Ill-Advised Musings On LLMs

I’ve got a cold, so I’ve been idly sitting around doing some slightly more thorough testing of my Emacs/mpv setup for asking LLMs what actor is on the screen of the movie I’m watching, and it’s led to me pondering just why some people are so (literally) incredibly enthusiastic about LLMs.

I mean, LLMs are fun. And useful for doing tedious programming things — for instance, the other day I asked ChatGPT to transform a li’l 150 line Javascript thing from Jquery to pure Javascript — and it did it flawlessly. (And also badly, engineering wise, since it did pointless stylistic changes on just about every line.) That would have been boring to do myself, so it’s nice that there’s a tool for that now.

But non-programming things? I just don’t understand the enthusiasm, because whenever I try to use an LLM for something, it’s never more than a toy. Somewhat useful toy, sure, but you can’t say in any way that it actually, like, works.

I wonder whether the enthusiasm is paradoxically based on how bad LLMs are in general. That is, when you chat with one of those things, it’ll give you the wrong answer a lot of the time, but then you say “but that’s not right”, and it’ll say “Good catch! You’re so smart! I’ve never seen anybody be that smart before; you must be a genius!” or variations thereof, depending on how high the company in question has dialled the knob that’s marked “Ass-Kissing” in the LLM console.

It’s just hard to be mad at a technology that’s consistently stupider than you are, and that always confirms your secret suspicion that you’re really, really smart yourself.

So I’m testing the same screenshot repeatedly with the same LLM (Gemini-2.5-Flash here) to try to see just what level of bullshit it’s giving me. This is Keanu Reeves from Even Cowgirls Get The Blues, and while he does look kinda untypical here, I think if you give somebody a cast list and ask who it is, I don’t think it’s that hard to pick him out as the most likely candidate.

But Gemini isn’t better than a throw of the dice here, even though I’ve tried in many ways to instruct it “if you don’t know, don’t guess”.

It even guesses correctly some of the time.

But let’s try some other movies…

It gets Dakota Johnson correct (from the cinematic masterpiece Madame Web).

Always.

This, from Dune Part Two, doesn’t seem right? Actually, I forget his name now… Oh, yeah, Dave Bautista.

No. So is it really just going through the list of cast members and picking one at random? What if I don’t tell the LLM what movie the screenshot is from…

Nope, makes no difference. It picked a guy from a different Dune movie, though?

It gets Timothée right — always, whether I tell it the movie or not.

Well, it’s not a totally bad guess — they aren’t totally dissimilar. (It’s Bill Pullman from Lost Highway, though.)

Yeah, if I tell it the movie, it gets it right.

Nnnnno.

So, once again, when I try to use an LLM for something, the result is somewhat useful — it’s better than nothing, but you can’t rely on it at all.

And people want to use these things to process social security applications and stuff.

Amazon apparently has a thing for letting you know who’s on screen (called “X-Ray”), and according to rumours, it’s based on hiring people in low-cost countries to tag every scene with the people on the screen. I thought identifying actors would surely be a solved problem by now, but I guess Amazon knows best.

Who is that actor on the screen? Emacs/LLM/Fun Redux

I’ve got a cold, so I futzed around a bit more with this…

To try to answer that eternal question that we all ask ourselves when we watch a move — who’s that, then? — I poked at ChatGPT to see whether it could tell me. And the answer was “kinda”, but it didn’t really want to, so the results were a bit hard to use in practice.

However, because Google dropped the “don’t be evil” motto (obviously), Gemini has no scruples whatsoever about identifying people. Which means that I can get it to output data in a more useful manner.

I asked it to return most famous movies, and also the IMDB ID. This means that I can display the actor headshot to see whether the answer is correct. I have to create an OSD overlay for mpv, which led me to further mpv fun:

Eurmh… Yeah, yeah, I should read the ImageMagick manual, but from bitter experience I know that it’s not super helpful on the “raw” formatting stuff.

OK, I’ll ask Google:

What? RBGA? Swap? That sounds suspiciously wrong, but I tried it anyway, and it was wrong.

That one was actually right! Except that it didn’t have -background transparent, so it didn’t really transpere (that’s a word).

There. I fixed it!

Oh, I love the helpful mpv error messages…

Anyway, after futzing around with this a bit more, I got it to work:

OK, what with all the back and forth between Gemini and IMDB, it takes several seconds to get the data. And…

That’s not a picture of Jane Russell. That’s Geoffrey Rush:

It’s kinda fascinating that Gemini gets the hard stuff right (identifying Jane Russell), but is bad at mapping that to imdb IDs. After testing some more, it seems to get the right ID about half of the time, and the rest of the time, it’s just a random person? Let’s try again.

Err… Let’s try again:

OK, Gemini isn’t very good at IMDB IDs. I’m really looking forward to all companies and gummints firing everybody and just relying on LLMs. That’s going to be so much fun…

Anyway, perhaps I should just ditch that LLM part and look it up in IMDB myself:

The problem is, of course, that there’s oodles of people with the same name in IMDB. I could filter for famousness, though. Hm…

Yup. But I don’t know whether this is better than the LLM’s guesses… People aren’t consistently named anyway (“Michael Fox” and “Michael J. Fox”, etc).

And showing both images would be annoying (and take even longer).

Oh well.

But looking it up in imdb works for Jane Russell, at least. So there you go.

And now I’m going to dedicate myself to feeling sorry for myself again. *cough* *cough*

Book Club 2025: Rent spel by Tove Jansson

Jansson’s books non-Moomin books were out of print for quite a while, but they were all released in new editions a while back, so I bought all the ones I hadn’t read yet, and I’ve slowly been making my way through them.

Most of them are very short (this one is just eighty pages long), and I guess you could say they’re kinda variable? Some of them seem very self-consciously “I’m writing for adults now, so there!”, and if I understand correctly, they weren’t well received in Sweden or Finland. And most of them were not widely translated. (I think that’s changed in the last decade or so.)

This one is absolutely wonderful. It’s not a novel per se, but it’s a collection of vignettes that are all about two women who live together, and they’re artists, and they spend a lot of time on a wind-blown little island outside of Finland. So it’s natural to read this as autobiographical, but I have really no idea whether it is or not: Of the two characters, the one that’s the analogue of Jansson’s partner is more of a viewpoint character than the one that’s Jansson’s analogue.

It’s a really sweet, funny and touching little book.

I’m not Swedish myself, but the language in the book seems more than a bit old-fashioned? That is, even if I’ve read Swedish all my life, there were more than a few words where I went “eh? eh?”. But perhaps that’s because Jansson is Swedish-Finnish (Swedish language but living in Finland), so the way of phrasing might be sorta archaic? I don’t know; just babbling.

Rent spel (1989) by Tove Jansson (buy new, buy used, 3.89 on Goodreads)

Book Club 2025: The New York Trilogy by Paul Auster

I read this in a Norwegian translation when I was… 20? I remember it clearly — it was during a very hot summer, and I spent the entire sweltering and bright night (living above the Arctic circle is fun) in bed reading this book. (And I seem to remember my cat being there at least some of the time — which is a bit odd, because he used to be outside all night in the summer… but he was pretty old by then, so perhaps he was just taking it easy.) Did I finish it in one night? All done by breakfast? I’m not sure, but I think so.

Wow, I still have that edition…

Anyway, this led to an obsession with Auster that lasted for a couple of decades, and I read all of his books up until, like 2005. Many of them were really good, but none were as magical as I remember these three novels were. So it’s with some trepidation that I’m finally re-reading this (and in English this time). Was I wrong back then? Is this book as great as I remember, or was it all due to that summer night and my cat?

No, I was right. It’s fantastic!

This time around I didn’t stay up all night to read it, but I did finish it over two evenings. (The cat died thirty years ago.) It’s such an exciting book — it’s got that early 80s post-modern freshness going on. (It was written between 1981 and 1984.) Everything has meaning and everything makes sense — it’s got some of that Pynchon-like paranoia, but in a very different way. There’s connections between everything, and there’s play with identity, and of course there’s a character (or two) with the same name as the author, and all that fun stuff. The trick is that Auster makes it all feel like it matters, so every new connection comes like a punch to the stomach.

You go *ouff* while reading it and kinda lose your breath for a second.

I’ve read most of Auster’s books, and I like a lot of them, but I guess you have to say that he peaked with his first novel. Well — collection of novels; there’s three short, interconnected novels here, but the first of these is the strongest, too. But all three are wonderful.

So that was fun book to read — not only because it’s as good as it is, but because it confirms what excellent taste I had as a twenty-year-old. (Yes, I know, I’m too modest.)

Oh, and it was nice to fondle a Penguin book again. As physical objects, they’re almost perfect books. The correct cream coloured paper, the correct floppiness of the paper, and it’s set in Bodoni, I think. Books that feel good hold and to read. (The cover design is horrendous, though.)

The New York Trilogy (1987) by Paul Auster (buy new, buy used, 3.86 on Goodreads)