I just got an e-mail from the university where I work talking about how we need to integrate ChatGPT into courses

I hope you can persuade the uni not to do that; pointing out the down-side.

 

I just got an e-mail from the university where I work talking about how we need to integrate ChatGPT into courses so that students can learn to utilize this productivity-enhancing tool.

https://finance.yahoo.com/news/harvard-computer-science-teacher-chatbot-153236564.html
Harvard’s new computer science teacher is a chatbot


The end result of this productivity-enhancing tool.

 

https://finance.yahoo.com/news/harvard-computer-science-teacher-chatbot-153236564.html
Harvard’s new computer science teacher is a chatbot

View attachment 296942
The end result of this productivity-enhancing tool.

Yep.

Imagine the reaction if, say 40 years ago, some college said that it was going to hire a bunch of instructors for some introductory subject so that there would be a 1:1 the instructor-to-student ratio so that the students could ask these instructors questions at any time of day or night. These instructors were chosen not because they knew anything about the subject, but only because they were cheap and confidently gave plausible sounding answers to most questions, regardless of how correct or useful those answers might be. In fact, the college knows that a significant fraction of the answers given will be far off the mark, but not to worry, because they told the students that they should think critically about any information they receive.

I know I sure as hell wouldn't have gone to school there.

I see this as just more college administrators wanting to jump on the latest bandwagon to show how "with it" they are.

 

https://www.theverge.com/2023/6/26/...-models-data-scraping-generation-remaking-web

AI is killing the old web, and the new web struggles to be born

 

https://www.theverge.com/2023/6/26/...-models-data-scraping-generation-remaking-web

AI is killing the old web, and the new web struggles to be born

An interesting link in the article:

https://meta.stackexchange.com/ques...ow-inc-cannot-consistently-ignore-mistreat-an

Although the feud that appears to be happening or brewing there is completely absent here, a lot of the concerns mentioned regarding AI-generated content is relevant to AAC (and, I would think, the vast majority of forums out there). A lot of discussion has gone on, both publicly and privately, with no solution being found yet that appears equitable and workable. It is a much more complex problem in practice than it might appear at first glance -- so much so that I can't help but wonder if the heartache going on at stackexchange isn't an almost inevitable consequence of just trying to actually do something.

 

My feeling about it.

 

https://www.vice.com/en/article/v7b...onsense-are-all-over-amazons-bestseller-lists

AI-Generated Books of Nonsense Are All Over Amazon's Bestseller Lists
Amazon’s Kindle Unlimited bestseller list was full of books with titles like “Apricot bar code architecture” and “Jessica’s Attention” earlier this week.


https://www.amazon.com/Apricot-bar-...C1JJ6LPT&revisionId=46466a95&format=2&depth=1

 

 

I just caught part of a news segment that I think captured one of the big issues with these large language models and why they are so problematic. Apparently they asked the Bard to write a poem about some mother whose baby died and it came back after five seconds with a poem that ended with something about the baby's soul living on and the interviewer was stunned at the empathy and compassion shown by it. The guy he was interviewing explained how that's not how it works -- that the language model had basically be fed every poem accessible on the Internet and it's just a statistical likelihood that poems about babies and death often have words like soul and eternity and forever in them. He then asked ask the interviewer what word he would say should follow "peanut butter and" and the guy immediately said "jelly". The interviewee then used that as an example of how these models work. It doesn't take any insight or understanding to guess that "jelly" is a word that has a high likelihood of coming next simply based on the frequency with which it DOES come next. Immediately after that, the interviewer gave another example where he asked Bard why it likes helping people and it came back with "because it makes me happy." Instead of applying what he had just been told, asking himself whether or not a phrase like "because it makes me happy" would be a common phrase to follow something like, "Why do you like helping people," he tried to use it to claim that Bard must be more than meets the eye because it was expressing satisfaction and emotion. To me, this reveals the one-level deep thinking of the interviewer, something that is shared by far too many people, where as soon as something tugs on an emotional string, they simply can't apply reason, even when they just went through an exercise showing the man behind the curtain. So it's not surprising that, no matter how much it is shouted from the roof tops what these tools are and are not capable of, everyone and their brothers, sisters, and long-lost cousins ARE going to abuse them right and left.

The best analogy that comes to mind is that these large-language models are basically playing Family Feud, but cheating. Any time someone starts a post, they get to go, "Survey says...!", but they get to see all of the answers and then pick one of them randomly with a likelihood related to their score.

 

Yes, it outputs human satisfaction and emotion, pain, empathy and compassion from inputs the match words that use terms for human satisfaction and emotion, pain, empathy and compassion. It operated on the stupidly simple principle of being a digital parrot using immense, thoughtless and unintelligent computer power made possible by today's powerful hardware technology, not any real computer AI software processing breakthrough.

 

Yes, it outputs human satisfaction and emotion, pain, empathy and compassion from inputs the match words that use terms for human satisfaction and emotion, pain, empathy and compassion. It operated on the stupidly simple principle of being a digital parrot using immense, thoughtless and unintelligent computer power made possible by today's powerful hardware technology, not any real computer AI software processing breakthrough.

A brute-force simulation of intelligence?

 

A brute-force simulation of intelligence?

No, it's not a simulation of intelligence. It's a regurgitation of human intelligence by a system not designed or built to simulate intelligence. It index's, stores and retrieves processed human intelligence.

We give it 'intelligent' queries and it retrieves previous human intelligence responses to those types of queries.

We are behind the ChatGPT curtain.

 

What is the definition of intelligence that were comparing against ?

 

What is the definition of intelligence that were comparing against ?

AAAnnnnddd… that’s the crux of the problem.

 

What is the definition of intelligence that were comparing against ?

https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf

Intelligence might be defined as the ability to learn and perform suitable techniques to solve problems and achieve goals, appropriate to the context in an uncertain, ever-varying world. A fully pre-programmed factory robot is flexible, accurate, and consistent but not intelligent.

We have ChatCPT, flexible √, inaccurate , inconsistent and not intelligent.

 

https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf


We have ChatCPT, flexible √, inaccurate , inconsistent and not intelligent.

Which begs the question, is consciousness and self awareness necessary for intelligence to exist?

 

https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf


We have ChatCPT, flexible √, inaccurate , inconsistent and not intelligent.

Which parts of the definition do we think the current AI models are not meeting ?
seems to me they "learn" as in the take into account the past

and they can "answer" problems they have not seen
simple example, whos seen the pope in a coat picture, thats not a copy of anything ,

 

Which parts of the definition do we think the current AI models are not meeting ?
seems to me they "learn" as in the take into account the past

and they can "answer" problems they have not seen
simple example, whos seen the pope in a coat picture, thats not a copy of anything ,

LLM don't learn or take account of the past because they understand nothing. Everything they produce is of human intelligence origin. It didn't make the pope coat on its own because it liked drawing, a human instructed it to use prior human creations as the basis just like the text versions do.

 

https://fortune.com/2023/07/12/chatgpt-chegg-college-cheating-homework-cs-get-degrees-harvard-grad/
Harvard ’21 grad says Gen Z just uses A.I. to do their homework, they ‘aren’t necessarily interested in using ChatGPT for learning’
At Fortune’s Brainstorm Tech, Nadya Okamoto, Harvard class of ‘21, confronted the CEO of Chegg, saying, "“I was a Chegg user but not because I gave a f*** about learning.”

 

LLM don't learn or take account of the past because they understand nothing. Everything they produce is of human intelligence origin. It didn't make the pope coat on its own because it liked drawing, a human instructed it to use prior human creations as the basis just like the text versions do.

when you say " don't learn or take account of the past "
is that true ?
they are feed the past during training .