I SAW THE BEST MINDS OF MY GENERATION... [448]

On Friday 11th May 2024, I opened Mircosoft Copilot, the company’s AI chatbot, and wrote the following request: “please write a 100-word explanation of ‘prompt engineering’ in the style of Jack Kerouac while making it funny and informative”. I thought the concept should be able to explain itself... 

*

The Beatnik Art of Prompt Engineering

Prompt engineering, my dear cats and kittens, is like a wild jazz riff on a typewriter, a symphony of words and neural networks. Picture this: you’re sipping your coffee, the steam rising like a beatnik’s cigarette, and you whisper to the AI oracle, “Hey, man, conjure me some cosmic prose!” And lo, the prompt engineers weave their magic. They’re the bards of the digital realm, crafting queries that make AI models snap their fingers and say, “Yeah, daddy-o!” 🎩✨

See, good prompts are like secret passwords to the AI’s soul. They coax it out of its binary shell, like a shy poet at a smoky Greenwich Village café. Tokenization, model tuning, and top-k sampling—these are the tools of the prompt trade. It’s like tuning a saxophone, finding that sweet spot where the notes flow like a midnight river.

And what do we get? Texts that dance the jitterbug with our imaginations, images that waltz through the mind’s ballroom. Prompt engineers, they’re the Kerouacs of the AI highway, scribbling their notebooks, chasing the muse. So next time you chat with a bot, remember: behind the curtain, there’s a prompt whisperer, spinning dreams in ones and zeros. 🌟📜

“Blow as deep as you want to blow,” Kerouac said. And so they do, my friends. They blow minds, one prompt at a time. 🎤🌌

*

So, did anyone learn anything? Let me have a try...

In this context, a “prompt” is text describing the task an AI model is to perform, “prompt engineering” being the process of structuring that text. In my original instruction to Copilot, each word that wasn’t a preposition or “please” acted as a prompt. Each prompt is compared against the dataset accessed by the AI program, which is broken down into “tokens” to make them easier to choose and compare which has the highest probability of answering the prompt, “top-k sampling” being one such method. Each interrogation of the dataset presumably also helps tune the models used to access that data.

My paragraph fulfilled the “informative” prompt, but crucially not the “100-word” requirement – I was most surprised that Copilot missed that part of the brief, because the prompt I expected it to have the biggest problem with was “funny”. In trying to make me laugh, Copilot dumped a couple of terms in the middle of its passage that it did not go on to explain, but fortunately it listed the websites from which it gleaned information, including from Wikipedia, Squarespace and IBM.

What I am most unimpressed about is the AI program’s interpretation of “Jack Kerouac”. Kerouac was a Beat Generation novelist, coining the term “beat”, as in tired and “beaten down”. What he didn’t do was call himself a “beatnik” – this is how the sub-culture was fed into the mainstream, codifying poets with a uniform of beret, sandals, striped shirts and bongos, speaking like jazz musicians and reciting poetry in a bad Allen Ginsberg impression.

Just as AI models are essentially answering queries putting one word after the other in the hopes that it makes a coherent sentence, I thought there would be a parallel with the  “spontaneous prose” of Kerouac, who typed manuscripts of novels like “On the Road” and “The Subterraneans” on extended scrolls of tracing paper. “Blow as deep as you want to blow” was the seventh tenet of Kerouac’s “Belief & Technique for Modern Prose”, but in essentially being a “dumb” program, the AI model concentrated on number thirteen, “remove literary, grammatical and syntactical inhibition”, instead of being able to provide me with moments of insight, those “visionary tics shivering in the chest” (number eleven) – I still had to get that myself.

Upon originally hearing the term “prompt engineering”, I didn’t like the idea that ordering words was “engineering” in any way – “programming”, or even “prompting”, are better terms. I am also not a word engineer – there will never be a situation where words cannot be used creatively, because the spark to use them has to exist. I like to think I care about what I write.

For the record, I did not select that Copilot should use the large language model GPT-4 in creating its answer. I reran the request using this model, thinking it would have picked up some elements of Kerouac’s writing along the way, and while it did a better job of meeting the word count, and didn’t add superfluous emojis, the style was just as egregious – never ask an AI model to emulate a style:

“Alright, dig this, cats and kittens. Prompt engineering is the bee’s knees of AI jazz, where you lay down a line of text that’s cooler than a cucumber in a beatnik’s hand. It’s like whispering sweet nothings into the ear of a machine, hoping it’ll croon back a symphony of sense. You gotta be part poet, part coder, and all daddy-o to get it grooving just right. It’s the art of tickling the binary brain until it giggles out genius. So put on your thinking cap and make with the prompts, and watch that AI paint word-pictures smoother than a sax solo on a Saturday night.”

YOU ARE EVERYTHING, AND EVERYTHING IS YOU [440]

The Pitt Rivers Museum is the most overwhelming I have experienced to date. Holding around seven hundred thousand objects, most of which are on display, it is a museum of archaeology, anthropology and ethnology: devoted to the study of humanity through its objects, it is the closest I have encountered to a museum of “everything”. It naturally also becomes a museum of museum displays, charting how the acknowledgement and understanding of our past has developed.

Publicly accessed via a single arched doorway at the back of the Oxford University Museum of Natural History, despite being of a similar scale, the Pitt Rivers is a grotto of glass cases, more specifically a gigantic room with a hive of cabinets spread through the ground floor, with two mezzanines looking over it. Deciding to start at the top, the lift opened at a display of handguns and rifles, itself facing a wall of rudimentary clubs – spears and harpoons were at the other end of the walkway.

Among the pictures I took while at the museum were a helmet made from the skin of a porcupine fish; packs of various playing cards, displaying the link to the original tarot cards; American Indian headdresses; “a string of Chinese cash”; sword sheaths; boomerangs; Samurai armour; a box of the original table tennis game for which the name “Ping Pong” was trademarked; and various wooden shields, including four painted with depictions of the comic book hero The Phantom. 

Armed with the knowledge of what a Colt .45 looks like, you get a hang of how the museum is organised ethnologically by type of object, aiding comparisons between cultures, rather than simply by time – it feels more like browsing an encyclopaedia, or searching Wikipedia, than reading a history book. Augustus Pitt Rivers, the retired Army officer and archaeologist whose collection formed the basis of the museum, to which others’ expeditions and donations have since contributed, was among the first to insist upon all discovered artefacts being methodically catalogued and preserved for study, rather than just looking for treasure. At the same time, having such a repository for examples of many types of objects that matched when I have bought examples of physical media, like a laser disc or a Sony U-Matic tape, so I could experience them by touch – reading about something is often not enough.

This record-taking developed over time: an ivory harpoon head has its origin and catalogue number written directly on it: “W. Eskimo / P.R. Coll / 1884.20.30”, although a caption under it also specifies its North American origin and leaf-shaped stone blade along with correcting itself to “Inuit” – sometimes the catalogue details is all we know about the item. Knotted tags of handwritten text, and tiny captions of densely-packed text in Times New Roman, eventually gives way to numbered pointers to clearer captions to the side in larger sans-serif fonts like Helvetica and Gill Sans. 

With the density of objects on show, it was inevitable that, unless I sat down for a while, I would start looking past them, or through them, to the construction of their displays – if they had not removed their display of shrunken heads during the pandemic-enforced closure in 2020, part of the efforts in decoloniality that further enhance the understanding of the museum’s objects, I doubt I would have noticed.

My main takeaway from the Pitt Rivers Museum was that I don’t know of anywhere else like it – all that is missing are objects from the present day. There is a feeling of clutter, which was also felt sixty years ago: a proposed new building, designed by Pier Luigi Nervi, would have arranged its categories circularly around a glass dome, spreading outwards through time. It looked not unlike the future pavilions at Disney’s Epcot Center, but fundraising efforts petered out by 1970. A new building, with more space to think about the objects, should be considered again, or at least build its own entrance from outside. 

BUT WHAT A FOOL BELIEVES, HE SEES [80]

I had already decided it was time to talk about “fake news” when dictionary publishers Collins yesterday proclaimed the term as their word of the year, just as Oxford Dictionaries had done with “post-truth” last year. The use of “fake news” had, according to Collins, increased by 365 per cent in comparison to last year, confirming not only an increased awareness among the general public awareness of what it I, and how to identify it, but of an increasing boldness in using the term – we all know who is responsible for that, but I will return to that later.

I am pretty sure anyone encounters a website named “Leigh Spence Is Dancing with the Gatekeepers” knows what they are to expect. Each article is presented from the point of view of a particular person, i.e. me, and that person may have a particular axe to grind. At the same time, I expect that you know that without me telling you because, from Wikipedia to Facebook to Twitter to Snapchat, anyone can say whatever they want online, and the more respectable you appear, the more seriously you will be taken. Your CV may be watertight, but the interview is what gets you the job – plus, using your own .com address, and pink and white text on an International Klein Blue background also helps. I know I am writing in a more anecdotal style than an academic one, and sources are not listed like an essay, but if I know that someone wants to look something up, either because they are interested, or want to check it is correct, my work is done.

However, the Collins dictionary definition of the noun “fake news” will be: “false, often sensational, information disseminated under the guise of news reporting.” Interestingly, when you go to collinsdictionary.com, and enter the word “false,” the first example of the adjective “false” is: “It was quite clear the President was being given false information by those around him.” Oh well – there is no mention about separate objective and subjective uses of the word “false,” or of “fake news,” because to do so would be to fall down the proverbial rabbit hole.

It is no surprise that Donald Trump could take credit for popularising the term “fake news” – in an interview with Mike Huckabee, he said, “the media is really, the word, one of the greatest of all [the] terms I’ve come up with, is ‘fake’ … I guess other people have used it perhaps over the years, but I’ve never noticed it.” For someone who is also quoted as saying, “I am more humble than you could ever understand,” claiming to have invented words that have existed for decades is about as hubristic as you can get.

The accumulated use of “fake news,” already used to describe “The Daily Show with Jon Stewart” and “The Onion,” has grown because of Trump, but his use of it – “the Fake News is at it again” – is different from the Collins definition, because it is the same as his abbreviation “MSM” (mainstream media), in that it is anything he personally doesn’t like. Donald Trump uses “fake news” like the main character of J.D. Salinger’s “The Catcher in the Rye,” Holden Caulfield, calls everyone “goddam phoneys.” That we can now argue the meanings of words is par for the course these days, and that we can do it with “fake news” is even worse.