Closed aolney closed 1 year ago
This would be relatively easy to do with the Macaw work from last year
This is a great thing to do. Can you also make glossary-type short answer flashcards? Stimulus - One of 12 sets of cranial nerves. It enables movement in the eye's superior oblique muscle. This makes it possible to look down. The nerve also enables you to move your eyes toward your nose or away from it. Response - Trochlear nerve
This is a great thing to do. Can you also make glossary-type short answer flashcards? Stimulus - One of 12 sets of cranial nerves. It enables movement in the eye's superior oblique muscle. This makes it possible to look down. The nerve also enables you to move your eyes toward your nose or away from it. Response - Trochlear nerve
Did you get this example from the text? If so, it would be good for me to look at the original page to see how it fits in the context of the chapter.
Jeff's stuff can be used for definitions, but what you have here is pretty elaborated and goes beyond what I normally think of when I think of a glossary entry.
No, I made it up; ideally, one could tune an algorithm to make such things to make them more or less detailed/abstract, so perhaps that is set to elaboration =10 and setting it to elaboration =1 would result in Stimulus - cranial nerve for the superior oblique muscle
I submitted a paper to the AIED LLM workshop that shows Macaw can do this an be as good as humans if you solve the duplicated answer options issue. BingChat is also very close to humans but tends to not follow the prompt correctly.
Pragmatically, BingChat is easier to work with for this unless you have a graphics card with ~45GB of RAM or a CPU with 256GB RAM, which is what Macaw 11GB needs
Use BERT to create options; might require upgrading AllenNLP