Prompt engineering

I have a bad habit, one I’ve been trying to break. (No luck so far.) When I interview people, I get too in my head about sounding awkward, so I overelaborate. I prompt my subject: Here’s the question, now here’s a menu of possible answers. Is it more ABC, or is it XYZ?

Consider this example from the previous edition of this newsletter, in which I interviewed former esports pro Anthony “vanity” Malaspina. I asked him (regarding his latest career ambitions): How did you get into finance and cybersecurity? Are those things you’re passionate about? Or is it just kind of like, this is what's going to put food on the table?¹

In the published interview, I whittle this down to just one sentence because, let’s face it, the actual question could have been just one sentence. The additional promptings are me sputtering — and worse, giving the subject an easy off-ramp. I’m preempting an unexpected answer with a binary of my own, setting up the respondent to choose one or the other option (in lieu of him taking the proverbial ball and running in some third, unforeseen direction).

In the grand scheme of things, this is a minor, personal quirk. But amid numerous discourses regarding uses and abuses of artificial intelligence — its discrediting of the perfectly honorable em-dash; its use by students to cheat on exams, essays and job interviews; its hallucinations regarding books that don’t exist — I noticed that there were some parallels between the weaknesses in my craft (the stuff I’d been laboring to excise) and the supposed workflows enabled by AI in creative fields.

I enjoy writing. (That feels too simple.)

I like the craft of putting together a sentence. (I already used “craft” in the previous paragraph; also “craft” in this context has a woody, dwarven vibe that doesn’t feel deserved.)

I like the work of putting together a sentence. (Not doing a very good job of it, am I? This somehow feels dumber than the simple version.)

I enjoy writing! The above paragraphs, regrettably, reflect my process. There’s a vision, hazy and unclear, of how the words ought to be arranged. I write and rewrite. A sentence reads well one moment, then stumbles over itself in the next. I fiddle with the knobs some more. The real thing is cobbled together, slowly, from false starts and dead ends. Meaning is excavated in the process. It works half of the time.

I could prompt an LLM to quickly generate an equally legible version² of this same newsletter³ , settling on a curated collection of sentences, read and validated by me as correct-sounding. But I think that workflow shares the same flaws I can see in my too-eager questions. The AI prompting me back (do you prefer generated option A or generated option B?) makes it less likely that I might stumble upon the odd “option 𐌔” — the option I would have chosen independently, if not prompted, or, more likely, the option I might have discovered through the work of sludging through a sentence, over and over again. I value this third option much more than the prompted responses.

Another example. Years ago, while reporting out a story somewhat outside of my normal beat⁴ , I spoke with visual effects workers: the people who help put together the vast majority of contemporary big-budget movies. The story was broadly a negative one, centering on the increasingly poor conditions VFX workers toil under, but there was one movie that I can recall an interview subject speaking fondly of. I won’t name it here, but the artist described it as a breezy assignment. The director had a very specific vision, one he articulated with ease and clarity, which made the production of the computer-generated graphics easy.

By contrast, the vast majority of other productions I heard about were disasters. A director would gesture vaguely at the visuals they wanted, artists would deliver some version of that, the director would get upset because it didn’t match what they were seeing in their head, the artists would start over, and on and on like that until the movie was finished. In one particularly egregious case, production (and rendering) time ballooned because the director couldn’t decide what size he wanted his characters to be. Do you like many of the big-budget movies you see these days? Does it seem like the process I’ve described above is good for art? Do you want to see that replicated in the future, while also putting a bunch of artists out of work?

AI offers an opportunity to cut the labor out of the equation without changing the underlying dynamic — to eventually deliver satisfying-enough sentences and visuals and songs. But I am doubtful of AI’s capacity to deliver on vision, and more than worried about the likelihood that it wipes out vision categorically.

Now and again, I like to respond to reader questions in my newsletter. (See: here and here.) If you’ve got questions for me — about video games, esports, criticism, etc. — drop me a line! My contact info is at the bottom of this newsletter. I’d love to hear from you.

Also: Some thoughts on post-sale Polygon

“Things are bad in games media” has unfortunately been a recurring theme of this newsletter — which I launched on the occasion of The Washington Post shutting down its video game vertical and laying me off. (I’m fine now; I was absorbed into a different team.) I just never expected to have to write something like this about Polygon.

If you’re not familiar: Last month, Vox Media sold Polygon to Valnet (an entity characterized by Kotaku, which broke the news, as a “click-farm powerhouse”). Many, many employees were laid off, with some writing duties farmed out to freelancers. The Wrap, which published an extensive exposé on Valnet and its owner, characterized the company’s whole shtick like so: “Employees are replaced by contractors, compensation plummets and writers who complain land on a blacklist that blocks them from working for Valnet sites altogether.” Valnet and its CEO are suing The Wrap for defamation in response. (Not the sort of behavior you might hope for from a steward of websites nominally engaged in publishing news.)

Polygon still exists, and there are some folks who remain employed there who I trust to do good work despite the circumstances, but the change in ownership can already be felt. The website, just recently redesigned, has been spackled over with video and banner ads. There’s been an uptick in conspicuously amateurish writing under unrecognizable bylines. (Sorry!) Oh, and Polygon’s Twitter account is live again. (“At the end of last year, Polygon exited Twitter. … We didn't need it. It was not a reliable source of traffic by any means, or engagement. It was just people being really upset, and visibly so, on the regular,” one of the site’s executive editors, Matt Patches, told me last year. Predictably, Polygon’s first tweet under new ownership garnered a bunch of replies about how it was good that games journalists had lost their jobs.)

In 2023, writing about a different click-farm powerhouse, GAMURS Group, I posed the question: What’s the point of click-farming?⁵

This, in fact, seems to be Valnet’s exact editorial strategy!

There were many different “eras” of Polygon⁶ , but each had a point beyond just churn. There are, after all, already dozens (hundreds?) of websites that clumsily paraphrase tweets and press releases. So who is served by throwing Polygon into the industrial vat of internet slop? (Perhaps a more apt phrasing: Who is enriched by it? No prize for guessing right.)

Last year, I cited a passage from a Substack essay by the journalist Matt Pearce about how journalism fits into the broader content ecosystem. Pearce wrote:

The last paragraph zooms out to social media, but there are less-than-scrupulous actors downstream of those platforms who, in light of the trends articulated by Pearce, view the production of good and original work as a wasted effort. We’re getting fucked by the platforms, so why bother? If a great article takes three days to write, that’s three days of missed ad revenue. You could publish three dozen ad-bearing pages in that time with minimal effort. Get the bag while there’s still a bag to get.

I think that’s shortsighted and a tragedy — particularly as it comes at the expense of a roster of journalists who did all sorts of weird, hairy, compelling, difficult work (not to mention the folks who did the exceedingly normal grunt work, but at a higher standard than is usual for an entertainment site). There aren’t enough life boats for all of them.

Thanks for reading ReaderGrev! Consider sharing it with a friend, on Discord, Twitter, LinkedIn, or even a subreddit where folks might appreciate it. Word of mouth helps this newsletter grow!

If you have a tip, I can be reached on Bluesky or via email at mikhail (at) readergrev (dot) com.

As a fun bonus: I was listening to Wolf Parade’s At Mount Zoomer on repeat while writing this newsletter. Here’s the album closer, Kissing The Beehive, for your listening pleasure.

1. This isn’t a perfect example, because the question I settled on in my edit was the second one, which was ultimately the more interesting one. But I didn’t know that when I was asking the question — I just sort of stumbled into it. I think I would’ve gotten there eventually, though, after the question I meant to ask: How did you get into …?

2. Easy dunk here: More legible! Getting that one out of the way before someone else tries it.

3. If I’m being honest, the “An AI Wrote This Article. You Won’t Believe What It Said.” genre of post is really losing its shine. You won’t catch me pulling that kind of stunt.

4. More on this in a future newsletter.

5. There are parts of this piece I sort of don’t care for anymore (iykyk) but it doesn’t terribly matter.

6. I’ve cherished Polygon across all of its weird and sometimes incoherent eras. The McElroy era made me a fan. But even before that, as a media nerd, Polygon stood out as the only place with the resources to publish something like this interactive on the work of the artist Roger Dean. (The live iteration of the page has broken a lot of the delightful scrolly-telling elements, but the Wayback Machine version, linked above, should get the idea across. Seriously, give it a scroll! It is the closest thing to Snow Fall I can think of in games journalism.) Around 2014, I applied for an internship at Polygon. (I beefed the interview, badly.) A few years later, I lucked into one of my first bylines ever thanks to the site’s former EIC Chris Plante. Eventually I joined another publication and — in a span of three years that included some of the biggest video game news stories of the decade: Covid-19, the launch of the latest console generation, Cyberpunk 2077’s troubled release, Activision sued by state and federal regulators, etc. — competed with Polygon’s crack editorial team.