We've discussed Cohen's work in our book _Calling Bullshit_, but the type of bullshit the Cohen focuses on — unclarifiable unclarity, particularly in academic writing — is not what LLMs produce so it strikes us as far less relevant to this course than Frankfurt's notion.
First, thank you for the link about CoT misrepresentation. I've written a fair bit about this on Bluesky etc but I don't think much if any of that made it into the course yet. We should add this to lesson 6, "They're Not Doing That!"
Your point about humanities courses is just right and encapsulates what we are trying to do. If someone takes the course and engages in the dialectical process and decides we are much too skeptical, great! If they decide we aren't skeptical enough, also great. As we say in the instructor guide:
"We view this as a course in the humanities, because it is a course about what it means to be human in a world where LLMs are becoming ubiquitous, and it is a course about how to live and thrive in such a world. This is not a how-to course for using generative AI. It's a when-to course, and perhaps more importantly a why-not-to course.
"We think that the way to teach these lessons is through a dialectical approach.
"Students have a first-hand appreciation for the power of AI chatbots; they use them daily.
"Students also carry a lot of anxiety. Many students feel conflicted about using AI in their schoolwork. Their teachers have probably scolded them about doing so, or prohibited it entirely. Some students have an intuition that these machines don't have the integrity of human writers.
"Our aim is to provide a framework in which students can explore the benefits and the harms of ChatGPT and other LLM assistants. We want to help them grapple with the contradictions inherent in this new technology, and allow them to forge their own understanding of what it means to be a student, a thinker, and a scholar in a generative AI world."
I'll give it a read. I must admit, the more I learn about the inner workings of LLM's the more I see them as simply the sum of their parts and nothing more. The rest is just anthropomorphism and marketing.
Whenever I see someone confidently making a comparison between LLMs and people, I assume they are unserious individuals more interested in maintaining hype around technology than they are in actually discussing what it does.
Someone saying "they feel" something is not a confident remark.
Also, there's plenty of neuroscience that is produced by very serious researchers that have no problems making comparisons between human brain function and statistical models.
Current LLMs are not the end-all of LLMs, and chain of thought frontier models are not the end-all of AI.
I’d be wary of confidently claiming what AI can and can’t do, at the risk of looking foolish in a decade, or a year, or at the pace things are moving, even a month.
That's entirely true. We've tried hard to stick with general principles that we don't think will readily be overturned. But doubtless we've been too assertive for some people's taste and doubtless we'll be wrong in places. Hence the choice to develop not a static book but rather living document that will evolve with time. The field is developing too fast for anything else.
I think that’s entirely the problem. You’re making linear predictions of the capabilities of non-linear processes. Eventually the predictions and the reality will diverge.
Every time someone claimed “emerging” behavior in LLMs it was exactly that. I can probably count more than 100 of these cases, many unpublished, but surely it is easy to find evidence by now.
Not quite, but it was the closest pithy quote I could think of to convey the point that things can be false for a long time before they are suddenly true without warning.
How about "Yes, they laughed at Galileo, but they also laughed at Bozo the Clown?"
We heard alllllll the same hype about how revolutionary the blockchain was going to be and look how that turned out.
It's a virtue to point out the emperor has no clothes. It's not a virtue to insist clothes tech is close to being revolutionary and if you just understand it harder, you'd see the space where the clothes go.
The post seems to be talking about the current capabilities of large language models. We can certainly talk about what they can or cannot do as of today, as that is pretty much evidence based.
This is an excellent point about scientific writing. We'll add something to that effect.
We have not taught this course from the web-based materials yet, but it distills much of the two-week unit that we covered in our "Calling Bullshit" course this past autumn. We find that our students are generally very interested to better understand the LLMs that they are using — and almost every one of them does, to vary degree. (Of course there may be some selection bias in that the 180 students who sign up to take a course on data reasoning may be more curious and more skeptical than the average.)
This is something we've given serious consideration, having taught a course called "Calling Bullshit" (http://callingbullshit.org) for almost a decade and having authored a book by the same name that gets downranked on various Amazon features because of its title.
But the bullshit is a term of art here, after the seminal 1986 article "On Bullshit" by Princeton philosopher Harry Frankfurt (later published as a little book). We strongly feel that it is exactly the right term for what LLMs are doing, and we make the case for that in lesson 2 of the course. (https://thebullshitmachines.com/lesson-2-the-nature-of-bulls...)
We're also concerned about accessibility for high school teachers etc., and thinking about what to do in that direction.
I'm curious: do you find "bs" to be any less offensive?
FWIW, I don't think you should cave on this. For me, your choice to use it over "hallucination" instantly elevated the insight of the lessons. I also think the authenticity of the voice of the lessons benefits from owning the decision to use it fully rather than compromising with the shorter "bs" version.
In the original drafts I had a long section on this, including some of the history of the GUI, the development of the mouse, etc. It was way too much for the main text when the point is just to set up a metaphor for students who have seen a Mac 128.
That said, we can and should do better in the instructor guide. Thanks for the reminder. I'll add some context there.
After talking to an awful lot of 18-20 year olds (our target audience) we decided we wanted to go with a "scrollytellying" style. I'm not a designer and I've done worked in that style before. After looking into a range of platforms — Vev and Closeread for Quarto deserve particular mention — I felt that Shorthand (https://shorthand.com/) was the best option for rapid development given my lack of experience in this whole process.
In general I've been very pleased. You don't have the fine scale control you do on a platform like Vev, but for someone like me that is probably a good thing because it keeps me from mucking around quite as much as I otherwise would with design decisions that I don't really understand.
The price is a bit steep for a self-funded operation and we're constrained a bit by the need to use their starter tier, but I feel like we are definitely getting our money's worth and customer support from Shorthand has been exemplary.
I'm surprised at the firefox problems; I did almost all the development in firefox. I know it's not your job to fix any of this, but if you are so inclined I'd be grateful for an email me with screenshots or descriptions of where things break.
Many comments about this, so I'll address them here.
We talked extensively with the 18-20 year olds who make up our target demographic and this "scrollytelling" style is their strong preference over the "wall of text" that I and most of my generation prefer.
What your comments make clear is that we need to develop a parallel version that is more less plain text for people who are using a range of devices, for people who have the same reading preferences that I do, etc.
Right now we're entirely self-funded and doing this on spare time but it's clear to me that an alternative version with a very clean CSS layout is the way to go, possibly with a pdf option as well.
I don't want to let versions proliferate too extensively, simply because this is very much a living document. Technologies are changing so fast in this area that many of the examples will seem dated in a year and — while we've tried to be forward-thinknig about this — some of principles may even need revision.
Yes—I should have stressed that. More than anything, we've talked at great lengths about LLMs with over a thousand undergraduate students who we have taught in our courses since ChapGPT 3.5 launched in Nov 22.
