wrote: post #289

Earlier quoted context omitted.

I still don't know what you mean by "filter". I asked, and you haven't elaborated. The only thing I can think of is that you (mistakenly) assume that I'm suggesting all students must move at the same pace? But that would be a very weird interpretation of me, since I've already expressed strong criticism of that idea: "Our current educational system is designed like an assembly line, based on age and social promotion.…

Sort students into different classes based on their ability and interest.

I have no objection to that, and I don't know why you think I would.

You seem to be assuming that "all students must meet these standards" means that all students must meet these standards at the same age and time, but I never said or implied that.

Re: AI makes the humanities more important, but also weirder

#292

Post by fallingknife »

mrguyorama wrote: post #267
fallingknife wrote: post #198

Earlier quoted context omitted.

Before college was a means to get a job, it was status signalling for the upper class by showing they could spend 4 years not working and learning things with no economic value that few could afford. There was never a time when a large portion of society went to school past 18 for any reason other than economic or status gain, and why should they?

>and why should they? Because modern life is radically more complicated than humans can naturally deal with. Your average peasant for millenia didn't need to understand Information security to avoid getting phished, didn't need to understand compounding interest for things like loans and saving for retirement (they'd just have kids and pray enough of them survive), didn't need to have some kind of mental model for th…

You don't have to understand infosec to not get phished. And education doesn't do a damn thing to help you resist those algorithms.

Re: AI makes the humanities more important, but also weirder

#293

Post by School-Cotton »

lapcat wrote: post #291
School-Cotton wrote: post #290

Earlier quoted context omitted.

Sort students into different classes based on their ability and interest.

I have no objection to that, and I don't know why you think I would. You seem to be assuming that "all students must meet these standards" means that all students must meet these standards at the same age and time, but I never said or implied that.

Right, but how do you filter students, if not by grades? With no grades, how do you know which ones are capable of a more rigorous program? (Similarly, later, how do you know which ones should get into the best universities? And so on)

Re: AI makes the humanities more important, but also weirder

#294

Post by jhanschoo »

vkou wrote: post #239
jhanschoo wrote: post #125

Earlier quoted context omitted.

> Hayden White's assertion that "history is fiction" Your comment made me dig up "Historical text as literary artifact". I have several thoughts on that, but one very important analogy stood out to me and is succinct enough to comment here. In statistics, one talks about the population, collecting data (sampling from the popluation), and building a model from the collected data. You may have guessed that the collecte…

The thing is, this statistics data-to-model only works on large datasets. History deals with very small, very heterogeneous datasets. Finding patterns in them (if they even exist) is very difficult, and it's why historical models and grand unified theories are so often bunk.

> this statistics data-to-model only works on large datasets.

> deals with very small, very heterogeneous datasets.

The population itself is a very heterogeneous "dataset", that becomes (perhaps) apprehensible after you interrogate it with respect to the statistical question you have in mind. Let's not forget that the neatness of the data is constructed from the narrowness of the question asked of it, and even still the data is expected (in the colloquial sense) to take on a minimally-informative, max-entropy distribution, except in respect to the statistic of interest. The critical questions levied against historiography also apply here: perhaps the question you are collecting data for is not related to the population the way you think it is.

> why historical models and grand unified theories are so often bunk.

A lot of history practiced today are self-aware of just how much narrative has to be invented to "make sense" of historical events, and so historians value diverse perspectives. One understands that the same events may not hold the same meaning across communities, and so seeks to record and compare how they are perceived across communities or individuals; that perception in that community's narrative is at least subjectively authentic for that community or individual.

To the extent that the historical goal is to predict political/cultural/economic future, perhaps political/cultural theorists/economists are better positioned.

Re: AI makes the humanities more important, but also weirder

#295

Post by fc417fc802 »

wizzwizz4 wrote: post #287
fc417fc802 wrote: post #281

Earlier quoted context omitted.

No, it doesn't. I don't find the linked thread remotely convincing regarding your (frankly preposterous) claim that preexisting non-deep-learning solutions are better at translation in the general case. Don't get me wrong. I'm sure that once fine tuned by a human for a specific language pair that such systems are better at performing literal translation. But the value proposition of deep learning here is that you don…

Found the other good post in that thread: https://news.ycombinator.com/item?id=35532941 . Google Translate and DeepL were capable of mostly-accurate Japanese translation two (or even 5) years ago (as seen in my previous link). Multiple people ran this text through GPT-4, and none of its translations were any good, as assessed by this Japanese-speaker. There's a reason we needed the Rosetta stone to give as an in for…

You originally made a broad claim about the technology as a class though. It's not particularly surprising that chatgpt (from 2 years ago no less) isn't state of the art at translations given that it isn't optimized for that (at least that I'm aware). The same approaches that are used to construct LLMs can be applied to construct a model intended for machine translation. That is where the value proposition lies.

Re: AI makes the humanities more important, but also weirder

#296

Post by lapcat »

School-Cotton wrote: post #293
lapcat wrote: post #291

Earlier quoted context omitted.

I have no objection to that, and I don't know why you think I would. You seem to be assuming that "all students must meet these standards" means that all students must meet these standards at the same age and time, but I never said or implied that.

Right, but how do you filter students, if not by grades? With no grades, how do you know which ones are capable of a more rigorous program? (Similarly, later, how do you know which ones should get into the best universities? And so on)

> Right, but how do you filter students, if not by grades? With no grades, how do you know which ones are capable of a more rigorous program?

The concept here is mastery rather than grades. When a student has mastered a subject, they can move onto more rigorous subjects. If they haven't mastered a subject, then they need to continue studying that subject. When we socially promote students who receive a grade of D, C, or even B, I wonder, what is the student missing? Why are we giving up and moving on just because an arbitrary amount of time has passed? And if the student has failed to completely learn things at the current level, what's going to happen when the student moves on to higher, harder levels? That's a disaster waiting to happen, compounding ignorance over time.

It appears that the "answer" to these questions is that our society cares more about ranking students than it does about educating students. So we allow the education to be incomplete for many or perhaps most students, as long as we already have our ranking from "best" to "worst" (at a given age). This is, in my opinion, an unfortunate consequence of the multiple conflicting purposes of the educational system.

> (Similarly, later, how do you know which ones should get into the best universities? And so on)

I already talked about the GRE, for example, so I don't know why you're still confused about that. (I'm not actually defending the GRE, or the SAT, another standardized test, but there are clearly ways to evaluate people separate from schools and grades.)

I do have some problems with the inequality of opportunities implied by "the best universities", but that's a whole other discussion. I've already mentioned that the conflicting purposes of the educational system extend to the college level too, are arguably even worse at that level. Think back to the very first quote I posted, from the A.I. cheater for whom the point of an Ivy League university was not to receive an education but to make the right social connections. How is academic achievement even relevant there? To me, this is a sign that the current system is fundamentally broken.

Re: AI makes the humanities more important, but also weirder

#297

Post by tharkun__ »

stanford_labrat wrote: post #170
gdubs wrote: post #141

There's a deeper educational problem that's been decades in the making which is that students are trained to see school and work as a series of never ending goals to achieve. The ultimate one is to 'get a job'. Well, now nobody can really even say what jobs will be available in 5 to 10 years with high confidence. Except maybe the trades – and we mostly cut those programs out of schools a long time ago. If college stu…

Just like AI seems like it's being used as a convenient scapegoat for the layoffs and trimming following the end of ZIRP, now we see it also being used to blame for the failures of our modern education system. Mostly which are that our system only rewards one thing in education: the grade. Not understanding, knowledge, intelligence, but instead a single number that is more easily gamified than anything. And this sing…

    ... our system only rewards one thing in employment: the metric. Not understanding, knowledge, intelligence, but instead a single number that is more easily gamified than anything. And this single number (your metric value) is the single most important thing for every level from junior to principal where it will unironically determine your entire future.
TFTFY

Re: AI makes the humanities more important, but also weirder

#298

Post by wizzwizz4 »

fc417fc802 wrote: post #295
wizzwizz4 wrote: post #287

Earlier quoted context omitted.

Found the other good post in that thread: https://news.ycombinator.com/item?id=35532941 . Google Translate and DeepL were capable of mostly-accurate Japanese translation two (or even 5) years ago (as seen in my previous link). Multiple people ran this text through GPT-4, and none of its translations were any good, as assessed by this Japanese-speaker. There's a reason we needed the Rosetta stone to give as an in for…

You originally made a broad claim about the technology as a class though. It's not particularly surprising that chatgpt (from 2 years ago no less) isn't state of the art at translations given that it isn't optimized for that (at least that I'm aware). The same approaches that are used to construct LLMs can be applied to construct a model intended for machine translation. That is where the value proposition lies.

I'm not sure whether you're missing the point by a mile, or whether I am.

The transformer architecture, later modified to create GPT models, was originally designed for translation. The modifications to make it do predictive text in a chatbot style make it much, much worse at translating: one of the biggest issues is that generative systems fail silently. Using the tech appropriately gives you things like Project Bergamot: https://browser.mt/.

Re: AI makes the humanities more important, but also weirder

#299

Post by csa »

kenjackson wrote: post #225
s1artibartfast wrote: post #218

Earlier quoted context omitted.

There is nothing wrong with going to school to obtain knowledge and skills to secure a job. The problem with the modern educational system is that it isnt very efficient at this task. Instead, most of the value relies on the screening that took place before the students even entered the institution, not the knowledge obtained while there.

Yep, this is a huge problem. I've long argued that we need value add metrics for colleges, and it probably won't be a single number, but rather a set of values depending on input values, e.g., some schools may deliver a lot of value for kids with 1550 SATs, but other schools may do better for kids with 1200. Today we simply use college as a proxy for intelligence, so people just like to go to the highest rated colleg…

> Today we simply use college as a proxy for intelligence, so people just like to go to the highest rated college they can to be viewed as intelligent.

Hmmm… I would say college is a proxy for social currency, of which intelligence is one type. In most cases, intelligence is the least valuable (imho).

Re: AI makes the humanities more important, but also weirder

#300

Post by dr_dshiv »

wizzwizz4 wrote: post #298
fc417fc802 wrote: post #295

Earlier quoted context omitted.

You originally made a broad claim about the technology as a class though. It's not particularly surprising that chatgpt (from 2 years ago no less) isn't state of the art at translations given that it isn't optimized for that (at least that I'm aware). The same approaches that are used to construct LLMs can be applied to construct a model intended for machine translation. That is where the value proposition lies.

I'm not sure whether you're missing the point by a mile, or whether I am. The transformer architecture, later modified to create GPT models, was originally designed for translation. The modifications to make it do predictive text in a chatbot style make it much, much worse at translating: one of the biggest issues is that generative systems fail silently . Using the tech appropriately gives you things like Project Be…

I get your point about the surprising news that gpt4 does so poorly at translation. I didn’t know that!

However, I think the idea is that LLM technologies have improved considerably since then. Do you still feel that Claude or ChatGPT perform worse than DeepL? It would be really nice to have an objective benchmark for comparison

Post reply on HN