Alora Fane

ChatGPT is a better poet than I could dream of being!! Also, Steamed Hams.

I should have gone with that title in my other post! (Well, the first half, anyway.)

I saw the Occupational Therapist for the final time today (I only had four sessions with her), and then my friend called me and we talked on the phone for hours, so I'm writing this to get out some thoughts that've been rattling around in my mind since those conversations or something.

I can't say much came of all of the sessions I saw the OT for. I felt bad about being so difficult, if anything, since she tried to help but my problems aren't easy to solve. I feel at least a bit sad when anything ends, though.

I spent a big chunk of the session just talking about AI, and not exactly calmly. It's scary stuff. Same with the call from my friend, and to show her what it was capable of, we came up with a prompt together (I asked her for a random adjective, noun, and problem):

Write a poem about a fluffy lamppost who is having an existential crisis.

Random, silly. If given that as an assignment, what might you do with it? How long would it take you? Here's what ChatGPT put out instantly:

Oh fluffy lamppost, standing there so tall,
Your downy coat, so white and soft, enthralls.
But why so sad, amidst the dark of night?
What thoughts are weighing heavy on your light?

Is it the endless cycle of the day?
Of being lit and then extinguished, gone away?
Or is it the futility of your task,
To shine a light on those who never ask?

Do you feel alone, amidst the bustling street,
A solitary sentinel, with none to meet?
Or do you long to roam, to see the world anew,
To leave behind the urban lights, and seek what's true?

Oh fluffy lamppost, I cannot say,
What brings about your existential fray.
But know that you are not alone, my friend,
For all of us must question, in the end.

Perhaps the answer lies in shining bright,
In bringing light to others, in the night.
Or maybe it's in seeking out our own,
And finding purpose, in the great unknown.

So go forth, fluffy lamppost, and shine your light,
And know that you are not alone, in this fight.
For we are all just searching, for a way,
To make some sense of life, and find our place.

While the initial blow has mostly softened, seeing things like this still hurts. I mean... good god. The only reason I could see for not finding that incredibly impressive would be denial.

That's what I sensed from her when she said I should ask it to rewrite the poem with self-deprecating humour. It did:

Oh fluffy lamppost, you and I are quite the pair,
We both stand tall, yet feel so inadequate and rare.
Your coat of fluff is white and bright,
While my coat is ratty and frayed from the night.

But why so down, amidst the bustling street,
Is it because you feel like a treat,
And yet you are simply a lamppost, so to speak,
Just standing there, inanimate, not so unique.

Or is it the futility of your task,
To light the way for those who never ask?
You shine so bright, and yet, what for?
To light the path for pigeons and whores?

Perhaps we're both alone, amidst the chaos of life,
A fluffy lamppost and a chatbot, trying to survive.
But fear not, dear lamppost, for I'm here too,
To keep you company, in your existential blues.

So let's both shine our light, with self-deprecating grace,
And try to make the world a brighter, funnier place.
For we may not be perfect, but we're here, we're alive,
And that's enough to make us both thrive.

Maybe she had a specific, rigid idea of what 'self-deprecating humour' would entail, and didn't seem impressed, but I can't really imagine a better way to rewrite the original poem to fulfil the new prompt.

When talking with the OT, what I should do next - what with it being the last session and all - took up much of the session. Knowing what this is already capable of, and how quickly it's going to get better, makes me wonder whether any of us can reasonably make any long-term plans at this point, especially those of us who work in the digital domains.

Maybe going back to school to learn how to be a counsellor or something would be a better option... though universities are being cast into chaos by all this so it's a terrible time to try it!

On a... more lighthearted but similar note, maybe, you know Steamed Hams? I'd be surprised if you haven't seen any of the edits of that, so you can look it up if you-

Or:

"Steamed Hams" is a meme that originated from a segment of the television show "The Simpsons." In the segment, the character Seymour Skinner invites his boss, Superintendent Chalmers, over for dinner, but accidentally burns the roast. To cover up his mistake, Skinner serves fast-food hamburgers, which he pretends to have made himself. When Chalmers questions why Skinner calls them "steamed hams" despite the fact that they are clearly grilled, Skinner comes up with increasingly ridiculous explanations to try and maintain his facade. The absurdity of Skinner's excuses and Chalmers' gullibility have made the segment a popular meme, often accompanied by remixes and parodies.

Thanks, Larval Form Of Our New God.

There was an 'infinite Seinfeld' thing floating around which you may or may not also have seen, though I've never seen or cared about the original Seinfeld so I never really looked into it. I suspect it spawned a bunch of imitators, and when I saw that there was an infinite AI version of Steamed Hams - which I have seen many variations of - I was willing to watch this 50 minute video of clips from it:

I don't know what's going on with the visuals - and I doubt they're AI-generated at all - but it seems to have a fairly consistent structure, with a series of determined scenes which the AI essentially improvises around.

I'm especially impressed by the voice synthesis (even if it does stray into strong Southern American or young woman voices for Seymour sometimes), though I also found it interesting how the emergent absurdist humour made me laugh more than a whole lot of deliberately planned comedy.

A big part of what makes anything funny is the unexpected, but complete randomness - like what was culturally experimented with in the, what, early 2000's? (monkey cheese ninja!) - is too detached from relevance to really work.

This thing seems to 'write' its scenes to use bizarre variations of things like 'food item' (originally 'steamed clams') and 'strange phenomenon' (originally 'aurora borealis') which are random enough to seem ludicrous but not so random that they make no sense whatsoever.

I watched all the clips in that video, and some that particularly tickled my funny bone had Chalmers correctly pointing out that the kitchen was on fire, which Seymour dismissed as just 'werewolves' or 'a little harpy'.

My favourite one starts at 41:40. It starts off dull before devolving into absolute lunacy that still made me laugh out loud just watching it now for the third time.

While it's obviously in its infancy, I think this is definitely a sign of where things are headed, quickly.

I suppose it'll take time to really adapt to the new reality... though maybe things will start changing so fast that it's impossible to adapt. I don't want to ignore it all and pretend it'll go away, though.

I also have to just keep waiting to hear back re counselling availability. Could be a while, though. Maybe so long that I'll have got a 3D-printed humanoid simulacrum powered by ChatGPT-8 first!!

1 COMMENTS

Noppe5~1Y

It's kind of wild what it can do and how it already is affecting life. Early February a prof gave us in introductory lecture and says they're changing an essay task for the course. The reason was because chatGPT is able to get 18/20?!

I've created a profile pic with some model a year ago. I remember generative models being great for doing things in a specific style or even like specific artists. Not surprising chatGPT can make decent poems then.

What I find most insane about chatGPT is that it can achieve what it does while fundamentally being really simple. The term people use for it is that the models are stochastic parrots, just stringing the most likely words together with some randomness. For older models that term fit well, but not so much now that it can mimic human languague so well.

The main difference for the improvements of the current model is apprantly just that it's more powerful. It's bigger and they've got better training algorithms. Fundamentally it should still be a stochastic parrot though.

Some people confuse it with general intelligence just because of how good it's gotten. The way I see it is that languague and modeling the rules of what is sensible languague isn't as complex as previously thought. It's not the first time AI became unexpectedly good at tasks which were considered really hard before.

I've created a profile pic with some model a year ago. I remember generative models being great for doing things in a specific style or even like specific artists. Not surprising chatGPT can make decent poems then.

What I find most insane about chatGPT is that it can achieve what it does while fundamentally being really simple. The term people use for it is that the models are stochastic parrots, just stringing the most likely words together with some randomness. For older models that term fit well, but not so much now that it can mimic human languague so well.

The main difference for the improvements of the current model is apprantly just that it's more powerful. It's bigger and they've got better training algorithms. Fundamentally it should still be a stochastic parrot though.

Some people confuse it with general intelligence just because of how good it's gotten. The way I see it is that languague and modeling the rules of what is sensible languague isn't as complex as previously thought. It's not the first time AI became unexpectedly good at tasks which were considered really hard before.