10 Comments

Very interesting post. I'll point out, if you don't mind, that you did not define "LLM" (Large Language Models, according to wikipedia, "a computational model capable of language generation or other natural language processing tasks. LLMs acquire these abilities by learning statistical relationships from vast amounts of text during a self-supervised and semi-supervised training process").

My experience with Chat has been mostly annoyance. My work involves very close attention to words, and Chat's gloss on them is often deceptive and wrong. And boring. I could see how certain areas of learning might use it more effectively, though.

I have regarded the development of general artificial intelligence as much more threatening to us as a species (speaking of doing without knowing). Your issues with LLMs were new to me, but I'd agree they're sinister too.

Expand full comment

Yes, I think the danger is that thinking will become what navigation without GPS now is.

Expand full comment

well done, Matt!

Expand full comment

merci, mon amie!

Expand full comment

Gee, Matt this is great.

It gets at something I have been thinking about, but points to the heart of the issue more directly than I had been able to articulate.

Expand full comment

Thanks Norman! I appreciate it. Some questions are so difficult from a practical standpoint, but this one seems straightforward, at least in terms of what we ought to do.

Expand full comment

"And so when you use a calculator, it may be the case that your mental math skills slips a bit."

The objection to calculators is that for many students their math skills slip more that "a bit". Sure, if for some reason you have to determine something like the square root of 157.6 to five decimal places, then a calculator is absolutely the best choice. But this can be a slippery slope, and some people will automatically use a calculator for something like 4 times 5, and quickly their basic math skills start to atrophy. This isn't theoretical speculation, and it's something that many math teachers experience in the classroom. It's also analogous to what Prof. Fritts is reporting. So the tool that is so useful for an experienced practitioner can be a liability for a beginner because they it allows them to avoid mastering the fundamentals of the discipline.

I think this is the big concern with LLMs and writing. Joan Didion is not going to have her writing chops degraded by ChatGPT, but the average student in a freshman comp class might be very susceptible to this.

Expand full comment

I couldn't use a calculator in school when I learnt basic arithmetic and I remain convinced that this was good for me. Even nowadays when my calculating skills have atrophied a bit, I can still do math on my head (or on paper, for that matter), which becomes really useful when in the supermarket with a limited amount of cash, or when sanity-checking results (I do computer science research). I have friends and colleagues who cannot do this (and I've met US undergraduates who were positively stunned by me doing long division with a paper and a pen) and unless they have access to a machine, there are things that they simply cannot do.

So even something as specialized and limited as a calculator reduces human capacity in meaningful ways.

Expand full comment

It's interesting how the general public has panicked over AI in general, thinking it will take over the world without understanding all the different types of AI. That LLM basically searches and extracts info from documents. It doesn't have the ability to think critically, ot discern fact from fiction, for example. I applaud you Matt for sitting down with your students to explore ChatGPT in class. Can they make the jump from simply producing a document with an LLM to critically reviewing the "draft" and creating a much better assignment?

Expand full comment

Well, the "general public" has been primed both for the panicking and the lack of understanding by the people who stand to make money from both as it regards LLMs. :)

Expand full comment