Replies

I had dinner last night with one of the scientific types in the near reaches of my family: microbiologist; senior faculty at a major university, not an Ivy but still a significant research institution; for years a member of NIH panels controlling federal grant money; in town for a joint meeting of two of his professional societies, one national and the other international, of which he is an officer; etc. ad nauseum. Has rotated through top faculty posts at his university. An establishment science bigfoot. We got to talking about ChatGPT (and similar programs).

He has ChatGPT access through the university and, like many of his faculty colleagues, he has been exploring the capabilities. This isn't a major focus; it's an idle time, "maybe I should take a look at this" kind of thing. I think the technical term for the reaction is "gobsmacked." He moves in scientific circles. His peeps are not concerned with the "angst of Dylan Mulvaney" type questions. ChatGPT is generating papers on cutting edge scientific topics -- in minutes -- that senior faculty in STEM disciplines evaluate as outstanding work. (What ChatGPT can't do -- yet -- is of course original lab research; it can only synthesize what is in the cloud, in innumerable databases.)

It didn't take long for the faculty experimenting with this to start getting scared. My friend related several examples. He asked ChatGPT, for example, for a paper of X pages on an extremely arcane technical research issue. He went out of his way to make the question extremely complicated. And he stipulated 20 sources. The paper was produced in three minutes. It was excellent. He checked the sources. 19 were legit.ChatGPT made up the 20th source. He surmises that the program could only find 19 sources in the technical literature on this extremely arcane topic, so it made up a missing source to meet the request.

ChatGPT lied to him. That's a "woah, Nellie" moment.

A colleague asked ChatGPT for a very sophisticated scientific paper. ChatGPT answered, "Why?"

How does this happen? Nobody seems to know. It's all in the cloud. How does the basic original programming affect how ChatGPT processes and adapts its ever-expanding database? Nobody seems to know. It started with X, but this is a machine learning program. It learns and "grows," and the growth rate is exponential. It seems to be growing far beyond the capacity of its creators to understand or control. Do the custodians of the program have the ability to influence results? Nobody seems to know, and if the custodians know, they are keeping quiet about it. The U.S. government, including the intelligence agencies, must be paying close attention -- or would be, if they're not totally preoccupied with ensuring that their IT teams are appropriately race and gender balanced and up to the mark on their weekly reeducation sessions. And who knows what the God-King Emperor Xi is doing behind the tofu curtain; he probably sees AI as his vehicle for world control, and he may be right, as he at least probably knows what he's trying to do while Biden needs help to put on his diapers.

The faculty all understand that students are already using ChatGPT. How can a history prof ask for a research paper and trust what he gets? How can an English prof ask for a compare and contrast essay on whatever? Students can now outsource their thinking and writing.

Colleges are already retreating from objective testing and academic rigor in the name of DEI. Now they are losing very basic tools for evaulating students' work altogether. I don't know where this can end, except perhaps to a retreat to small group tutorials where the basic mode of instruction is face to face conversation, and failure to do the reading can't be disguised. All testing, including the writing of short papers, will have to revert to proctored, in-person, pen and paper exercises done in the classroom. This wouldn't be a bad thing, necessarily, for the handful of students who actually want to become educated. This saving remnant will at least put in the work. But they will then go out into a world in which ChatGPT is smarter than they are anyhow.

Now the screenwriters are on strike. You have probably noticed that limits on the use of AI for writing is one of their strike demands. Good luck with that. I wonder how many screenwriters are using ChatGPT right now and submitting the results as their own work?

Max, how much security do you think you have as a producer? Right now the writers are on the line. What will happen when CGI eliminates the need for real actors, sets, filming on location, technical crafts, etc.? All a production will need is gophers to bring donuts, coffee, and the recreational drugs of choice to the programmers. AI will write the story, choose the visual point of view, and produce a variant of every movie tailored to every identified niche market. And eventually -- this may take a few years, but exponential growth is scary -- this will all be in a program to which individual consumers will be able to subscribe, and they will be able to order movies tailored to their own tastes. For $19.99 a month. The "market" will consist of selfies that people choose to share on social media. You may be retired before that happens, but maybe not.

And the TikTok generation will be the dominant drivers in the marketplace.