Texas professor fails entire class from graduating- claiming they used ChatGTP • Hacker News Board

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#231

Post by Xcelerate » Tue, May 16, 2023, 7:06 PM UTC

hnfong wrote: post #154

Xcelerate wrote: post #113

Earlier quoted context omitted.

> The amount of non-technical people who have very little intuition around the limitations of LLMs is actually fairly startling Is this surprising though? To be honest, I don’t think we have any such thing as “AI experts” currently. LeCun, Hinton, etc. were early pioneers in neural networks and Hutter did a lot of the first work with reinforcement learning, but none of these people seem to agree on much now with rega…

It isn't surprising to the least, and given how little humanity knows about quasi-human level intelligence, I would suggest taking any hard claims about limitations of LLMs with a grain of salt. > Progress is increasingly driven by trying things and seeing what works rather than a theory-first approach, and whenever this occurs, a lot of mysticism tends to accompany the process. This sounds wrong. It sounds as if you…

> This sounds wrong. It sounds as if you're arguing that empiricism and the usual scientific method ends up in mysticism

I think we’re on the same page... I’m arguing that the current approach to creating better LLMs is more akin to an art involving educated guesses and tinkering than it is a hypothesis-driven process with randomized controls (e.g., traditional engineering disciplines or applied physics / chemistry).

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#232

Post by nixpulvis » Tue, May 16, 2023, 7:08 PM UTC

olyjohn wrote: post #224

masswerk wrote: post #163

Earlier quoted context omitted.

One of the possible outcomes of the current trends is that we may rediscover the need for a strongly curated knowledge base as a common reference of conversational truth. (E.g., a classic encyclopedia, advisably more than a single one.)

I'm wondering if something like the old Yahoo! will make a comeback. I remember how it was kinda difficult to get your site listed on there, since someone had to vet the link and add it to the site. Honestly, I would love something like this. Sometimes I want to buy some sneakers, but all that comes up is Amazon and a bunch of trashy spam when I try to find something. Searching for things has become nearly impossible…

What if you could just blacklist domains from your search?

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#233

Post by circuit10 » Tue, May 16, 2023, 7:09 PM UTC

JohnFen wrote: post #215

circuit10 wrote: post #209

Earlier quoted context omitted.

If it had no model of what it was writing about it wouldn’t be able to write anything coherent. It obviously has a world model, it’s not as advanced as what a human has but it’s there

> If it had no model of what it was writing about it wouldn’t be able to write anything coherent. It has a model of the structure of writing on the internet. It can produce text that adheres to that model. That's all that's required for coherent writing.

And the structure of writing in the Internet includes a lot about the world, so a good model of that must contain some form of world model to make accurate predictions

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#234

Post by circuit10 » Tue, May 16, 2023, 7:12 PM UTC

hammyhavoc wrote: post #214

circuit10 wrote: post #209

Earlier quoted context omitted.

If it had no model of what it was writing about it wouldn’t be able to write anything coherent. It obviously has a world model, it’s not as advanced as what a human has but it’s there

The clue is in the name, it's a large language model . When I asked ChatGPT "Does ChatGPT have a world model?", I got this output: "ChatGPT, like other GPT models, does not possess an explicit world model. GPT (Generative Pre-trained Transformer) models, such as ChatGPT, are trained on a large corpus of text data from the internet and do not have direct access to real-time information about the world. They rely solel…

What does this prove? It proves either that someone trained ChatGPT to spit out this information or that this view is very common on the Internet, which doesn’t necessarily make it right

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#235

Post by shalmanese » Tue, May 16, 2023, 7:12 PM UTC

thadt wrote: post #28

The ability to 'cheat' on written assignments has always been a possibility. With the rise of the Internet, it became easier to source it from more places - with a counter balancing rise of plagiarism detectors. Now with Google, Grammarly and ChatGPT, it is trivial to copy/paste a pipeline with enough entropy to defeat many if not most detectors. I suspect the next generation of tools will be finish this cycle - we w…

> If you want someone to produce a piece of writing about something, how they produced it is less important than the result. If it is correct, clear, and gets the job done, then good.

No, this fundamentally misunderstands the difference between work and pedagogy. With a work product, the end result is all that matters because it is a fundamentally new contribution for which the contribution is prized. If a marketer can get a 0.1% better conversion rate using ChatGPT, then that's the best course of action.

Pedagogy is a deliberate simulacrum of work under artificial conditions where learning is done via the process. Nobody cares that you've built the 10 millionth version of tetris, the work product at the end of it is fundamentally uninteresting. The work product is solely there to force you to engage in the process and deliberately useless so it can be shaped towards maximal learning.

Nobody mistakes, for example, that someone who uses a forklift to lift a 300lb bar or pays a friend to do it for them is getting their reps in at the gym because it's intrinsically obvious that the work of moving metal has no value but is instead, meant to induce an internal change in the athlete. Using ChatGPT on an assignment is no different from hiring a forklift to do your workout but people seem to want to get confused about it.

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#236

Post by pg5 » Tue, May 16, 2023, 7:12 PM UTC

sidewndr46 wrote: post #39

binarymax wrote: post #20

To the teachers I know, I’ve been recommending that if they are paranoid about GPT, they can ask the students to cite material from a physical textbook in their assignments. Even if the students type the textbook into ChatGPT at least they read the material. This is beside the point of whether current education needs to be improved, and whether GPT is going to force some change here - but this is a useful stopgap. te…

As someone who was a student over a decade ago, I'd lose my fucking mind if I had to use physical books. There was simply not enough time to go get books off the shelf and complete all my assignments. Everything was done digitally, I can't think of a single exception in college. I once turned in a math exam written using LaTeX and the professor initially declined to grade it. I did concede that I hadn't followed the…

I was doing my undergrad during the same time period, and I always went with the physical book option when one existed.

I for some reason find it much quicker to read and find stuff in a physical book. I also hate reading on a screen (didn't have any e ink device).

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#237

Post by MandieD » Tue, May 16, 2023, 7:14 PM UTC

erehweb wrote: post #129

AJ007 wrote: post #111

Earlier quoted context omitted.

How many people believe horoscopes and astrology carry any meaning? Most will never comprehend LLMs or even the simplest concepts of machine learning.

Autocorrect is fairly well understood and an LLM is an extension of that. (Yes, I know it's more complicated - still, it's a reasonable first approximation)

I wouldn't say that autocorrect and autocomplete are well understood by most users, but are well-accepted and not feared and/or fetishized. Telling my non-tech relatives and friends that "ChatGPT is extremely fancy autocomplete" has settled most of them down a bit.

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#238

Post by knaik94 » Tue, May 16, 2023, 7:18 PM UTC

hnfong wrote: post #161

AJ007 wrote: post #111

Earlier quoted context omitted.

How many people believe horoscopes and astrology carry any meaning? Most will never comprehend LLMs or even the simplest concepts of machine learning.

While we're on the topic of stroking our epeens, do you know how horoscopes and astrology work? Most people here with that condescending attitude will never comprehend it either. FWIW, it's one of the topics I've been trying to understand and come up with a consistent theory for a while. (Not to say those who believe in horoscopes and astrology know how it works though...)

There is no complete theory, because having a complete theory would require a complete theory about cognition. But the general understanding is that horoscopes and astrology play into the fallacy of personal validation. People are bad at evaluating how generic certain descriptions are. It is a well researched and replicated result. The arguments supporting horoscopes and astrology are also written in a way to make them impossible to disprove, making them unscientific.

https://en.wikipedia.org/wiki/Barnum_effect

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#239

Post by qrio2 » Tue, May 16, 2023, 7:21 PM UTC

harha_ wrote: post #136

I'm a bit scared of this progress trend. I tried ChatGPT once and it was so convincing that I can imagine how one could easily get addicted to it. I'm deliberately avoiding using it to avoid dependence.

I get what you mean, but to me that's a bit like avoiding using a shovel because you can dig with your hands. IMO learning these tools, how to use them well, and properly is going to be really helpful. That said I do already feel that dependence creeping in

Re: Texas professor fails entire class from graduating- claiming they used ChatGTP

#240

Post by hammyhavoc » Tue, May 16, 2023, 7:27 PM UTC

circuit10 wrote: post #234

hammyhavoc wrote: post #214

Earlier quoted context omitted.

The clue is in the name, it's a large language model . When I asked ChatGPT "Does ChatGPT have a world model?", I got this output: "ChatGPT, like other GPT models, does not possess an explicit world model. GPT (Generative Pre-trained Transformer) models, such as ChatGPT, are trained on a large corpus of text data from the internet and do not have direct access to real-time information about the world. They rely solel…

What does this prove? It proves either that someone trained ChatGPT to spit out this information or that this view is very common on the Internet, which doesn’t necessarily make it right

In the same breath: what does saying it must have a world model prove? Why must it? For every occasion that it can impress with seeming reasoning, there's plenty of times where it just falls down.

The more balanced view is that we simply don't know, it's assumption on both sides of the fence—Schrödinger's world model. If something resembling a world model is emergent from the sheer scale of data, perhaps, that's a very interesting idea, but it's clear that it still doesn't understand, in the same way that a physics simulation doesn't understand what it's doing either, but it can still output useable information whilst doing something incredibly complex. Ergo, I'd say it's far more likely that there isn't anything out of the ordinary going on.

If I can't get something as simple as a valid NGINX configuration out of it without hallucinations, despite supplying it with the original Apache .htaccess file, documentation and the URL rewrites it will require, why would it have a world model? How an NGINX configuration works is much more simplistic than how the world works with its many systems. Especially when you consider that NGINX is inherently language-based, which should be what it excels at with pattern recognition et al, but it's dumb pattern recognition to a fault with "this commonly follows that" in the training data.