3 ms·

more an indictment of the midterm in this case, only one question actually required any form of math and it was extremely basic. The rest you could BS your way

more an indictment of the midterm in this case, only one question actually required any form of math and it was extremely basic. The rest you could BS your way to a partial score just based on the context of the question. The California and Texas one was a layup

rictic 4y ago
I agree, but that doesn't explain why it did better than the rest of the class
- Tenal 4y ago
  [dead]
- cyber_kinetist 4y ago
  The majority of the test is basically kissing the professor’s feet to make him feel his opinions validated. These kinds are tasks are perfect for GPT-4 to generate (whereas for the students, they have a bit more dignity than that…)
  - Balgair 4y ago
    'Fun' Scenario: Some AI company comes along and starts selling AIs for 'educational purposes'. Somewhere in the 100's to 1000's of dollar range, not quite as much as a textbook, but a double digit fraction of that cost. They start training the AI based on the feedback they get from what the students have turned in. So, if a student uses their AI as an 'assist', the company then mandates that the student send back to the AI model all of the comments and grades that were given on the submitted work. So the AI will learn the school, department, and professor's quirks. Think about that selling point! Who are you going to choose to spend your cash on? The AI 'tutor' that is trained on a generic dataset, or the one that is specifically trained on your professor/TA/teacher/etc ? Sure, it's all marketing maybe, but then again, maybe it's not. Multiply this over all the students in those 400+ person classes, and you've got a very large addressable market.
qgin 4y ago
The midterm might not be a great one, but still what software could have received an A five years ago, just given the raw text dumped into it as the input?

3 ms·

more an indictment of the midterm in this case, only one question actually required any form of math and it was extremely basic. The rest you could BS your way

by ren_engineer 4y ago

rictic 4y ago
I agree, but that doesn't explain why it did better than the rest of the class
- Tenal 4y ago
  [dead]
- cyber_kinetist 4y ago
  The majority of the test is basically kissing the professor’s feet to make him feel his opinions validated. These kinds are tasks are perfect for GPT-4 to generate (whereas for the students, they have a bit more dignity than that…)
  - Balgair 4y ago
    'Fun' Scenario: Some AI company comes along and starts selling AIs for 'educational purposes'. Somewhere in the 100's to 1000's of dollar range, not quite as much as a textbook, but a double digit fraction of that cost. They start training the AI based on the feedback they get from what the students have turned in. So, if a student uses their AI as an 'assist', the company then mandates that the student send back to the AI model all of the comments and grades that were given on the submitted work. So the AI will learn the school, department, and professor's quirks. Think about that selling point! Who are you going to choose to spend your cash on? The AI 'tutor' that is trained on a generic dataset, or the one that is specifically trained on your professor/TA/teacher/etc ? Sure, it's all marketing maybe, but then again, maybe it's not. Multiply this over all the students in those 400+ person classes, and you've got a very large addressable market.
qgin 4y ago
The midterm might not be a great one, but still what software could have received an A five years ago, just given the raw text dumped into it as the input?