ChatGPT3 Passes Its Finals
But that doesn't mean it's getting an MBA. Yet. - Plus Why new MacBooks are slower
Professor Christian Terwiesch of the University of Pennsylvania’s Wharton School decided to give the final exam for his operations management MBA course to GPT-3. It passed! With a grade of B-.
Terwiesch found that it did very well at basic operations management and process analysis questions but made mistakes at 6th grade level math. It was not good at handling more advanced process analysis questions either. However ChatGPT3 responded well to human hints and even learned over time so the hint was no longer needed. Terwiesch also had ChatGPT3 create exam questions. He found them well worded and at time humorous but they needed "substantial adjustments before becoming usable exam questions."
So what were his takeaways?
- Terwiesch believes his experiment shows that we still need a human in the loop in order to take the best advantage of ChatGPT3. It can't run things on its own.
- He will not allow ChatGPT3 to be used during his exams. He compared using ChatGPT3 to calling a friend for the answer.
- He says he will use ChatGPT3 when teaching case discussions to emulate consultants who sometimes have compelling recommendations that are wrong.
- And he said he will now need to come up with more challenging assignments that push students to think more creatively in ways that ChatGPT3 can't.
- Plus he intends to use ChatGPT3 to improve his own productivity as a teacher. The test questions it created needed work but even so, he thinks it might cut his test creation time in half.
In his paper, he showed how ChatGPT3 answered several of the questions on the exam. One interesting effect was how ChatGPT3 handled the answer to Question 3. The question said "To determine the bottleneck in the cranberry processing plant, we need to identify the step in the process with the lowest capacity." And then it gave lots of details the student would use to answer the question. ChatGPT3 incorrectly calculated which step had the lowest capacity. But then ignored that calculation and correctly identified the bottleneck.
That's interesting because it shows that ChatGPT3 isn't thinking linearly. It's assembling answers without always correctly relating the sentences to each other.
----
Previously on how Apple handles Solid State Storage:
The base model M1 and M2 versions of the MacBook Air, each offered 256GB of storage but they did it in different ways.
The M1 version split the storage between two 128GB chips.
The M2 used one 256GB chip. That's not a problem for most uses but it means you can't read and write simultaneously from two chips.
That can slow down transfers of large amounts of data and cause slower performance if the RAM gets filled up, and the drive is used to swap data.
Well now that same effect is happening in the MacBook Pro. The new BASE level 14-inc MacBook Pro running on an M2 chip has slower read write speeds than the older version with an M1 chip.
Previously iFixit noted that the M1 version had two NAND chips on the front of the motherboard and two on the back for a total of 512 GB of storage. 9to5Mac's Derek Wise opened up the new M2 Pro MacBook Pro and noted that only one storage chip is visible on the front of the board. He didn't look at the back but it stands to reason there's only one on the back too. Once again the difference will not be noticeable except in a few edge cases.
That only applies to the base level 14-inch MacBook Pro with 512GB of storage. And Ars Technica found that M2 MacBook pros with higher levels storage -- 1 TB and higher-- had faster internal storage speeds than the comparable M1 MacBook pros.