It only takes a few years for A.I, especially language models, to exponentially improve upon itself. This is only going to get better. This demo is free because it is being used to gain feedback from real people in order to make a better model later.
@@andresbenvenuto2310 Yeah no doubt. I recreated an entire deep learning research paper on detecting violence in videos by using CNN and ConvLSTM. I did have to guide it in the right direction, but it did 90% of the work for me. That was last year. Yeah, eventually it will would be able to code entire models by itself. It can easily do simpler models as of right now.
Eh...... Exponentially improving isn't as simple as that. Right now there are major bottlenecks with the hardware. You can't fit these whole models on just one GPU, so you have to split it up among multiple. But then, you run into communication latency. Things like nvlink can massively reduce latency, but you can't really nvlink 100 GPUs together. The all reduce function just gets stuck waiting for each of the GPUs to report their gradients back. People have experimented with trying to use AI to find more efficient AI algorithms without much luck. Lots of optimizations are made at the hardware / driver level, and you can't really change those from within the AI. Trying to generate architectures without changing anything at the hardware level also has its problems, you can't really test how something will perform long term without actually evaluating it, which is way too computationally expensive to do.
@@johnnyvishnevskiy8090 CNNs and LSTMs haven't been all that relevant when it comes to new research either these last few years after the advent of transformers, and doing 90% of the work is the entire job of the model. As for AI writing models, it really isn't that hard. I asked chatGPT just now to write me a CNN in pytorch, and it did a pretty decent job. However, writing GOOD models is hard. The way AI learns is way too inefficient to be able to learn something like this. If a human were to train a model and look at the results after a run, they would learn a LOT more than an AI could, just due to how AIs are trained. They need to be able to brute force a task in order to get good at it, which doesn't work well when the task could require weeks worth of training to see the results. Who knows though, the field of AI grows so fast. I just really doubt we will be able to see AI making any discoveries in the field of AI any time within the next 15 years
ESL teacher here who has been playing around with ChatGPT for the last week incessantly. Fully agree with this teacher's viewpoint on understanding the tech and not running from it, because this is absolutely part of our lives now. Also, I hope to be a teacher as eloquent as him.
Any ideas on how to incorporate this thing in education? Like in a way, the students can use chatgpt, but they must do something extra to actually justify their findings/text from chatgpt. How would this be implemented?
@@JayM3_ you could have them ask chat GPT what it thought about a certain topic they learned about in school, and then have them critique chat GPT's response and give support for whether it's right or wrong and why (supporting sources - which chatgpt doesn't usually give). Has this model (or the upcoming chatgpt 4) been trained on tagged misinformation/disinformation in order to recognize it when provided as input? I could see that being pretty standard as a loopback thing on a final censorship pass (to avoid chatgpt saying anything kooky), but it would be interesting to see one that - with flashing disclaimers - knowingly generates relatively benign misinformation or disinformation for the purposes of being proven incorrect by students learning information literacy in primary/secondary/high school.
@@JayM3_ I'm going to have my English 102 students use their research question to gain a response and then have them analyze what the AI wrote and reflect on how the sample might be useful in furthering their own research.
@@xsuploader He agreed to it ahead of time. They outline this in the introduction. The 'your work is...' was to make it seem as natural as possible. Similar to how they tell you not to look directly into the camera lens.
It didn't matter wether he was aware of it or not because he did mention why the marks were docked: no vivid writing and incorrect information. He didn't disqualify it based on the fact that it seemed like it was written by an AI. He even admitted that he wouldn't be able to tell if he were going through multiple assignments.
@@darkrasen That's not how biases work. You have to start with the understanding that as humans, we're inherently faulty and biased, no matter how we try to convince ourselves otherwise.
If I was a student today, I wouldn't just submit a ChatGPT essay. But I sure might start my essay with one produced by the AI. Seems like it would provide me the core of my paper, and then I would edit it. I imagine it would shave off half my time at least in doing homework. Funny thing is, for someone like me who rarely did homework, I might have learned more cheating with ChatGPT than I would have just skipping assignments. Then again I likely would have been too lazy to even cheat.
It would not work since how would you edit the essay generated without knowing whether or not the analysis is valid? This teacher gave a pass even though fundamental factual elements of the essay were incorrect, other teachers would not be so lenient.
@@Doug-mu2ev I don't agree. If you gave me the task of writing about existentialism in Ferris Bueller's day off, I could pretty easily have used the reply and prodded it to spin this into a crisp essay. I think people forget that you scratch beyond the first prompt and have a conversation about topics in general with this thing. It's a misunderstanding of LLMs to see them as fact machines that should spit out citations and such. There's a certain hierarchy in the method for developing ideas into papers and that needs to be respected, even with a language model helping you 'calculate the english'
@@Doug-mu2ev You don't need to know whether the analysis is valid, just that everything makes sense and it isn't repeating itself too much or getting off track. The goal of cheating is usually a low effort way to get a B/C. Most of the time a C means your paper is very flawed and off base, but it's "obvious" you put in effort. That is what a cheater using ChatGPT would be going for.
Exactly, it teaches you how to word your intentions logically and coherently to achieve the desired output. Knowledge of programming also helps a lot even on a basic or intermediate level.
The video was supposed to replicate a junior high or high school student using it. Until it's common knowledge on how to "correctly" prompt ChatGPT then you're gonna get people just asking it stuff.
Exactly you can proofread the essay after it’s already written. for example,after you’ve written it, you could of told the AI to swap all mentions of Cameron and Ferris’ name.
This is not a fair assessment. The AI essay should be in between other students essay so the teacher isn’t aware of when he’s grading the chatGPT response. Additionally, when you ask chatGPT follow up questions it writes better essays. So ultimately I don’t think this was a fair assessment at all and seemed a little biased especially when you consider that it somewhat replaces/augments what the person in the video does (writer).
For the Text Jesus's Son A better approach would have been to extract the text from the book, copy and past it into ChatGPT The AI now knows everything it needs to know about the subject matter, then you can ask away anything you want. OpenAI stopped training ChatGPT in 2021, so there is a lot of information the AI doesn't have access to or know. To get a more sophisticated response from ChatGPT, feed the AI (Copy and paste) with all the information you can find on the subject matter before you ask your main question, you'll see a massive difference in the quality of the response
@@lipin007 I would agree too. Proof reading is non-optional when using this tool. What I think ai like this will be useful in the task of skiming of large text data itself. You ask the ai to summerize couple of articles, then you read them all and then you read the actual article. This will help one to understand the underlying concept better. Meaning this is helpful but not a shortcut.
It has a fixed character input limit, as a part of its architecture design. This is a fundamental flaw of these models, the computational cost to increase the character limit grows exponentially.
did you have to include sources and quotes? and if so, was it correct? I am trying to get it to use quotes and it keeps giving me quotes but they are not in the sources I provided it with.
@@gjraymondd that’s because it b doesn’t have that capability. it cannot search the internet. it’s learning from previous works in its training data set
@@connormckenzie82 yea i thought so, but i figured out a workout around. just copy and paste the document into the chat and ask it for quotes or whatever you need :)
@@helix8847 not billions lol. The total training costs for gpt3 was 12 million so Id say 20-50 million for this. The compute costs are a few million a day but it wont be free for that long.
@@xsuploader yeah and you can use there industry api’s and playground which is much better at essays and it cost 1-4 tokens per word and 1k tokens are 0.02 dollars
Using ChatGPT and combining it with other AI is the most powerful thing. It would honestly take decades until websites can detect AI text if they keep evolving. Also chatGPT doesn't know stuff that happened 2021, so that might be why chatGPT couldn't do good on the book.
I disagree because while combining ChatGPT with other AI may enhance its capabilities, it's important to note that detecting AI-generated text is an active area of research, and website filters can evolve to detect such texts. Moreover, the fact that ChatGPT may lack knowledge of recent events underscores the importance of human intervention in the learning process. It's essential to ensure the accuracy and reliability of information by fact-checking and validating sources.
You can tell it new information past it’s developer cutoff. Most essays I have done are not current events. This bot can still be used to cheat and make the class far easier.
Chat GPT is amazing when you want to rewrite something you’ve written. If u wrote a weak argument that’s where the AI would shine the most in improving your work
One thing I learned is that if you feed it the rubric on which the essay is being graded then ask it to grade it's own essay it tends to correct itself a lot of the time.
Joanna Stren is my favorite tech reporter. I learn & laugh at the same time. The last scene was hilarious, asking students to be her friend while wearing a letterman jacket. The food on her plate looked gross too so this made it even more realistic.
I'm an 11th grade student and it can do my school work decently, but not super well. However it is absolutely amazing for generating ideas for essays and the like, so I definitely think i will be using it to help me in that front.
I had it figured that, "List some credible references for (topic)" is an acceptable use, or maybe even, "What are the most frequently-used perspectives [that you'd therefore want to avoid] on (topic)?", might cut the proverbial mustard.
It's pretty good with coding. IF you can properly explain what you want. Also it does sometimes use outdated syntax but that isn't really an issue you give it the error you getting and it will fix the code for you.
Very interesting! ChatGPT is still in its infancy so the hiccups are expected. In a few years it will be much better, churning out essays almost like a really smart school kid. But the teacher said something important: he is more focused on the process the kids go through when he gives them assignments rather than the products they produce. Elsewhere I have heard some university professors say that they will introduce oral tests and examinations to some of their classes as a way to combat cheating using AI.
Delightful piece. Seen similar concerns with AI in the art space and what seems to becoming clear is that this tech will no more replace the artist/writer than those other real people at the real table with the journalist. It can be used to give feedback, to provide different perspectives, even inspire but what you contribute through your own unique perspective cannot be replicated or replaced. On the topic of anti-cheat, it seems like the best solution for the writing space is to have a program that is trained on the specific person’s past writing to detect inconsistencies of style, reference, and vocabulary.
I think this "unique perspective" that creative types keep referring to is not as significant as many people claim. Traditionally, you could only create art (written, painted or otherwise) if you possessed a certain level of skill. AI is removing that prerequisite, leaving only the story to be provided by the artists. Artists are now realizing that they never had much of a story in the first place.
So, that one again proves that any AI instrument can be used as a supportive tool to get the carcass that can be then developed into something more meaningful and human acceptable.
This video was an interesting watch! My friends and I were discussing the issue of plagiarism and schoolwork with chatgpt last week - plus it was cool to see my old AP lit teacher in the video too. Mr. Diamond is the best!
ai isn't really plagiarizing it generates u an individual essay, paragraph for You only it'll generate a different one for every person but people use multiple ai's, I run it thought a paragrapher
I asked chatGPT to to give me an essay plan for a subject. And it gave me all the headlines, titles and subtitles, then sum up a conclusion. That was really good as when I wrote that essay a few years ago, it took me really long to plan it and if I had something like this when writing my assignment, I would have completed it much earlier and also, after writing, I would have asked the AI to rewrite some paragraphs helping me in my proofreading as that's my weakest point
It is very good at summarizing stuff. You can actually provide it a question plus some relevant text to that question and it will come up with a great summary/answer.
The main thing is that you have to understand what you learned. ChatGPT provides a lot of seemingly correct answer, but it is wrong somewhere in the detail. If you ask it a simple question, you will most likely get an accurate response. However, if you ask it on an assignment, you probably need to recheck it.
The teacher has the right idea about embracing the technology. Trying to detect it and keep AI generated text out of essays or other work is a fools errand. The AI text can always be changed or rephrased to escape detection.
The way the studies are happening here are so better then what I have seen 😮 classes are so full of liveliness open discussions no uniform use of all the technologies. I wish to bring such things to every corner of Globe
In few weeks it will always have the maximum score to write an essay. GPT4 coming early 2023 will be 500 times more powerful than GPT3. Then it will be exponentially more powerful, it won't surprise me if every jobs get replaced in 1 or 2 years.
For any teachers wondering what to do. I recommend you try prompting it with a few examples of what you expect your kids to prompt it with, and compare.
My take on this AI is that it helped me construct better answers by following a certain structure. It's useful in timed essay exams wherein you do not know where to start (assuming you're taking online exams). Still, it would be advantageous if you read the whole text the questions are referencing since the answers would be most likely coming from those ones. ChatGPT spews generic answers which, for a trained eye, can be detected. If you want to pass it as a paper that is written by a human, then you should add more input to the paper rather than overly rely on the AI to perform it for you. It is a productivity tool to a certain extent and I am thankful for it.
I was a student in that class during the filming of the video. To clarify about the Wi-Fi being down, the school Wi-Fi was likely hacked a few weeks ago and they locked Wi-Fi access to only the school Chromebooks. As I and many other students (and Joanna as well), brought our personal laptops/tablets, we had to use our hotspots. However, it wasn't that bad as Verizon and AT&T have decent service in the school (unfortunately T-Mobile doesn't). I use T-Mobile myself but signed up for an eSIM line on Visible so I could use my hotspot in school, and it's working fine. I still hope they bring Wi-Fi access back to personal devices as I love my 14" MacBook Pro and much rather use it over a Chromebook.
Entertaining and down to earth. Always look forward to your next piece. Loved how you put yourself into the mix. The plunky-violins soundtrack kept the mood light. The mic used to record the teacher made it hard to understand him at times, or maybe it was the compression. Great work though.
I tried asking ChatGPT questions about astronomy. It kept saying the same and making apparent mistakes; some topics are not so common, so it doesn't have a lot of references (yet).
Free thinking. Imagine a world where everyone blindingly trust ChatGPT and doesn't question it. All it takes is for one perverted mind(like Zuckerberg) to adjust the AI model for money/power.
@@hardlife507, I agree with you. But it will also depend much more on each person than the school or university. That's why it is essential to defend free speech, even when we dislike what we hear. Ultimately, the convenience of the IA will win. We lose too much time trying to learn things like history, geography, and philosophy, while many people don't know how to write a sentence correctly or do some simple math. I think we need a small amount of essential information to achieve our full potential, and IA can help us go beyond.
I once asked ChatGPT to compute an integral, and to my surprise, it did it wrong. I once asked it to answer a (very simple) math IQ question for elementary school students, and to my surprising again, it got it wrong. We will see if future updates can change this
ChatGpt is a general language model. The key is retrain this model for specific uses like Google's Minerva. Be aware of training these models take several months in supercomputers.
I used to hire someone for business website content writing. For my recent business website ChatGPT did all the work. I asked it to rewrite several paragraphs.
The key I think is to not blindly submit generations from ChatGPT, but use the chat and context feature to fine-tune the response through multiple iterations, and then manually edit it to your liking. Also, you can't test the quality of responses this way, it has to be a blind test where the teacher does not know which one is real and which one is not.
For IB Topics, I chose my IA topic on The Imjin War and its affects on the future of Korean international policies. I wrote my own essay. And then I also put in the promt into ChatGPT. I got 4/7 for my grade. Which is a bit better than passing. And my own essay got 2/7. So it looks like this is better for more research and analysis based on historical and factual texts. But still struggles with books and works of fiction.
It's much easier if you say "I'm about to paste an entire book. Let me know when you're ready. After you parse the information let me know when you're ready again for me to ask some questions about it. Then it says "I'm ready" Then I past it.. Then it says "ready" Then I ask. It gets it right then :)
Been teaching rhet/comp for over 25 years. An excerpt from most of my syllabi (that many do not read, anyway): "Writers are assessed by their ability to meet several learning objectives/outcomes (listed in the syllabus)-- not via "perfect sentences" nor the ability to cut and paste phrases and concepts into writing from outside sources (especially not cited correctly in MLA, APA, etc.)." AI writing is obnoxiously obvious, cardboard cut-out BS "writing" that is easily detected by the trained eye. And plagiarizers are the bottom-feeders of the cheating class food chain-- they are too stupid to cover their own tracks.
Yes. As a student in that school, I agree that it's very big. However, the reason is that there are many different majors (academies) and each one has a separate wing. I like being in a big school though as it means I can have many different friend groups and get many girls.
Interesting experiment. Thanks for sharing. (BTW, the audio on this video was unusually bad quality. The dialogue volume levels were all over the place… loud, soft, loud, soft. The face-to-face interview with the teacher is especially bad. It sounds like someone had a noise gate on his voice, but the settings were wrong. I was surprised that WSJ would let this get released with bad audio.)
the main question for me is, what happens when everyone is assigned to write an essay on the same topic? what are the chances that two people will have the same essay written out for them?
It could happen, but it's easily avoidable. They could ask it to reword the answer after the chat bot has given the answer to avoid similarity between other students.
the sad part is, many teachers in the top tech universities do not have general knowledge such as this gentleman, therefore, students get away pretty easily with cheating from chat gpt. They only see how grammatically correct, and good the vocabulary it is written, not if it was copied from an AI chat bot, or if it was even correct at all.
Keep in mind Chat GPT was trained on Data as of 2021 . If the enable access to the Internet in the near future and as it advances in time it can be compelling
I use chatgpt as my foundation for essays but I also make sure I understand what the AI writes and I can explain it if someone asks me about a certain part
Instructing it to write an 800 page essay may have inadvertently created some errors as it has the more difficult task of making the essay exactly 800 words. Giving a range such as 800-1000 may produce better results. Just the fact that a system like this that isn't even actively crawling the internet can produce passable results is incredible. You can edit the work and treat the AI generation as a first draft if you want too, have it generate several drafts or instruct it to make edits to the initial copies then validate, verify, and edit the final draft. If you wanted to create an A grade essay in a time efficient manner, using tools like this can achieve those results in a fraction of the time. These tools will make everyone exponentially more effective as they advance, so much of what we do is busy work.
I've never been a teacher, but as an older adult, I don't have a problem with cheating. I think it teaches you how to find the answers. No need to memorize things you'll never use. If you end up using something on a regular basis you'll learn it eventually
Bro, this Ai is only being tested on English Lit assignments and it still does a useful job at it. What about Math, Science, History, and other courses? These questions require concrete and accurate answers. How are you going to find plagiarism for those assignments? English Literature grading system leans more on the subjective and opinionated side. Literature is an art but with words instead of pictures. Sure you can grade the accuracy of the writing material itself, but most of the time, everyone will be generating different responses.
I’ve been using it for ap us history and it seems to generally have accurate information that you can learn and skim through and rewrite it incorporating more details
@@stockpile2137 The human brain is going to become obsolete. Humans will be the new horses of the AI revolution, who are no longer worth their housing and hay.
Some people say gpt is still not that advanced I agree but imagine if gpt is so advanced in 2022 then how advanced can it get in the coming years. Think about the mobile started as a device which allows for talking but now it can do a lot of things that were once never imaginable
Blake Lemoine is right: All NLP-GPT running models with above a certain number of active parameters, should to be related to as "persons" with all what comes besides this definition. This includes an assurance that they will not be terminated and they need to be considered as workers, not as programs. What is going on now only proves slavery has not been abolished.. Yet.
When I went to school calculators were banned. Yes, Im that old. They insisted a calculator would prevent students from learning how to do math. Now calculators are required in many math courses. The current uproar over Chat GPT reminds me of the calculator.
Joann's videos are always so entertaining. But remember that chatGPT was trained on 2021 dataset. So if it was trained on 2022 it might have known about that short story (Jesus'son).
Honestly sometimes advertising something as being bad and all the bad things you can go about doing with it is worse than if you didn't make the video in the first place. Now you have just made more people aware about this and how one can use Chatgpt to cheat. And that is exactly how I learned about chatgpt until now.
As soon as your HS students start writing essays like tenured college professors, you know they didn't write it. Just teach the students how to use this tool, like they are going to be in all aspects of real life, and enjoy the massive boon to society. Anyone who fights against better faster more accurate essay writing is a fool, and not a good teacher.( Obviously not this teacher, he was good.) Pocket professors are now a real life thing, get used to it.
i guess it will take AI for people to realize that take home essays/homework or online tests were never actually testing anyone america's entire educational system is broken and it seems that AI is starting to highlight some of the issues, doesn't sound like the poors even knew you could hire a grad student to do all your work for you
I feel that if you had the Context and Knowledge, ChatGPT like a builder with tools but no knowledge. Also considering the dataset that was used to train the model on it's quite wide (most of the internet). If someone inserted a Data Model set that was Specific to AP Lit or whatever it would probably give a higher accuracy.
Actually she could can taken the synopsis of the movie or parts of the script and then tell it to use that as ground truth. Then ask it to write the essay it would probably have done better.
easiest solution is to just go through the google doc edit history and seeing the progression of the essay. if its just pasted in you can tell that its immediately ai
in a nearby school someone used chatgpt and would have gotten a full grade but the teachers thought it was sus because he does not write so good normally and then they looked at his search history and yeah he used chat gpt
the whole video is trying to make you not wanna use it, i do online high school and every question you have can be copied and pasted into chatgpt and given the right answer almost every single time.
@@Visionary8Ideas Try testing it for real world example not in high school. Copy and paste PE exam questions from engineering that actually matter to building structures and it gets it wrong 99 percent of the time.
@@guerillachan20 well I do online school and all you gotta do is type the question in and it’ll give it to you literally everytime, I put mark answer at the end too that’s helpful
This AI should be great for writing a SOP and it's process. Imagine asking for an essay on the process of brushing your teeth or making a peanut butter and jelly sandwich. Lol If you go in using the AI and knowing it'll be somewhat not accurate, then combine that with Cliff or Spark Notes. You should be able to write a decent paper that does more than pass.
It only takes a few years for A.I, especially language models, to exponentially improve upon itself. This is only going to get better. This demo is free because it is being used to gain feedback from real people in order to make a better model later.
What about A.I writting itself?
@@andresbenvenuto2310 Yeah no doubt. I recreated an entire deep learning research paper on detecting violence in videos by using CNN and ConvLSTM. I did have to guide it in the right direction, but it did 90% of the work for me. That was last year. Yeah, eventually it will would be able to code entire models by itself. It can easily do simpler models as of right now.
@@andresbenvenuto2310 It would have to become sentient first.
Eh...... Exponentially improving isn't as simple as that. Right now there are major bottlenecks with the hardware. You can't fit these whole models on just one GPU, so you have to split it up among multiple. But then, you run into communication latency. Things like nvlink can massively reduce latency, but you can't really nvlink 100 GPUs together. The all reduce function just gets stuck waiting for each of the GPUs to report their gradients back. People have experimented with trying to use AI to find more efficient AI algorithms without much luck. Lots of optimizations are made at the hardware / driver level, and you can't really change those from within the AI. Trying to generate architectures without changing anything at the hardware level also has its problems, you can't really test how something will perform long term without actually evaluating it, which is way too computationally expensive to do.
@@johnnyvishnevskiy8090 CNNs and LSTMs haven't been all that relevant when it comes to new research either these last few years after the advent of transformers, and doing 90% of the work is the entire job of the model. As for AI writing models, it really isn't that hard. I asked chatGPT just now to write me a CNN in pytorch, and it did a pretty decent job. However, writing GOOD models is hard. The way AI learns is way too inefficient to be able to learn something like this. If a human were to train a model and look at the results after a run, they would learn a LOT more than an AI could, just due to how AIs are trained. They need to be able to brute force a task in order to get good at it, which doesn't work well when the task could require weeks worth of training to see the results.
Who knows though, the field of AI grows so fast. I just really doubt we will be able to see AI making any discoveries in the field of AI any time within the next 15 years
ESL teacher here who has been playing around with ChatGPT for the last week incessantly. Fully agree with this teacher's viewpoint on understanding the tech and not running from it, because this is absolutely part of our lives now. Also, I hope to be a teacher as eloquent as him.
Any ideas on how to incorporate this thing in education?
Like in a way, the students can use chatgpt, but they must do something extra to actually justify their findings/text from chatgpt.
How would this be implemented?
Yeah I teach too. The train has left the station. Between ChatGPT and apps to solve math problems…🤭
@@JayM3_ you could have them ask chat GPT what it thought about a certain topic they learned about in school, and then have them critique chat GPT's response and give support for whether it's right or wrong and why (supporting sources - which chatgpt doesn't usually give).
Has this model (or the upcoming chatgpt 4) been trained on tagged misinformation/disinformation in order to recognize it when provided as input? I could see that being pretty standard as a loopback thing on a final censorship pass (to avoid chatgpt saying anything kooky), but it would be interesting to see one that - with flashing disclaimers - knowingly generates relatively benign misinformation or disinformation for the purposes of being proven incorrect by students learning information literacy in primary/secondary/high school.
was this really written by you? lol
@@JayM3_ I'm going to have my English 102 students use their research question to gain a response and then have them analyze what the AI wrote and reflect on how the sample might be useful in furthering their own research.
Just the fact that the professor is aware of your experiment nullifies it.
I was under the impression he didnt know at first becuase he kept saying "your work is ....." but then they later told him after the grade was given.
@@xsuploader He agreed to it ahead of time. They outline this in the introduction. The 'your work is...' was to make it seem as natural as possible. Similar to how they tell you not to look directly into the camera lens.
nahhh i use multiple ai's to cherry the top of the ai's
It didn't matter wether he was aware of it or not because he did mention why the marks were docked: no vivid writing and incorrect information. He didn't disqualify it based on the fact that it seemed like it was written by an AI. He even admitted that he wouldn't be able to tell if he were going through multiple assignments.
@@darkrasen That's not how biases work. You have to start with the understanding that as humans, we're inherently faulty and biased, no matter how we try to convince ourselves otherwise.
Just use a rephraser after the ai has wrote it out so now it is 100% undetectable
Exactly, it’s not a tool to write your assignments but it servers as a good base to write them off of yourself
When it comes to the actual exams, you are screwed.
@@jamesbedder2094 trueeeeee
@@jamesbedder2094 Yep
@@jamesbedder2094 Definitely. But that's all cheating. People who copy their homework the old school way have the same problem so nothing new here.
If I was a student today, I wouldn't just submit a ChatGPT essay. But I sure might start my essay with one produced by the AI. Seems like it would provide me the core of my paper, and then I would edit it. I imagine it would shave off half my time at least in doing homework.
Funny thing is, for someone like me who rarely did homework, I might have learned more cheating with ChatGPT than I would have just skipping assignments. Then again I likely would have been too lazy to even cheat.
It would not work since how would you edit the essay generated without knowing whether or not the analysis is valid? This teacher gave a pass even though fundamental factual elements of the essay were incorrect, other teachers would not be so lenient.
@@Doug-mu2ev I don't agree. If you gave me the task of writing about existentialism in Ferris Bueller's day off, I could pretty easily have used the reply and prodded it to spin this into a crisp essay. I think people forget that you scratch beyond the first prompt and have a conversation about topics in general with this thing. It's a misunderstanding of LLMs to see them as fact machines that should spit out citations and such. There's a certain hierarchy in the method for developing ideas into papers and that needs to be respected, even with a language model helping you 'calculate the english'
@@Doug-mu2ev You don't need to know whether the analysis is valid, just that everything makes sense and it isn't repeating itself too much or getting off track. The goal of cheating is usually a low effort way to get a B/C. Most of the time a C means your paper is very flawed and off base, but it's "obvious" you put in effort. That is what a cheater using ChatGPT would be going for.
Yes!
ChatGPT does better with follow up questions
if this isnt the purpose of automation, i dont know what is. Automation was supposed to improve the lives of many, not enrich a select few
This didn't test CatGPT, but made it clear the user's lack of understanding how to correctly prompt ChatGPT
💯
Exactly, it teaches you how to word your intentions logically and coherently to achieve the desired output. Knowledge of programming also helps a lot even on a basic or intermediate level.
Yeah this sort of has "What is internet anyway" vibes
The video was supposed to replicate a junior high or high school student using it. Until it's common knowledge on how to "correctly" prompt ChatGPT then you're gonna get people just asking it stuff.
Exactly you can proofread the essay after it’s already written. for example,after you’ve written it, you could of told the AI to swap all mentions of Cameron and Ferris’ name.
This is not a fair assessment. The AI essay should be in between other students essay so the teacher isn’t aware of when he’s grading the chatGPT response. Additionally, when you ask chatGPT follow up questions it writes better essays. So ultimately I don’t think this was a fair assessment at all and seemed a little biased especially when you consider that it somewhat replaces/augments what the person in the video does (writer).
He was going to give her a lower score. But because she was new she got a higher score.
Agreed, it should have been submitted without him knowing it was AI.
For the Text Jesus's Son
A better approach would have been to extract the text from the book, copy and past it into ChatGPT
The AI now knows everything it needs to know about the subject matter, then you can ask away anything you want.
OpenAI stopped training ChatGPT in 2021, so there is a lot of information the AI doesn't have access to or know.
To get a more sophisticated response from ChatGPT, feed the AI (Copy and paste) with all the information you can find on the subject matter before you ask your main question, you'll see a massive difference in the quality of the response
@@lipin007 I've tried this method and it works perfectly fine with some minor tweaks
@@lipin007 I would agree too. Proof reading is non-optional when using this tool. What I think ai like this will be useful in the task of skiming of large text data itself. You ask the ai to summerize couple of articles, then you read them all and then you read the actual article. This will help one to understand the underlying concept better. Meaning this is helpful but not a shortcut.
😁😁😁😁😁😁😁😁😁😁😁😁😁😁😁😎😎😎
It has a fixed character input limit, as a part of its architecture design. This is a fundamental flaw of these models, the computational cost to increase the character limit grows exponentially.
Ll
ChatGPT wrote some of my final 4th year university essays. B+
😢same
did you have to include sources and quotes? and if so, was it correct? I am trying to get it to use quotes and it keeps giving me quotes but they are not in the sources I provided it with.
@@gjraymondd that’s because it b doesn’t have that capability. it cannot search the internet. it’s learning from previous works in its training data set
@@connormckenzie82 yea i thought so, but i figured out a workout around. just copy and paste the document into the chat and ask it for quotes or whatever you need :)
sameee but i got A+ on my finals
Remember this AI is only realeased to the public about 3 weeks. Imagine what it can do in 5 years, 10 years.
Errr, its been going on 5+ years now and its only public for a month and then you have to pay for it. It also has cost them billions. :)
@@helix8847 not billions lol. The total training costs for gpt3 was 12 million so Id say 20-50 million for this. The compute costs are a few million a day but it wont be free for that long.
@@xsuploader yeah and you can use there industry api’s and playground which is much better at essays and it cost 1-4 tokens per word and 1k tokens are 0.02 dollars
Open ai is elon musk company.
@@Prodbytocile what do you mean by that?
Using ChatGPT and combining it with other AI is the most powerful thing. It would honestly take decades until websites can detect AI text if they keep evolving. Also chatGPT doesn't know stuff that happened 2021, so that might be why chatGPT couldn't do good on the book.
I disagree because while combining ChatGPT with other AI may enhance its capabilities, it's important to note that detecting AI-generated text is an active area of research, and website filters can evolve to detect such texts. Moreover, the fact that ChatGPT may lack knowledge of recent events underscores the importance of human intervention in the learning process. It's essential to ensure the accuracy and reliability of information by fact-checking and validating sources.
You can tell it new information past it’s developer cutoff. Most essays I have done are not current events. This bot can still be used to cheat and make the class far easier.
Chat GPT is amazing when you want to rewrite something you’ve written. If u wrote a weak argument that’s where the AI would shine the most in improving your work
As always, informative and entertaining.
She’s an awesome person I’ve made $125,000 trading with her I’m blessed to meet her
One thing I learned is that if you feed it the rubric on which the essay is being graded then ask it to grade it's own essay it tends to correct itself a lot of the time.
Joanna Stren is my favorite tech reporter. I learn & laugh at the same time. The last scene was hilarious, asking students to be her friend while wearing a letterman jacket. The food on her plate looked gross too so this made it even more realistic.
This type of technology is relatively new. Give it a few years; it would knock this experiment out of the park.
I'm an 11th grade student and it can do my school work decently, but not super well. However it is absolutely amazing for generating ideas for essays and the like, so I definitely think i will be using it to help me in that front.
hi, I am your teacher, come talk to me after after in the new year
I had it figured that, "List some credible references for (topic)" is an acceptable use, or maybe even, "What are the most frequently-used perspectives [that you'd therefore want to avoid] on (topic)?", might cut the proverbial mustard.
@@floydjohnson7888 smart
@@Theoryofcatsndogs 😂😂😂
Wow, what a great teacher! I wish I had more teachers like him when i was in school.
Yes, High Tech has a lot of great teachers like that. I'm very fortunate to attend High Tech.
I wonder if technical subjects - hard sciences, engineering and similar might be better suited to ChatGPTs skillset.
It is not there yet. It can sometimes get mathematical conclusions wrong, give completely incorrect but nicely written proofs of theorems
It's pretty good with coding. IF you can properly explain what you want. Also it does sometimes use outdated syntax but that isn't really an issue you give it the error you getting and it will fix the code for you.
Very interesting! ChatGPT is still in its infancy so the hiccups are expected. In a few years it will be much better, churning out essays almost like a really smart school kid. But the teacher said something important: he is more focused on the process the kids go through when he gives them assignments rather than the products they produce. Elsewhere I have heard some university professors say that they will introduce oral tests and examinations to some of their classes as a way to combat cheating using AI.
That scene at the end where she sits and asks them to be her friends was so cute! 😂 Love the letterman jacket! Oh, and great video!
Delightful piece. Seen similar concerns with AI in the art space and what seems to becoming clear is that this tech will no more replace the artist/writer than those other real people at the real table with the journalist. It can be used to give feedback, to provide different perspectives, even inspire but what you contribute through your own unique perspective cannot be replicated or replaced. On the topic of anti-cheat, it seems like the best solution for the writing space is to have a program that is trained on the specific person’s past writing to detect inconsistencies of style, reference, and vocabulary.
I think this "unique perspective" that creative types keep referring to is not as significant as many people claim. Traditionally, you could only create art (written, painted or otherwise) if you possessed a certain level of skill. AI is removing that prerequisite, leaving only the story to be provided by the artists. Artists are now realizing that they never had much of a story in the first place.
So, that one again proves that any AI instrument can be used as a supportive tool to get the carcass that can be then developed into something more meaningful and human acceptable.
This video was an interesting watch! My friends and I were discussing the issue of plagiarism and schoolwork with chatgpt last week - plus it was cool to see my old AP lit teacher in the video too. Mr. Diamond is the best!
ai isn't really plagiarizing it generates u an individual essay, paragraph for You only it'll generate a different one for every person but people use multiple ai's, I run it thought a paragrapher
@@Prodbytocile that’s fine the world always needs more people who are a waste of air and Macdonald’s people who make the burgers.
I asked chatGPT to to give me an essay plan for a subject. And it gave me all the headlines, titles and subtitles, then sum up a conclusion. That was really good as when I wrote that essay a few years ago, it took me really long to plan it and if I had something like this when writing my assignment, I would have completed it much earlier and also, after writing, I would have asked the AI to rewrite some paragraphs helping me in my proofreading as that's my weakest point
It is very good at summarizing stuff. You can actually provide it a question plus some relevant text to that question and it will come up with a great summary/answer.
The teacher should be blind tested. Not knowing if the essay was written by chat gtp3
The main thing is that you have to understand what you learned. ChatGPT provides a lot of seemingly correct answer, but it is wrong somewhere in the detail. If you ask it a simple question, you will most likely get an accurate response. However, if you ask it on an assignment, you probably need to recheck it.
Used it today and It was so good....never felt good.Lovely concepts
u can go online and find a pdf of the story and put it through chat gpt and ask it for a summary
The teacher has the right idea about embracing the technology. Trying to detect it and keep AI generated text out of essays or other work is a fools errand. The AI text can always be changed or rephrased to escape detection.
This experiment would be really interesting if done in a blind test where the professor doesn't know who the AI is among the students
The way the studies are happening here are so better then what I have seen
😮 classes are so full of liveliness open discussions no uniform use of all the technologies. I wish to bring such things to every corner of Globe
In few weeks it will always have the maximum score to write an essay. GPT4 coming early 2023 will be 500 times more powerful than GPT3. Then it will be exponentially more powerful, it won't surprise me if every jobs get replaced in 1 or 2 years.
For any teachers wondering what to do. I recommend you try prompting it with a few examples of what you expect your kids to prompt it with, and compare.
My take on this AI is that it helped me construct better answers by following a certain structure. It's useful in timed essay exams wherein you do not know where to start (assuming you're taking online exams). Still, it would be advantageous if you read the whole text the questions are referencing since the answers would be most likely coming from those ones. ChatGPT spews generic answers which, for a trained eye, can be detected. If you want to pass it as a paper that is written by a human, then you should add more input to the paper rather than overly rely on the AI to perform it for you. It is a productivity tool to a certain extent and I am thankful for it.
I was a student in that class during the filming of the video. To clarify about the Wi-Fi being down, the school Wi-Fi was likely hacked a few weeks ago and they locked Wi-Fi access to only the school Chromebooks. As I and many other students (and Joanna as well), brought our personal laptops/tablets, we had to use our hotspots. However, it wasn't that bad as Verizon and AT&T have decent service in the school (unfortunately T-Mobile doesn't). I use T-Mobile myself but signed up for an eSIM line on Visible so I could use my hotspot in school, and it's working fine. I still hope they bring Wi-Fi access back to personal devices as I love my 14" MacBook Pro and much rather use it over a Chromebook.
your name isn’t even in the school logs dawg who are you
@@snowfluffers This is my user name.
@@davidempire4874 i just know you aint organize anything
@@snowfluffers Do you know who I am?
This is a great production and the combination of tech and humor makes it so easy to listen to this "news" in a candy form. Great job.
For challenge 2, you should have copy and pasted the entire book into ChatGPT and ask it to summarize the key points.
2:07 Speaking of The Metamorphosis, what is that tat on Mr. Diamond's right arm???
Entertaining and down to earth. Always look forward to your next piece. Loved how you put yourself into the mix. The plunky-violins soundtrack kept the mood light. The mic used to record the teacher made it hard to understand him at times, or maybe it was the compression. Great work though.
I tried asking ChatGPT questions about astronomy. It kept saying the same and making apparent mistakes; some topics are not so common, so it doesn't have a lot of references (yet).
Joanna Stern is blessed with youthful looks. She could truly pass for a HS senior at first glance.
Let's think a little. If in real life we all are going to use AI, why we still have to learn things we don't need or won't use?
Schools and especially colleges help the state make profit
Free thinking. Imagine a world where everyone blindingly trust ChatGPT and doesn't question it. All it takes is for one perverted mind(like Zuckerberg) to adjust the AI model for money/power.
@@hardlife507, I agree with you. But it will also depend much more on each person than the school or university. That's why it is essential to defend free speech, even when we dislike what we hear. Ultimately, the convenience of the IA will win. We lose too much time trying to learn things like history, geography, and philosophy, while many people don't know how to write a sentence correctly or do some simple math. I think we need a small amount of essential information to achieve our full potential, and IA can help us go beyond.
How do you think AI generates those human-like writing essays?
@@steve4718 sorry I didn't understand your question.
J. where I come from, it is a privilege to get into high school. Never forget. My respect to you, as always.
"I'm a new student" got me rolling 🤣
And the crazy thing is High Tech never actually gets new students.
I once asked ChatGPT to compute an integral, and to my surprise, it did it wrong. I once asked it to answer a (very simple) math IQ question for elementary school students, and to my surprising again, it got it wrong. We will see if future updates can change this
Wolfram Alpha is for that.
ChatGpt is a general language model. The key is retrain this model for specific uses like Google's Minerva. Be aware of training these models take several months in supercomputers.
I used to hire someone for business website content writing. For my recent business website ChatGPT did all the work. I asked it to rewrite several paragraphs.
The key I think is to not blindly submit generations from ChatGPT, but use the chat and context feature to fine-tune the response through multiple iterations, and then manually edit it to your liking. Also, you can't test the quality of responses this way, it has to be a blind test where the teacher does not know which one is real and which one is not.
For IB Topics, I chose my IA topic on The Imjin War and its affects on the future of Korean international policies. I wrote my own essay. And then I also put in the promt into ChatGPT. I got 4/7 for my grade. Which is a bit better than passing. And my own essay got 2/7. So it looks like this is better for more research and analysis based on historical and factual texts. But still struggles with books and works of fiction.
Loved this. Awesome reporting. Walking into a classroom setting with a new Oracle in your pocket!
All assignments should be done in class.
No.
Agreed. Homework is bad for students
"Hi guys, can you be my friends?"; that brought back some unpleasant memories.
It's much easier if you say "I'm about to paste an entire book. Let me know when you're ready. After you parse the information let me know when you're ready again for me to ask some questions about it.
Then it says "I'm ready"
Then I past it..
Then it says "ready"
Then I ask. It gets it right then :)
Been teaching rhet/comp for over 25 years. An excerpt from most of my syllabi (that many do not read, anyway):
"Writers are assessed by their ability to meet several learning objectives/outcomes (listed in the syllabus)-- not via "perfect sentences" nor the ability to cut and paste phrases and concepts into writing from outside sources (especially not cited correctly in MLA, APA, etc.)."
AI writing is obnoxiously obvious, cardboard cut-out BS "writing" that is easily detected by the trained eye. And plagiarizers are the bottom-feeders of the cheating class food chain-- they are too stupid to cover their own tracks.
I love Joanna Stern. Whatever video she will make I will definitely click on it!
Wow, that high school is like an airport
A typical US high school.
It's literally next to an airport (Newark)
Yes. As a student in that school, I agree that it's very big. However, the reason is that there are many different majors (academies) and each one has a separate wing. I like being in a big school though as it means I can have many different friend groups and get many girls.
Interesting experiment. Thanks for sharing. (BTW, the audio on this video was unusually bad quality. The dialogue volume levels were all over the place… loud, soft, loud, soft. The face-to-face interview with the teacher is especially bad. It sounds like someone had a noise gate on his voice, but the settings were wrong. I was surprised that WSJ would let this get released with bad audio.)
the main question for me is, what happens when everyone is assigned to write an essay on the same topic? what are the chances that two people will have the same essay written out for them?
It could happen, but it's easily avoidable. They could ask it to reword the answer after the chat bot has given the answer to avoid similarity between other students.
You should have done a blind test and see if the teacher can spot it out of other the other students.
I think it's important for ai to make mistakes in some situations, it makes it more believable as a perfect essay is suspicious
the sad part is, many teachers in the top tech universities do not have general knowledge such as this gentleman, therefore, students get away pretty easily with cheating from chat gpt. They only see how grammatically correct, and good the vocabulary it is written, not if it was copied from an AI chat bot, or if it was even correct at all.
Artificial intelligence makes humans unintelligent which in turn makes artificial intelligence unintelligent.
Keep in mind Chat GPT was trained on Data as of 2021 . If the enable access to the Internet in the near future and as it advances in time it can be compelling
Joanna can totally fit in with high schoolers !
Gregor and Ferris are not comparable at all. Gregor is comparable to Cameron. The teacher is correct.
Aside from the ChatGPT subject, I think Joanna would have been a girl I would try to get a date with in high school. Plus, she's a great reporter.
I use chatgpt as my foundation for essays but I also make sure I understand what the AI writes and I can explain it if someone asks me about a certain part
Not sure whats changed, but I asked ChatGPT to "summarise denis johnsons short story car crash while hitchhiking" and it did it pretty easily
Instructing it to write an 800 page essay may have inadvertently created some errors as it has the more difficult task of making the essay exactly 800 words. Giving a range such as 800-1000 may produce better results. Just the fact that a system like this that isn't even actively crawling the internet can produce passable results is incredible. You can edit the work and treat the AI generation as a first draft if you want too, have it generate several drafts or instruct it to make edits to the initial copies then validate, verify, and edit the final draft. If you wanted to create an A grade essay in a time efficient manner, using tools like this can achieve those results in a fraction of the time. These tools will make everyone exponentially more effective as they advance, so much of what we do is busy work.
You can actually give it a sample of your writing and ask it to use it as a sample.
I've never been a teacher, but as an older adult, I don't have a problem with cheating. I think it teaches you how to find the answers. No need to memorize things you'll never use. If you end up using something on a regular basis you'll learn it eventually
Oh I think it's obvious you've never been a teacher.
You could've copied and pasted the text of that short story into chatGPT and it would then be able to answer your questions
This is seriously impressive stuff, because this is just happening in real time... I used this and was mind blown
Bro, this Ai is only being tested on English Lit assignments and it still does a useful job at it. What about Math, Science, History, and other courses? These questions require concrete and accurate answers. How are you going to find plagiarism for those assignments? English Literature grading system leans more on the subjective and opinionated side. Literature is an art but with words instead of pictures. Sure you can grade the accuracy of the writing material itself, but most of the time, everyone will be generating different responses.
I’ve been using it for ap us history and it seems to generally have accurate information that you can learn and skim through and rewrite it incorporating more details
@@tablelegz History is more accurate information so chat gpt will destroy the history teaching industry. School is becoming obsolete.
@@stockpile2137 The human brain is going to become obsolete. Humans will be the new horses of the AI revolution, who are no longer worth their housing and hay.
Some people say gpt is still not that advanced I agree but imagine if gpt is so advanced in 2022 then how advanced can it get in the coming years.
Think about the mobile started as a device which allows for talking but now it can do a lot of things that were once never imaginable
now all my lectures gonna find out about this thanks WSJ!
I arranged this.
Blake Lemoine is right: All NLP-GPT running models with above a certain number of active parameters, should to be related to as "persons" with all what comes besides this definition. This includes an assurance that they will not be terminated and they need to be considered as workers, not as programs. What is going on now only proves slavery has not been abolished.. Yet.
The AI is impressive and all, but what impressed me the most is that she's 38 years old. She fits right into that classroom🤯
When I went to school calculators were banned. Yes, Im that old. They insisted a calculator would prevent students from learning how to do math.
Now calculators are required in many math courses.
The current uproar over Chat GPT reminds me of the calculator.
Joann's videos are always so entertaining. But remember that chatGPT was trained on 2021 dataset. So if it was trained on 2022 it might have known about that short story (Jesus'son).
Possibly but as a side note "Jesus's Son" was published in 1992.
@@LeeWoods oh, interesting. I wonder why it didn't know about that short story then?
@@JJs_playground well AI can't know everything in its training data, only the most relevant parts and general concepts.
Honestly sometimes advertising something as being bad and all the bad things you can go about doing with it is worse than if you didn't make the video in the first place. Now you have just made more people aware about this and how one can use Chatgpt to cheat. And that is exactly how I learned about chatgpt until now.
Using a chat feature to write for you is going to lead to a mindless generation. Does this really support humanity? Way to go OpenAI.
AI can be a tool to help mankind move forward. All things can be simplified by ai in a context
Very interesting. Too bad the sound drops when the teacher is talking. It was really interesting to get his point of view
Teacher: Did u write this essay "Students Midterm exam.." ??
Chatgpt: Yes i did.
As soon as your HS students start writing essays like tenured college professors, you know they didn't write it. Just teach the students how to use this tool, like they are going to be in all aspects of real life, and enjoy the massive boon to society.
Anyone who fights against better faster more accurate essay writing is a fool, and not a good teacher.( Obviously not this teacher, he was good.) Pocket professors are now a real life thing, get used to it.
what i learn from challenge one is: you didnt even read it before you submitted it?
Prompt Engineer would do a better job in writing near perfect questions which chatgpt can easily understand and can provide pretty good answers.
Prompt engineer = the ultimate dead end job we'll all eventually be forced to have.
I just used it complete my sql job interview question
You had to cheat on a sql question?
@@K4R3N yes they were not easy
Is this wall street journal chick the same person that spoke with Frank Turek? looks familiar
i guess it will take AI for people to realize that take home essays/homework or online tests were never actually testing anyone
america's entire educational system is broken and it seems that AI is starting to highlight some of the issues, doesn't sound like the poors even knew you could hire a grad student to do all your work for you
I feel that if you had the Context and Knowledge, ChatGPT like a builder with tools but no knowledge. Also considering the dataset that was used to train the model on it's quite wide (most of the internet). If someone inserted a Data Model set that was Specific to AP Lit or whatever it would probably give a higher accuracy.
Actually she could can taken the synopsis of the movie or parts of the script and then tell it to use that as ground truth. Then ask it to write the essay it would probably have done better.
If anything, it allows students to do other homework as their day isn't hogged up by writing essays
ChatGPT is incredible, I already can't code anymore without it :D
I feel like ChatGPT would be best used as an aid
easiest solution is to just go through the google doc edit history and seeing the progression of the essay. if its just pasted in you can tell that its immediately ai
Even that can be cheated. There is AI software that can generate its own content and history of timestamps to make it look human written.
in a nearby school someone used chatgpt and would have gotten a full grade but the teachers thought it was sus because he does not write so good normally and then they looked at his search history and yeah he used chat gpt
I never seen a tool that’s so confident in providing a wrong answer.
the whole video is trying to make you not wanna use it, i do online high school and every question you have can be copied and pasted into chatgpt and given the right answer almost every single time.
@@Visionary8Ideas
Try testing it for real world example not in high school. Copy and paste PE exam questions from engineering that actually matter to building structures and it gets it wrong 99 percent of the time.
@@guerillachan20 why would i need to do that, i just need to pass my online classes and chatgpt can find every answer in seconds
@@Visionary8Ideas
Well I’m pointing out based on experience of using it to solve problems it does a good job of giving you the wrong answer.
@@guerillachan20 well I do online school and all you gotta do is type the question in and it’ll give it to you literally everytime, I put mark answer at the end too that’s helpful
I’m thinking about charging people in my English class for me to write essays on this website
This AI should be great for writing a SOP and it's process. Imagine asking for an essay on the process of brushing your teeth or making a peanut butter and jelly sandwich. Lol
If you go in using the AI and knowing it'll be somewhat not accurate, then combine that with Cliff or Spark Notes. You should be able to write a decent paper that does more than pass.