Hold onto your hats, everyone — there’s stunning news afoot. It’s hard to believe, but it looks like over-reliance on chatbots to do your homework can turn your brain into pudding. At least that seems to be the conclusion of a preprint paper out of the MIT Media Lab, which looked at 54 adults between the ages of 18 and 39, who were tasked with writing a series of essays. They divided participants into three groups — one that used ChatGPT to help write the essays, one that was limited to using only Google search, and one that had to do everything the old-fashioned way. They recorded the brain activity of writers using EEG, in order to get an idea of brain engagement with the task. The brain-only group had the greatest engagement, which stayed consistently high throughout the series, while the ChatGPT group had the least. More alarmingly, the engagement for the chatbot group went down even further with each essay written. The ChatGPT group produced essays that were very similar between writers and were judged “soulless” by two English teachers. Go figure.
The most interesting finding, though, was when 18 participants from the chatbot and brain-only groups were asked to rewrite one of their earlier essays, with the added twist that the chatbot group had to do it all by themselves, while the brainiacs got to use ChatGPT. The EEGs showed that the first group struggled with the task, presumably because they failed to form any deep memory of their previous work thanks to over-reliance on ChatGPT. The brain-only folks, however, did well at the task and showed signs of activity across all EEG bands. That fits well with our experience with chatbots, which we use to help retrieve specific facts and figures while writing articles, especially ones we know we’ve seen during our initial scan of the literature but can’t find later.
Does anyone remember Elektro? We sure do, although not from personal experience, since the seven-foot-tall automaton built by Westinghouse for the World’s Fair in New York City in 1939 significantly predates our appearance on the planet. But still, the golden-skinned robot that made its living by walking around, smoking, and cracking wise at the audience thanks to a 78-rpm record player in its capacious chest, really made an impression, enough that it toured the country for the better part of 30 years and made the unforgettable Sex Kittens Go to College in 1960 before fading into obscurity. At some point, the one-of-a-kind robot was rescued from a scrap heap and restored to its former glory, and now resides in the North Central Ohio Industrial Museum in Mansfield, very close to the Westinghouse facility that built it. If you need an excuse to visit North Central Ohio, you could do worse than a visit to see Elektro.
It was with some alarm that we learned this week from Al Williams that mtrek.com 1701 appeared to be down. For those not in the know, mtrek is a Telnet space combat game inspired by the Star Trek franchise, which explains why Al was in such a tizzy about not being able to connect; huge Trek nerd, our Al. Anyway, it appears Al’s worst fears were unfounded, as we were able to connect to mtrek just fine. But in the process of doing so, we stumbled across this collection of Telnet games and demos that’s worth checking out. The mtrek, of course, as well as Telnet versions of chess and backgammon, and an interactive world map that always blows our mind. The site also lists the Telnet GOAT, the Star Wars Asciimation; sadly, that one does seem to be down, at least for us. Sure, you can see it in a web browser, but it’s not the same as watching it in a terminal over Telnet, is it?
And finally, if you’ve got 90 minutes or so to spare, you could do worse than to spend it with our friend Hash as he reverse engineers an automotive ECU. We have to admit that we haven’t indulged yet — it’s on our playlist for this weekend, because we know how to party. But from what Hash tells us, this is the tortured tale of a job that took far, far longer to complete than expected. We have to admit that while we’ll gladly undertake almost any mechanical repair on most vehicles, automotive ECUs and other electronic modules are almost a bridge too far for us, at least in terms of cracking them open to make even simple repairs. Getting access to them for firmware extraction and parameter fiddling sounds like a lot of fun, and we’re looking forward to hearing what Hash has to say about the subject.
I did a double-take at Elektro who was the thumbnail image for this post in my RSS reader. “Is that Thinko the robot from Beauty and the Robot (better known as Sex Kittens Go to College)?” Indeed it was. I’d have used a pseudonym if I’d appeared in that one, too.
I just did a deep dive into using ChatGPT over the last 6 weeks or so, and I am simultaneously impressed, scared, and disgusted.
I’m writing a book and simultaneously learning to be a writer. Basically, learning a skill and taking on a project to promote that learning. I’m using ChatGPT as an alpha reader, as in “how can I make this text better”?
Firstly, ChatGPT uses unconditional positive regard, and it’s easy to get hyped up about a project because it will gush about it and tell you that you’re a natural and your project will transform the world. (Instead, tell it “you are a tech reviewer, you know what works, what sells, and what is a dumb idea. Analyze my idea and tell me whether it will work.)
Secondly, ChatGPT is really good at writing short stretches of text, but the text is what people online typically write and is not ready for publication. Actual writers will take a line of text and shorten it, remove worthless phrases, use targeted verbs instead of adverbs (“whispered” instead of “spoke softly”), and so on.
Also, ChatGPT can write short stretches of text that seem reasonable, but anything longer than a couple of paragraphs and you start to see cracks in the edifice. Point of view is not consistent, central thesis tends to wander, that sort of thing. If you want it to write you a story (say), you have to take the story and upgrade it to be consistent and interesting.
But it’s also creative. It suggests lines that are inventive and unique, funny or ironic, and that come out of nowhere; ie – that aren’t suggested by the source material. Sometimes it’s scary good at this. Ask it to create a poem some time and see for yourself.
And finally, as research for my project I’ve been reading newly published books (similar to my book idea), and I note that some books seem to have sections that feel for anything like they’ve been written by ChatGPT, these books are self published on Amazon, and it makes me wonder…
Just about anyone can type in an idea for the plot line of a book and tell ChatGPT “now, write me the book” and it’ll do that. And that gets me a little scared, because the average public reader won’t be able to tell the difference between a high quality writer (Stephen King) and a self-published ChatGPT novel based on a 2 page plot description.
I wish I got this idea to write a book 5 years ago, because it looks to me like everyone and their dog can generate a book from almost nothing using ChatGPT, and I’m a little nervous that books themselves – great literature, things that are popular and good reading – will be lost in a torrential deluge of books that are “mediocre, but serve their purpose”.
I’m just a tiny bit uneasy over the idea that writing, especially good writing, will soon disappear as an art form.
The em dashes are the biggest ChatGPT tell. “–” Did you prompt AI to write this? A meta joke?
No, I’m learning to write well and just recently discovered the difference between comma and em-dash, where you use one versus the other and so on.
So I’m trying to use more em-dashes in written prose. Trying to use all the tools available.
(Apropos of nothing, I’ve decided not to use ChatGPT to write my words for me. I use it for critique and suggestions, but I’m the actual author.)
If it does it is because people don’t want to pay for it. e.g. newspapers, etc.
Excellent points ; the LLM’s will duly use these suggestions after they scrape hackaday later
Bite my shiny golden arse.
I wonder how many students will figure out that if AI can do your homework, it can do whatever job you think you want.
Am I the only one who can’t tell between “AI” and “Al” at first sight, especially if an article mentions both?
From what I see, the Proxima Nova font used here does provide an alternate lower case “L” that’s rounded at the bottom. Could we at least give it a try? Pretty please?
+1
+1
That’s only half of the problem. Proxima Nova isn’t on any of my devices, and Hackaday isn’t serving a copy of it (and probably won’t because it’s a commercial font from Adobe).
How about a website refresh, with a libre font that makes the difference between Al Williams and his evil clone clear?
N/m, it is being downloaded from typekit.com, it’s just not showing up for me. shrug
I blame the font nerds and design geeks. They all love the sans serif fonts.
I hate them. I find them much less easy to read, to begin with.
Also, if you have a last name with a sequence of ayes and ells, you often can’t tell easily if you have a typo when you enter your name in a web form. Try “William” or “Philip” in your typical shipping address form and see what it looks like. Or “Illinois”.
I think Al should just change his name :-)
Thank you for pointing out the North Central Ohio Industrial Museum! That is just around the corner from me and I didn’t even realize it was there. Next time my wife goes shopping in Mansfield I might have to stop there and check it out.