Muhaimin, Abdul. "Power of Academic Integrity." 𝘞𝘰𝘯𝘥𝘦𝘳 𝘰𝘧 𝘵𝘩𝘦 𝘮𝘪𝘯𝘥. Jan. 29 2025, Week 10.

 Recently, There was a huge academic purge which stripped us out of quite a few liberties and freedoms. To not beat around the bush, Chat GPT has become quite a big problem affecting the academic integrity of American High School students. The main problem is that it is really hard to find out if a text is written by AI conclusively, so how does it work?


There are two characteristics of a writing which can help identify if a blog is written with Artificial assistance are perplexity and burstiness. Perplexity is how surprising a piece of writing is, human writing has a high perplexity score but AI writing is very monotonous. Burstiness is the variety in the complexity. My previous blog has a burstiness score of 0.65, for reference AI generated paragraphs have a burstiness of 0.16. Unfortunately, formal writing is often written to follow a simple structure because of its inherent nature of formality, so even though this metric is really powerful in finding AI usage in a normal setting it fails in a formal setting. So even though AI generated paragraphs can be sieved with this metric, if the cheater fortifies their paragraph with human modification, it is really hard to separate it from.




One way which is a bit better at this— is watermarking text. So how is text watermarked? Is it just writing Chat GPT after every other word? Well to understand how to encode hyper data in text we must learn how Chat GPT works? AI Chat models are just essentially autocorrect on steroids, they predict what text will come next but on a essentially infinite scale. But, Open AI scientists realized if the most likely text was picked then the text was very boring, so AI randomly picks the text depending on the probability distribution. Here is where the watermarking happens. 20% of the possible words are randomly put on a black list and the AI can only pick the letters in the white list. If a text has a really high concentration of white list words it can be determined almost instantly with an almost 100% probability that they used AI.

Yet, Open AI does not employ watermarking in its text, this is because if there was a definitive way to prove if a piece of text was written by AI, they would lose 20-40% of their audience. This is the proof that Open AI is not making tools for the student, they are making them for their stockholder. Therefore, you must remember that Chat GPT cares about your literary potential. So the important lesson of this blog is the power of integrity and knowledge, learn about all the tools in your tool belt. You may even learn to use them in a safe setting, but never grow dependent on them, because they don’t care about you and they will do anything for your money.

Comments

  1. Yumi LiuJanuary 29, 2025 at 11:01 PM

    Hi Abdul! This was a really informational blog to read. I had no clue of the scientific, mechanical stuff that was happening behind the scenes for ChatGPT. AI writing has always striked me as oddly intricate writing but at the same time possible, for a super intellectual person. However, it all makes sense why AI writing sounds the way it sounds after you explained the function of white list and the black list. Just Curious, what would you think of the possible implementation of an AI tutor? It seems that you are against dependency on AI because of how it inhibits learning, however, in a case where AI would be aiding a student’s growth, how would you think of it?

    ReplyDelete
  2. RitishaFebruary 4, 2025 at 9:26 PM

    Hi Abdul,
    With recent issues of Artificial Intelligence being used in class as well throughout the world to replace human tasks, this article is very interesting! Especially the information you provided with the percentages regarding the “burstiness” scores. The information you provided in the next paragraphs were also interesting to look at. As someone who is not every well versed in the computer world, your blog still managed to interest me and make me wonder about the true mechanisms behind Chat GPT. And I totally agree! AI should only be used as a tool, otherwise frequent reliance on it can lead to stunted development as it hinders your creativity.

    ReplyDelete

Post a Comment

Popular posts from this blog

Ruhika, Week 16 - Memories Bring Back, Memories Bring Back You

Image
As this is the last blog I will ever write, I wanted to share one memory I will remember with all of the people in my class. AP Lang was a rollercoaster of a class… the amount of stress it caused me is immeasurable. However, I am so grateful that I got this experience since I think this class is the one which has the most tightly connected community. Suhas Bathini: I will not forget how terrible of a dancer you are. I am glad that the Gartic telephone allowed me to become friends with you. Ananya Bhardwaj: Your Michael Jackson impression is amazing. I will need a full length cover. Yuhaendan Bhaskaran: You are a bad mafia, letting your hands give you away.  James Chan: Your stage presence is amazing, which was evident in your POAS. I have had the pleasure of having English with you twice and getting to witness incredible performances more than twice. Devyani Choudhury: You whispering under your breath, “[A 3] is not the same to me,” will always make me giggle. Ritwik Deshpande: I a...
Read more

Michelle Li, 3/19 Week 13 - Person of American Significance

Image
David Koresh, my Person of American Significance. Image source: Financial Times You might have noticed my face turning a little red after my POAS presentation this Monday; I was embarrassed. I genuinely thought my performance was awkward and that my memory lapses were too obvious. But my tablemates and friends quickly reassured me that I did well, which I was grateful for, but didn’t truly believe. I think it’s because I don’t notice how I deliver my speeches. When I’m presenting, I switch into a different mindset where I cut out filler words, enunciate things for emphasis, and use hand gestures. It happens near automatically now through a phenomenon called implicit memory ---a type of long-term memory that causes us to unconsciously recall behavior or skills, typically through repeating these habits in the past. Since delivery has become second nature, I only focus on the content of what I’m speaking about—which I’ll admit was lackluster and cut out important information I wanted to ...
Read more

2025- Abdul Muhaimin-Week 16

Image
  This year might have been the most memorable bout of time in my recent history; even after excluding the mind bending outwardly world events, this year was still filled with the highest highs and lowest lows of my life. This blog seems like a farewell letter, partly because this is, after the end of this year I am leaving American High to join Ohlone Connections . It seems strange how I will not be seeing the wonderful friends I made during APeng, and largely in American High. So as we close this year I ask you to think about some of the memorable parts of this year, some lenses tell a more interesting tale. One lens to view the year through, could be the “WAR AGAINST COLLEGE BOARD.” It started with a training montage where we met are lieutenant, Mrs. Smith,  and as the year progresses so did our skills. Who doesn’t remember the sleepless nights spent doing annotations, or mental connections forged through rhetorical analysis. In the middle of this epic we face off against...
Read more