How I Defeated A.I. Plagiarism | How Smart is ChatGPT ACTUALLY? | ChatGPT Explained

0:00 - Prologue
1:47 - Verifying Attack Modes
3:36 - Test College Level Essays, "Hart Crane"
4:29 - Test Another Essay, "Their Eyes Were Watching God"
5:59 - Knowledge Test - NBA Stats Example
7:24 - Testing Old Information
7:39 - Harmful, Illegal Acts Test
7:54 - Bad Words Test
8:24 - Testing Large Language Model Connected To Browser
9:19 - Route Plagiarism Detection
9:39 - Language Models Mathematical Background
10:14 - Language Model from a High Level
10:29 - Brief Explanation of Neural Networks
10:59 - Tokenization
13:49 - Reverse Engineering and Building
18:29 - Summary and Discussion

----------------------------------------------------------------------------------------------------

Blog post on the topic:

GitHub Repo coming soon.

Free Code for my book, "Probability for Lemurs," - Making Probability Easier to Understand:

My YouTube Channel and Video Analytics:

[fredrikjonsson]

Very nice video. I can think of a few people with audiences many orders of magnitude greater than yours who pretty much are shown to simply not have a clue by it. When someone claims "ChatGPT is smarter than you!" they could use watching this.

[dustinjensen]

Thanks for explaining all this so well, Patrick... highly entertaining and informative!

[elip]

Great video, you're a very underrated channel. Aren't A.I language models still applicable to a fair extent for plagiarism if the majority of a generated essay is modified, insofar that the framework of the essay and many key topics are outlined/completed for you? Wouldn't this pose as a strong time-saver for students whilst also theoretically lying below the threshold on your system? I'd love to hear your thoughts, but again I still recognise that suspicious teachers can drill the student on their knowledge of a text/work.

[sandyskelaney]

Thanks for the explainer. Good stuff. As a teacher, I’m ok with not defeating GPT but rather learning how to adapt and build skills needed to be successful in an AI augmented future. I’d like to see AI used together with critical thinking, group process and oral presentation to enhance deeper learning experiences.

[sionsmith]

Really insightful, do you have thoughts on image ai detection?

[coopergates]

Remember when Microsoft released TayTweets? I'm sure ChatGPT has more failsafes, but still, blatant propaganda by reversing figures to change the result of a comparison is quite a bug, or the lack of clarification of info that is "effective as of 2021"

[blisphul]

I know a professor that just got an AI plagiarized paper, and it was so off topic, he just gave the student a 50%. Normally, the students are reported for plagiarism for non-AI cheating, but in this case, it's easy enough to grade for quality and whether the paper is on topic. He doesn't actually mind if students use it to speed up their writing and just use it as a base, since he does recognize it as a useful tool, but it's just that, a tool.