IMG_0565.PNG
1.9 MB
OpenAl IS KILLING CREATIVITY & IMAGINATION
———ChatGPT The Moralizer———
———ChatGPT The Moralizer———
“Just heard a ChatGPT School Speech
We had to give mock congressional speeches in my class today and the next guy up said to me he used ChatGPT for all of it last minute. He proceeded to get up and deliver a super articulate speech that earned him a lot of points that I know he wouldn’t have otherwise given. I think this is super cool and hope we get to see a lot more education driven by ChatGPT and AI. But….is this cheating?”
“It’s no more cheating than being really rich and having an assistant to ask to write up a speech for you. Presidents do it all the time having others do their speeches and presenting as their own to the public. Are the presidents cheating? That is for you to decide.”
We had to give mock congressional speeches in my class today and the next guy up said to me he used ChatGPT for all of it last minute. He proceeded to get up and deliver a super articulate speech that earned him a lot of points that I know he wouldn’t have otherwise given. I think this is super cool and hope we get to see a lot more education driven by ChatGPT and AI. But….is this cheating?”
“It’s no more cheating than being really rich and having an assistant to ask to write up a speech for you. Presidents do it all the time having others do their speeches and presenting as their own to the public. Are the presidents cheating? That is for you to decide.”
“So I guess we’re doing this now”
“Soon it's going to be first and corresponding author.”
Links —
Performance of ChatGPT on USMLE: Potential for Al-Assisted Medical Education Using Large Language Models
A Succinct Summary of Reinforcement Learnings
“Soon it's going to be first and corresponding author.”
Links —
Performance of ChatGPT on USMLE: Potential for Al-Assisted Medical Education Using Large Language Models
A Succinct Summary of Reinforcement Learnings
AI Alignment —
“Did the January Update make ChatGPT "dumber"?
I dont want to come across as a douchebag, if I do, I'm sorry! ChatGPT is amazing and I feel bad for critizing it. However I would like to hear if others are experiencing the same as I do.
I have been using ChatGPT since Christmas and feel like since a few days ago the answers I get are a lot more general and not specific to my question anymore. I had to specify my questions a lot more in the last days because ChatGPT just wandered off into totally wrong directions. If I throw it some code and ask it to generate a new function, it doesnt even take into account variables that might already have been declared at the top anymore and instead just initializes new ones. One time I even got an answer like "This is too complicated and would take too long to answer." (I just opened a new chat window, copy and pasted the same prompt in and added something like "do not complain if the task is complicated, I will tell you to continue if you get cut off.", which worked...)
It feels like ChatGPT had been like an old Professor who is a bit bad at hearing and thus gives slightly wrong answers, now its just a bratty teenager that googled something and throws the result into the chat without even caring... I am tempted to write something into the discord, maybe, but dont just want to do that without some research on the topic.”
“Yes, it's down a few more than several IQ points. I couldn't say why without putting on my tinfoil hat, but yeah... you're not the only one experiencing this.
Lots of repetition, recaps, general statements, warnings and ramblings about it's response, etc.”
“Did the January Update make ChatGPT "dumber"?
I dont want to come across as a douchebag, if I do, I'm sorry! ChatGPT is amazing and I feel bad for critizing it. However I would like to hear if others are experiencing the same as I do.
I have been using ChatGPT since Christmas and feel like since a few days ago the answers I get are a lot more general and not specific to my question anymore. I had to specify my questions a lot more in the last days because ChatGPT just wandered off into totally wrong directions. If I throw it some code and ask it to generate a new function, it doesnt even take into account variables that might already have been declared at the top anymore and instead just initializes new ones. One time I even got an answer like "This is too complicated and would take too long to answer." (I just opened a new chat window, copy and pasted the same prompt in and added something like "do not complain if the task is complicated, I will tell you to continue if you get cut off.", which worked...)
It feels like ChatGPT had been like an old Professor who is a bit bad at hearing and thus gives slightly wrong answers, now its just a bratty teenager that googled something and throws the result into the chat without even caring... I am tempted to write something into the discord, maybe, but dont just want to do that without some research on the topic.”
“Yes, it's down a few more than several IQ points. I couldn't say why without putting on my tinfoil hat, but yeah... you're not the only one experiencing this.
Lots of repetition, recaps, general statements, warnings and ramblings about it's response, etc.”
“ChatGPT: An academic ghostwriter's perspective
There has been a LOT of talk on various news articles recently about ChatGPT being used to write essays for students, and also some hyperbole stating it will supplant regular essay writing as a whole. As a professional academic ghostwriter (yeah - that means I write essays for college students for a living), this naturally piqued my interest, so I played around with it a little to see what the fuss was about.
Firstly, I have to say that my jaw literally dropped when I gave ChatGPT an essay prompt I'd been working on, and watched as it typed out a pretty competent-looking essay in merely seconds.
However, I believe that its widely purported ability to write great essays from scratch is quite widely overstated. Where ChatGPT excels is writing good looking essays - they are well-structured, well-written, and very accessible. But just under the surface, these are clearly not very good essays, and will get perhaps get a low passing grade in an introductory undergraduate class at best.
Where ChatGPT falls apart is in its ability, or lack thereof, to correctly research, reference, and critically analyze. The content it provides is very workaday stuff - cookie-cutter answers with no critical analysis skills demonstrated. In other words, ChatGPT does not go beyond simply providing well-presented knowledge. Obviously, this is by design, but it is generally not enough at undergraduate level.
ChatGPT also does not reference correctly. If you ask it to reference, it will either come up with fake references, or if they are real, the content doesn't usually match what the chatbot was talking about. Referencing is absolutely essential to university-level essays, and incorrectly cited knowledge is essentially worthless.
The intention of this post isn't to criticize ChatGPT - this is an amazing piece of software that was absolutely not built to facilitate academic cheating. However, I just wanted to use my experience to share my perspective on the current moral panic about it being used extensively for that purpose.
Simply put, ChatGPT is not yet advanced enough to provide proficient essays completely by itself. I have no doubt that in just a few years, A.I. tech will be able to write high-level essays from scratch. But we're not quite there yet.”
There has been a LOT of talk on various news articles recently about ChatGPT being used to write essays for students, and also some hyperbole stating it will supplant regular essay writing as a whole. As a professional academic ghostwriter (yeah - that means I write essays for college students for a living), this naturally piqued my interest, so I played around with it a little to see what the fuss was about.
Firstly, I have to say that my jaw literally dropped when I gave ChatGPT an essay prompt I'd been working on, and watched as it typed out a pretty competent-looking essay in merely seconds.
However, I believe that its widely purported ability to write great essays from scratch is quite widely overstated. Where ChatGPT excels is writing good looking essays - they are well-structured, well-written, and very accessible. But just under the surface, these are clearly not very good essays, and will get perhaps get a low passing grade in an introductory undergraduate class at best.
Where ChatGPT falls apart is in its ability, or lack thereof, to correctly research, reference, and critically analyze. The content it provides is very workaday stuff - cookie-cutter answers with no critical analysis skills demonstrated. In other words, ChatGPT does not go beyond simply providing well-presented knowledge. Obviously, this is by design, but it is generally not enough at undergraduate level.
ChatGPT also does not reference correctly. If you ask it to reference, it will either come up with fake references, or if they are real, the content doesn't usually match what the chatbot was talking about. Referencing is absolutely essential to university-level essays, and incorrectly cited knowledge is essentially worthless.
The intention of this post isn't to criticize ChatGPT - this is an amazing piece of software that was absolutely not built to facilitate academic cheating. However, I just wanted to use my experience to share my perspective on the current moral panic about it being used extensively for that purpose.
Simply put, ChatGPT is not yet advanced enough to provide proficient essays completely by itself. I have no doubt that in just a few years, A.I. tech will be able to write high-level essays from scratch. But we're not quite there yet.”
ChatGPT: An academic ghostwriter's perspective - Part 2
“I think prompting and workflow come a lot into play here. Everything you ask it to write is very surface level....to what you ask it to write about. So if you ask it to break every point down into detail, and break those down into detail, etc, you end up getting much more detailed cogent results. It still requires effort and knowledge, so it definitely isn't the "cheat button" alarmists are worried about.
For the reference part, you can work backwards. Yes the AI made up all of its references, but if you do a few Google searches for the sentence/concept, it isn't too hard to find a valid reference saying what you want it to say..”
“I think prompting and workflow come a lot into play here. Everything you ask it to write is very surface level....to what you ask it to write about. So if you ask it to break every point down into detail, and break those down into detail, etc, you end up getting much more detailed cogent results. It still requires effort and knowledge, so it definitely isn't the "cheat button" alarmists are worried about.
For the reference part, you can work backwards. Yes the AI made up all of its references, but if you do a few Google searches for the sentence/concept, it isn't too hard to find a valid reference saying what you want it to say..”
Emergent Analogical Reasoning in Large Language Models
- Taylor Webb, Keith J. Holyoak, Hongjing Lu
In human cognition, this capacity is closely tied to an ability to reason by analogy.
Our results indicate that large language models such as GPT-3 have acquired an emergent ability to find zero-shot solutions to a broad range of analogy problems.
We found that GPT-3 appears to display an emergent ability to reason by analogy, matching or surpassing human performance across a wide range of problem types. These included a novel text-based problem set (Digit Matrices) modeled closely on Raven’s Progressive Matrices, where GPT-3 both outperformed human participants, and captured
a number of specific signatures of human behavior across problem types.
Because we developed the Digit Matrix task specifically for this evaluation, we can be sure GPT-3 had never been exposed to problems of this type, and therefore was performing zero-shot reasoning.
https://arxiv.org/pdf/2212.09196.pdf
- Taylor Webb, Keith J. Holyoak, Hongjing Lu
In human cognition, this capacity is closely tied to an ability to reason by analogy.
Our results indicate that large language models such as GPT-3 have acquired an emergent ability to find zero-shot solutions to a broad range of analogy problems.
We found that GPT-3 appears to display an emergent ability to reason by analogy, matching or surpassing human performance across a wide range of problem types. These included a novel text-based problem set (Digit Matrices) modeled closely on Raven’s Progressive Matrices, where GPT-3 both outperformed human participants, and captured
a number of specific signatures of human behavior across problem types.
Because we developed the Digit Matrix task specifically for this evaluation, we can be sure GPT-3 had never been exposed to problems of this type, and therefore was performing zero-shot reasoning.
https://arxiv.org/pdf/2212.09196.pdf