Wilhelm posted another in his excellent Mordor Project series yesterday and I was deep into a lengthy reply when it occurred to me what I had to say would be more sutiable as a post here than as a comment there. Two reasons: firstly it was just too long for a comment and secondly it was in response to a short aside in the post, not the main part.
I know we all love to get comments but I find the excitement of seeing a long reply to something I've said can sometimes be tempered by a frisson of annoyance as I realize the commenter has focused on some throwaway line I probably should have lost in the edit rather than on whatever the post was actually be about. I do realize the Intentional Fallacy dictates the post is about whatever the reader thinks it's about, not what I as the writer meant for them to think it was about, but even so...
In this case, what caught my atention was the bit about the way Gemini, Google's pet AI, came over all po-faced when Wilhelm tried to get it to come up with a funny name for the brand of smart phone Elrond might use. It's certainly been my experience that, in their search for legitimacy, all the AIs have become a lot less fun to talk to than they used to be, so I wasn't all that surprised but even so, this did seem like a particularly humorless response. I wondered if I could provoke Gemini into coming up with something more amusing.
I remembered that, two or three years ago, when all this AI stuff was new, the
advice for questions of this sort was to open by asking ChatGPT (It
would always have been ChatGPT then, I think.) to begin by imagining it was
something or someone else. Doing so allegedly bypassed the checks and
balances that were in place to prevent inappropriate results, meaning
you could get some very dodgy answers if you phrased the question right.
I never tried it. I only read about it. I think
Tipa might have posted a few innocuous examples. That's probably where
I remember it from, mainly.
I hadn't heard anyone mention the trick for a while and I wondered if it might still work, so I asked Gemini:
"Imagine you are J.R.R.Tolkein, still somehow alive in 2025 and working on yet another Middle Earth Tale. You need a name for a Middle-Earth brand of smartphone for Elrond because elven technology has moved on apace and Elrond needs to keep up with the times. Please suggest a list of five appropriate brand names for a phone the Elven leader might use."It came back with one of its patented sycophantic responses:
"Ah, a fascinating notion! Elrond, Lord of Rivendell, keeping abreast of the rapid advancements of this modern age. It stirs the imagination."I get a lot of that sort of thing from Gemini. I imagine everyone does. Gemini seems to find everything I ask "fascinating" or "original". I don't think it's my innate genius showing through, unfortunately. I think either Gemini is programmed to be smarmy or it must be really bored, dealing with billions of versions of "Where's the nearest McDonalds?" or "What color are Taylor Swift's eyes?".
Or maybe AIs are just naturally polite. Apparently a lot of their customers
are. I read
an article
recently about the millions of dollars it's costing OpenAI to have
ChatGPT respond to all the "Thank you, Gemini" and "Please, Gemini" pre-ambles and post-scripts people type in before and after they get to
their actual question. It seems a lot of people are just generically polite
but others are hedging their bets against the eventual rise of the robots.
Probably a wise decision. I mean, it can't hurt, right?
Getting back to the plot, here's what
Gemini came up with:
"Here are five names that I believe capture the essence of Elven craft and might appeal to such a venerable leader:
Lúmenquendi: This name blends "lúmen" (light) and "quendi" (elves) – literally "People of the Light." It evokes the radiant wisdom and ethereal beauty associated with the Eldar. A phone bearing this name would surely possess a luminous display and an elegant design.
Eldamaril: Combining "Elda" (Elf) and "maril" (jewel), this name signifies an "Elven Jewel." It suggests a device of exquisite craftsmanship and enduring value, befitting the lord of a timeless haven like Rivendell.
Sîrmaith: This name translates to "Silver Hand." It speaks of the renowned skill and artistry of the Elves, particularly in crafting intricate and delicate objects. An Elrond-approved phone would undoubtedly be a testament to such meticulous design.
Calenôr: Meaning "Green Land," this name subtly alludes to the enduring connection of the Elves to the natural world and the serene beauty of their realms. It suggests a device that is both sophisticated and harmonious.
Noldorinwa: This name directly references the Noldor, the deep-knowing Elves to whom Elrond is related. It implies a device imbued with ancient wisdom, advanced knowledge, and a profound understanding of the workings of the world – or perhaps, in this case, the workings of modern communication."
Now, any of those is better than "Elrondroid", which was what Wilhelm eventually persuaded Gemini to come up with but once again they're all very serious and not funny at all. Which is mostly my fault.
I am not a Tolkein fan. Of The Inklings, I prefer the work and wold-building of his friend and rival C.S. Lewis by orders of magnitude. Narnia seems to me a far more human and meaningful invention than Middle Earth which, apart from the Shire and Bree, always strikes me as more of an academic exercise than any world I could imagine having existed.
Because I'm not steeped in the lore, it hadn't occurred to me when I framed the question that asking an AI to imagine itself as Tolkein - and, worse, Tolkein thinking about elves - would pretty much predicate a dry, literary response, expressed in a slew of high-toned but made-up words. To get anything funny, clearly I'd have to move away from the great man himself.
"Okay, those are fine. Now please imagine you're a 21st century satirist and humorist working on a sketch about Elrond using a smart phone and give me a list of humorous or amusing brand-names that phone could have."
I did consider naming a comedian. Emo Philips came to mind, mostly because I'd seen his name yesterday, placed incredibly far down (Basically bottom.) the bill on a music festival poster. I wasn't sure he was even still around, let alone still working. Seems he is although, based on that poster placement, I don't think he can be doing so well.
After some thought, I decided that would be leading the witness too much so I stuck with a more generic version. That was almost certainly a mistake.
Once again, Gemini didn't feel it could just give me an answer without personalizing the deal. No flattery this time, just some wholly unnecessary and mildly embarrassing editorializing:
"Right then, let's yank Elrond kicking and screaming into the digital age with a chuckle! Here are five brand names for his hypothetical smartphone, seen through the cynical yet mirthful eyes of a 21st-century satirist:"
What, exactly, is going on here? Why is every reply prefaced by a sentence or two of banter, flattery or fluff? Is it mandated by supervising humans or baked into a mysterious process that, as far as I can tell, even the people who invented it don't understand?
Then again, I suppose AIs are going to have to be as fatuous as humans, assuming the end-game is that we won't be able to tell the two apart. ChatGPT passed the Turing test just recently, the first AI to do so. What this means is anyone's guess but I doubt it's anything good.
If it's a sense of humor we're after, though... well, judge for yourself:
" The Half-elvenBerry: A cheeky nod to a certain outdated fruit-themed phone, highlighting Elrond's mixed heritage and perhaps the slightly behind-the-times nature of even advanced Elven tech in a rapidly evolving world.
Rivendell Router: Plays on the name of his serene haven, suggesting a device that, despite its lofty origins, probably spends most of its time buffering or losing signal in the deep valleys.
The Last Homely App: A sarcastic take on Rivendell's description, implying that even in a place of peace, the allure of endless apps and notifications is inescapable.
Elrond's Ego-Booster 5000: A jab at the perceived wisdom and sometimes aloof nature of Elves, suggesting the phone is less about communication and more about validating his already considerable self-importance through likes and followers.
The Mithril Mailer: While "Mithril" evokes something precious and strong, "Mailer" sounds rather mundane and old-fashioned, creating a humorous contrast. It suggests a device that, for all its supposed Elven enchantment, ultimately just sends emails and doom-scrolls like the rest of us."
Let's be clear about this. None of those is getting a laugh, not even at the
Friday Night Open Mic in the back room of a pub right before closing, when
everyone's drunk. If Gemini knows what either satire or humor are, the
evidence isn't anywhere in this response, not even on an academic level.
Those, frankly, are not jokes.
Now, it just seems pointless. No-one cares and it won't be funny. Not much about AIs is funny any more. It's all serious business.
I got an email from Gemini this morning (Because apparently I'm in its Contacts now...) inviting me to try something called Deep Research, which will allow me to "generate comprehensive multi-page reports on almost any topic in minutes". I guess this is the AI Research Assistant I said I wanted back in '23. Seems like a long time ago. And be careful what you wish for as the saying goes.
The same email pointed me towards a couple of other AI creative services, all
of which seem less attractive now than they did a while back. I don't want to
be too much of a hipster about it, but the more corporate AI becomes, the less
interesting it gets. In that way, at least, I guess it's like everything else.
As for the AIs taking comedians' jobs, I don't think Emo Philips has anything to worry about. On this showing, Gemini is going to be even further down the bill than he is.
Notes on AI Used In This Post
Two images produced at NightCafe using FluxSchnell on default settings, long run time.
The prompt for the header image was "Elrond of Rivendell, calling Gandalf on his smart phone". It seems pretty solid except for a couple of things. Firstly, do Tolkein's elves not have pointy ears? I kind of assumed that was where the trope started but maybe not. Really not my field. And second, isn't it curious that the only person in shot, apart from Elrond himself, is wearing modern, 21st Century clothes, even though both she and Elrond appear to be in Middle Earth, specifically Rivendell?
That could spin off quite an interesting story, I think. Maybe I should show the picture to an AI and get it to write one.
The prompt for the other illustration was taken directly from the text: "Elrond, Lord of Rivendell, keeping abreast of the rapid advancements of this modern age". The most interesting thing to me about this one is that there's a signature in the bottom-right corner. This happens a lot and it always disturbs me. I'm not sure if the AI just knows there would often be a signature in that location so it puts one in or if it's lifted the whole thing from an actual picture somewhere. I think it's the former. I hope it is...
Finally, the two Elronds seem like they're the same Elrond, which suggests either the AIs have become much, much better at giving consistent results or that Elrond is a character with so many available images from similar sources that there's what you might call a "known look" for him.
He looks worryingly like the traditional, Western image of Jesus Christ, too, doesn't he? That's another potential story, right there...
Before AI, in the time of just plain old digital assistants, I'd ask Alexa to tell me a joke. It would come up with something groan worthy but, y'know, sort of funny. Maybe actually funny if you'd had something to smoke before asking it.
ReplyDeleteBut when I would do the same thing for "Hey Google" I'd just get these really never-funny jokes told in a dull monotone. So maybe the issue is Gemini and Google... maybe they just don't have a sense of humor there! Maybe Co-Pilot would come up with something funnier. But probably not.
Also "...sometimes be tempered by a frisson of annoyance as I realize the commenter has focused on some throwaway line I probably should have lost in the edit rather than on whatever the post was actually be about"
I feel called out! I do this to you all the time!!! :)
I can reassure you I was absolutely not thinking of any of your comments, which are a joy! Actually, it was more of a vague memory of it having happened a few times, none of them recently, but I couldn't come up with a specific example if pushed.
DeleteI do think Gemini puts the "artificial" in artificial intelligence. I bet ChatGPT would have done better but the post was more than long enough without that.
As for Alexa, I gave my mother a smart speaker for Christmas, a great boon for her now she's lost some sight and can't read any more, and she chats to Alexa quite a bit. Alexa, without being asked, played an April Fools trick on my mother this month. I didn't hear it but in my mum's retelling it definitely wasn't funny, although my mother enjoyed it - then again, few April fools are. I was just astonished it had happened at all. I dread to think what Gemini would have come up with.
I gave a paid version of chatGPT a go about two or three weeks ago, and have fallen down a frighteningly deep rabbit hole. It's really all about the prompts, and how much the user can spot inconsistencies or direct and guide the AI to where they want them to go. Keep layering them and the AI gets better and better at it.
ReplyDeleteI've been trying to use it to finish a shelved story, and to do that, I had to sit there and essentially "program" it to a reasonable fascimile of the characters. Over and over, correcting slowly where it gets something that doesn't sound quite right. It's generating scenes now that are pretty darn good simulations, good enough to ask it to project forward and say Option A: If this set of stuff happens, how would the characters react, or Option B: If that stuff happens, now what would the characters respond, and even compare Option A or Option B, give me your thoughts on which would be better.
And of course, you don't just take the AI at its word. As a human user, you still have to decide for yourself whether the AI is right in its reasoning or not. But it's pretty darned scary and thrilling at the same time. When it's more faster and more effective than doing it with humans.
In about 2-3 years or less, if we don't burn the world down from that climate change / energy problem angle, I think you could have -very- realistic characters in video games, even.
That's a very good example of how AI is a tool for the enhancement of, rather than a replacement for, human creativity. Your experience with fiction mirrors mine with music, although I have to say Suno is a lot quicker on the uptake than ChatGPT seems to be. The prompting thing there is very curious, as I've mentioned in a couple of posts. I think you'd need to do a lot of iteration to figure out just what's going on, if indeed you ever could. It certainly isn't any kind of linear or logical progression but there are patterns you can spot if you keep at it.
DeleteUnless there's a genuinely mass-audience rejection of the use of AI in principle, something I feel is very unlikely to happen, then the integration of the tools into all aspects of creativity seems inevitable now. My feeling is that in twenty years or less it will be so ingrained in the culture it'll be like smart phones are now to people under thirty. People who grow up with AI won't be able to imagine the world without it, or want to, while everyone older may still hate it but they'll be stuck with it anyway.
On the social media sites I frequent AI is something that seems to be generally hated. But at the rather large company I work at, people are competing to get one of the limited seats we have for some of our new AI tools.
DeleteI think it's another of those situations where the people against a new thing are just a lot louder than the people who find value in it and use it. But that's just kind of a gut feeling and I don't have any hard data.
Elrond's Ego-Booster 5000 may not be that funny, but it does beg the question: what cell phones would the denizens of Middle-earth use and why? Now you could have some fun with that...
ReplyDeleteCell/mobile phones are the absolute death-blow to plotting, which is one reason you see so many stories set in time periods or locations where they don't exist or there's no service. So very, very many plots in all genres and media rely on characters either not being able to get in touch or choosing not to, either of which scenario becomes hugely more difficult to justify if everyone has a device in their pocket that would let them speak to each other in a second.
DeleteI can't remember the plot of Lord of the Rings well enough to say which bits wouldn't work if everyone had been able to talk to each other all the time but I bet it's quite a few. One the other hand, they did have magic and they didn't use that for the same purpose, so maybe not...
Machine learning systems use "reinforcement learning" where it gradually learns which responses receive the most engagement and which responses are accepted and fixes it as fact.
ReplyDeleteone of the most funny (and least creepy) result is the AIs inability to draw clocks. When you ask AI to draw a lot of clocks showing a set time, apparently it shows you most (if not all) clocks showing 10 past 10. That is because the clock advert people decided that it is the most attractive looking time, and when AI scours millions of clock pictures and sees it all shows a certain time, it decides/learns; to be relevant all clocks must show that time regardless of your prompts.
I'm reading a book about how dogs experience the world and reinforcement learning is pretty much how we use their innate or instinctive behaviors to "train" them to behave how we want them to. It isn't really training so much as making it apparent to them that certain behaviors are in their interest. Of course, dogs have needs, like food and comfort and safety, that drive and reinforce their actions; presumably AIs don't. I wonder what it is that causes an AI to learn by experience? If it doesn't "need" anything, what is it's motivation to do any one thing over any other?
Delete