Welcome back to Foxing Around, where I talk about something other than Birds, Books, and Bikes.
At work, I have been tasked with using a Language Learning Model, or LLM, to make my work more efficient. These are popularly known as “AI” despite not being artificial intelligence. I am generally “anti ai,” especially in the creative field, as it has been homogenizing visual art for years, and the written word is next—my beloved em dashes have been called “sus” because LLMs are equally fond of them—though they tend to put ugly spaces before and after. I’ve avoided LLMs as much as I can, but recently I’ve been forced to cross my personal line in the sand, at least in my professional life.
LLMs are much more powerful than the original chat program, ELIZA, which I used gleefully back in the olden days of typing away at a monochrome screen in the computer lab. ELIZA repeated parts of phrases back to you in the form of questions, like a caricature of an analyst, but it was enough to make many people feel “heard,” and if there’s something people in an alienated culture where we stare at bleeping glass tiles instead of talking to one another want, it’s to feel heard.
I’ve been following MIT sociologist Sherry Turkle since her first book, The Second Self, was introduced to be by a Rutgers Professor who taught an interdisciplinary course on literature and science, where William Gibson’s short stories from Burning Chrome were on the syllabus, among others. Turkle has been watching humans interact with technology for over forty years, and is not an “anti” anything; she sounded the warning bells about social media, how Zoom calls are not the same as being in person (in her book Reclaiming Conversation, which I can’t recommend highly enough). Lately, she’s been talking about humans with relationships with bots, the latest evolution of Eliza.
I don’t use chatbots outside of work. This is all written by me, not “polished,” and while as a lonely 13-year-old, ELIZA was fun to mess with, I don’t chat with a bot and make it pretend to be Karlach from Baldur’s Gate 3, as tempting as that might be. Karlach is a butch half-demon with a heart of gold, voiced by Samantha Beart. A lot of chatbot activity has replaced “fanfic,” where fans chat with character bots of their own invention, which are uncannily similar to popular ones. I bet if these existed when I was eleven, I would be chatting with Ariel and Ookla from Thundarr the Barbarian.
In my professional life, I’ve been using an LLM to build workflows that automate my most tedious tasks, by using a chatbot. If you haven’t been forced to use an inept chatbot in place of human customer service yet, I envy you; instead of browsing a FAQ page, they mimic chatting in a chat to a customer service rep in a little window. Except when I’ve worked with a human at the other end, I’ve gotten results; these tend to spam you with search results from their website, and insult your human intelligence by claiming to be Artificial Intelligence. There is a lot of human labor behind “AI” and content moderation, which 404 Media covered recently. The fake “AI Grocery Store” that Amazon ran was actually a thousand underpaid Indian workers watching people load their grocery carts and tallying it manually.
The higher end LLM models that my employer has subscribed to are much more effective than a crap chatbot used to avoid having any real form of customer service.
It’s so competent that it’s a little scary. I’ve had to correct it here and there, but if I am specific, it does what I ask, and then saves what it learned as a skill that I can call. It’s saved me plenty of time, even counting in the time spent configuring it. This is called “prompting,” and once you get good at it, it is frighteningly addictive.
So much that I have begun dreaming about having a robot concierge, like JARVIS in the Iron Man movies, that can fix every little thing. This isn’t odd; personal assistants are luxuries that the rich relish, and who wouldn’t want one that isn’t exploiting another human’s labor? I have a robot vacuum and floor mop that works great. If they made one that could scrub shower tile, fold laundry, and clean stovetops, I don’t want to share what I’d be willing to pay to have that time back in my life.
The frightening part isn’t the willingness to subcontract all activities that are the slightest inconvenience. “Clean the stove, take out the recycling, and reorganize my books by author’s last name.” That’s an easy one, but by not organizing them myself, I wouldn’t have found books that I forgot I had.
What’s frightened me is that in two weeks, I’ve become accustomed to using the language of prompts in my internal monologue.
The other day, during a nap, I caught myself thinking, “sleep another half hour. Then do the laundry and begin packing for your trip.”
Saying checklists out loud may be normal to some of you, but I normally talk to myself in the second person. “Okay, you’ve got another half hour to sleep… then get ready for the trip.” Maybe it’s the LLM’s frightening competence that made me want to “prompt myself.” No matter what it was, I don’t like it.
If I had a personal Jeeves-bot valet, I would spend a lot more time reading books, looking at birds and other wildlife, and riding my bike. But I like figuring things out. I don’t want to turn into a Bertie Wooster who can’t survive on his own; I can’t imagine Gussie Fink-Nottle asking someone else to collect his newts! The joy is in the discovery. Over the past few days I used Merlin to help me find birds for the “Great Big Day.” Merlin does not use AI, but it can identify birdsongs, which makes you know what to look for up in the trees. And it will tell you some of the behavior, show you various plumage of male and female. This helped me see my first Common Yellowthroat the other day! My photo is terrible, or I’d share. (LLMs can’t make the birds pose outside of foliage … yet.)
One of the tasks I used the LLM for recently was helping pore through spreadsheets to help me solve a problem. But it was still me who solved the problem; my employer is adamant that we are to use this as a new tool, that it is only as smart as how we use it, and its great value is freeing our time to think and work on important projects that can’t be solved by a chatbot. And that is what I am using it to do.
On all my web search widgets, I have AI turned off. It is a processor-intensive application and wasting it, as I’m sure you know, wastes electricity and water to the tune of dozens to hundreds of web searches. Another caveat: because websites can have code that weaponizes AI chatbot web searches, it is unwise to use give an AI chatbot access to any web browser that is connected to your password manager. The bot searches in the background so you may not even know what site it hit, which can mimic a real one and get your credential data when the bot attempts to login.
I am looking for a small rugged camera so I can take better photos while I’m on my mountain bike. It would be simple to use an LLM, or the AI search all the engines are foisting, to “help.” But I enjoy the search, and I sure as hell don’t trust an LLM knows which sites are legit and which are just Amazon link farms, to skip reviews tagged “as part of an online promotion,” and so on. I could probably prompt it to do so, but where’s the fun in that?
Fascinating about (almost?) crossing over. I'm retired and not yet needing AI. My imagined usage would be for compliments. Flattering insights about my writing. Encouragements to not give up. Unlike fellow writers I don't long for fame, just the feeling of being understood.' At this point I might not care how I get it:) A self-made self might do the trick. But I wonder about the hidden costs.
If B&H photo out of NY still exists, they're the ones to ask about the camera.
KC likes using chat for workflow stuff, but I'm adamantly against it. I need to figure stuff out for myself. But my job isn't as complex as his and intrinsically rewards curiosity.