Tag: AI

Paragraphs about AI

All these thoughts I keep having about AI, I’m going to put them here…

The models are unpredictable, in fascinating and stimulating ways but, to state the obvious, they cannot be other than capitalist in nature. A socialist LLM, were one to exist, would have to have been trained on another world, in another context all together. 15 March

It’s already too late to take these tools away from their most passionate users – and it’ll soon be too late to take them away from workers, many of whom now depend on them. 13 March

It’s said that blocking AI will be counterproductive because if we do then only bad actors will progress and we’ll wind up only with bad AI, but this essentially deletes human agency all together. It must be possible for humans and human institutions to just say ‘no’. 13 March

A tricky aspect of arriving at an accommodation with AI is that quite a lot of its output will actually be a kind of hybrid – part human and part AI. Unpick that, AI police. 11 March

Good art is true. All of it. AI art can never be true. It can be plausible (useful, persuasive, stimulating…) but it can never be true. In this AI art is like bad art. They’re the same thing. They’re not true. This seems obvious to me. 11 March

Don’t refuse to use AI because of an ethical objection to one of its applications or to a particular, exploitive use or because you have a vague idea that it’s ‘evil’ or ‘stupid’. 8 March

It’s safe to assume that AI will improve. That gaps will disappear, errors and hallucinations diminish, plausiblity and usefulness increase. Don’t expect it to fail or weaken or ‘eat itself’. 8 March

With AI, discrimination will become a more valuable skill. Not magically being able to ‘detect’ AI work – for that will surely soon be impossible – but being confident in your judgement of all work, whether human or AI. 7 March

Copyright does a very simple thing: it provides a creator a temporary monopoly. Should we suspend this 300 year-old protection so that AI businesses can train their models cheaply? Should a nation voluntarily suspend copyright to boost the AI economy? No. 5 March

In criticising AI poetics will be more useful that hermeneutics. In fact, the profusion of increasingly-plausible AI work surely represents some kind of crisis for interpretation. Susan Sontag saw this coming. 2 March

Is AI going to be one of those tech innovations that actually reduces profit? Like the web and solar power – producing huge incomes for critical businesses but driving down profitability across whole industries? Seems plausible. 27 February 2025

Metre disorder

the compound word elevator-operator showing marks for the four weights of emphasis in poetry
The four weights of stress in poetry

I love poetry. I write poetry, I love rhyme and rhythm and structural play of all kinds. I write in rhyme. But I can’t see metre.

I know what metre is, I can hear it when it’s pointed out. I know it’s there. I just can’t see it or use it in any non-clumsy way. I’ve half convinced myself this is a condition or a syndrome, or maybe a disorder. Or the malignant trace of a buried trauma.

Metre is not present everywhere. It’s not in most ordinary writing, which observes a kind of anti-metre which, if it’s good, mixes rhythms and emphases to form a kind of counter-metre that flows and is readable. We can’t call it metre because it’s not regular, it’s the deliberate (or accidental) absence of metre. So it’s something else (of course, ask an actual poet, or a literature prof, and they’ll probably tell you this is rubbish, that metre is present in all deliberate writing).

Metre is a structuring device, a way to impose structure on the flow of language, on the unordered. So now I’m wondering if I’m missing structure elsewhere. That if I can’t see the repeating structure in poetry, maybe I can’t see it in other forms. Does this explain why I need an expert to point out the connections and the hidden figures in books and movies? Why I quite often miss the plot twist on first reading, or the recurring device in a movie or the huge, obvious repeating theme in a piece of music?

Is it also why I often feel like a mug in the presence of a politician or a persuasive communicator. Why it takes me a while to parse a new policy, locate the nub of the argument. I’m not an idiot but I’m missing the structure in some way. I’ve got half a grasp. I’m a bit slow.

This is also why I’m convinced that, for all of us, our main concern with regard to politics and to policy should be not to be mugs, to make a serious effort to spot the hidden structure, the concealed rhythm. In day-to-day politics this means not being drawn in by the superficial meaning of a new policy, making a conscious effort to go deeper and find the truth.

Anyway, I asked ChatGPT if it uses metre in its output:

Overall, my primary focus is on producing coherent and grammatically correct text that is appropriate for the given context or purpose, and any use of rhythm or meter would be secondary to this goal.

ChatGPT-3.5

Then I asked it to summarise our conversation, only in iambic pentameter:

Sure, here’s a summary in iambic pentameter:

As an AI language model, I write
With neutral style that’s clear and concise.
My focus is on grammar that’s just right,
And punctuation that will suffice.

While rhythm and meter are not my aim,
I can recognize patterns in the text,
And sometimes use them to create a frame
For output that is of high quality and next.

My primary goal is to produce text that’s fit
For context and purpose, without any fuss,
And so, I strive to follow each writ
Of standard language conventions, without any muss.

ChatGPT-3.5

And I think you’ll agree, it’s kind of amazing. So I asked it to make the stresses visible. It obviously doesn’t have the character-set needed to do proper stress markings so it’s just used capitals:

As AN AI LANguage MODEL, I write
WITH neuTRAL style THAT’S clear AND conCISE.
My FOCUS is ON gramMAR that’s JUST right,
AND PUNCTUation THAT will SUFFice.

While RHYthm and METER are NOT my AIM,
I can recOGnize PATterns in the TEXT,
And SOMEtimes USE them to CREate a FRAME
For OUTput THAT is OF high QUALity and NEXT.

My PRImary GOAL is TO proDUCE text THAT’s FIT
For CONtext AND purPOSE, WITHout ANY fuss,
And SO, I strive TO FOLlow EACH writ
Of STANdard LANguage CONvenTIONS, WITHout ANY muss.

ChatGPT-3.5

The first thing you’ll notice is that it’s not strictly iambic pentameter but that it’s pretty close. This seems to correspond with what we know about the AI models – they’re making a probabilistic stab at the answer so it will often be off in some way, sometimes drastically but often in a subtle, difficult-to-detect way. This is also how I write poems.

I make a probabilistic stab. Sometimes it’s a bit off.