Boom times
This week: end of the tech boom, FTX and AI writers
Welcome to another free edition of The Terminal. If you’d like to become a paying subscriber and get access to more content, hit the button below.
I co-host the podcast Down Round, which you can subscribe to here (and follow us on Twitter.)
End of an era
I think it’s pretty clear by now the software-led tech boom of the 2010s is pretty much over. The combination of pandemic stimulus and tech-enabled remote work kept things ebbing along a few years longer than they perhaps would have otherwise, but the party’s over. There are the mass layoffs — helpfully tracked here — plunging stock prices and missed revenue guidance, but all of those can be blamed on a generally flagging economy. The rest of it is in the vibes.
You don’t have to look far for those vibes. Elon Musk ascending to the throne of Twitter atop a pile of middle manager skulls. Mark Zuckerberg pumping the spoils of a decade of social media dominance into brute forcing the existence of a parallel universe nobody wants to spend any time in. Legions of platform businesses collapsing on their own rickety scaffolding. The form and function of the smartphone growing stagnant, such that it is no longer really possible to discern the difference between iPhone models. The slushy cultural output of a legion of streaming services chasing what few potential subscribers remain. Elizabeth Holmes going to jail.
It’s unsurprising that it should be experienced as a vibe. If the last ten to fifteen years of technological development should be remembered in any way specifically, it is in the total convergence of the tech industry and mass culture, such that is no longer possible to imagine experiencing culture in a way not shaped by the hardware and software of some tech firm or another. Technological culture is how we experience the turning screw of increased interest rates and less free money flowing through the economy. Less startups are founded, less platforms thrive, less founders are elevated to the status of cultural icons.
For much the same reason, it’s hard to imagine what comes next, or what the inevitable next boom will be driven by. It’s hard to imagine, for example, Twitter actually dying regardless of how Musk chooses to run it, or a platform like Instagram withering away entirely despite how poorly Meta updates it for the tastes of a new generation. It might be short-sighted of me to say so, but it feels like we’re locked into the platform economy we live in thanks to the sheer accumulation of users and data we amassed over the past decade of crucial early tech development.
All of this, more succinctly:
Magic beans
One of the best pieces of writing on the collapse of FTX was Matt Levine’s Bloomberg newsletter about the balance sheet the imperilled exchange was shopping around to potential investors and rescuers after its shoddy finances were laid bare by a frenzied customer bank run. In it, Levine focuses on the fact most of FTX’s ‘assets’ were native crypto tokens it had invented from thin air and had a barely theoretical value which would absolutely not be realised if they were sold on the open market:
Last week I was shocked that one of the main assets of FTX — one of the main assets it relied on to be able to pay out customer balances — was a token it had just made up. But I was wrong! It was two tokens that it had just made up! FTX’s two largest asset balances, “before this week,” were $5.9 billion of FTT ($553 million at post-crash prices last Thursday) and $5.4 billion of SRM ($2.2 billion post-crash). Something like two-thirds of the money that FTX owed to customers was backed by its own tokens that it had made up […] It is striking that the balance sheet that FTX circulated to potential rescuers consisted mostly of stuff it made up. Its balance sheet consisted mostly of stuff it made up! Stuff it made up! You can’t do that! That’s not how balance sheets work! That’s not how anything works!
As the contagion from the FTX collapse continues to spread, imperilling various crypto exchanges, tokens, protocols and projects, one of the interesting observances is that the impact on the broader economy seems to be basically. There’s no doubt many retail investors and casual gamblers who bought into the company’s aggressive marketing have personally been torched, but the damage in the aggregate, while potentially extensive within the crypto shadow economy, seems basically constrained within in. The general cautiousness of regulators and institutional money to jump in with both feet probably contributed in this regard. (Maybe for the best that SBF’s extensive lobbying efforts weren’t fully successful.)
As this tweet puts it, it pretty well establishes crypto as largely what it always seemed to be: a “casino economy” which is “unrelated to actual financial functions”. I don’t think there are many actual crypto faithful who would disagree that is the current state of play, even if they believe it will serve a richer function somewhere down the line.
There’s also something darkly funny about the general response from within the crypto world that this merely proves that centralised institutions and exhanges are suspect, and that the only way forward is to return to the yeoman fantasy of everyone custodying their own assets — such as by putting their bitcoin on a hardware wallet, for example. I think the blackpill here is that you’ll never convince a huge portion of the population engaging with the monetary system to have absolute responsibility for their own keys and money.
There’s something uncanny about the FTX scenario in the end, which places it in a different realm to your usual fraud or Ponzi scheme. On the one hand it was a vast, sloshing pit of customer deposits, poorly managed and incompetently gambled by a bunch of speed addicts, which ended up as little more than a vault filled with illiquid company scrip, which at least rhymes with other luminary examples from the history of financial fraud. But maybe it’s the religious intensity of the crypto and blockchain world that gives it a different character. It reminds me of the newsletter I wrote about NFTs almost two years ago — evidence of our “epoch of high weirdness” and a “shadow world unmoored from reality and the rational laws we like to believe govern us”.
Vaguely related bonus – here’s a tweet I cosign as being probably at least somewhat true:
Poetic interlude
Robot scribes
I’ve spent the past couple of weeks playing around with Lex, the AI writing assistant developed by tech writing collective Every. Describing itself as a “thought partner who’s always ready to help while you’re writing”, it’s a pretty simple application in execution. It takes OpenAI’s GPT-3 and integrates it into a barebones browser-based word processor, allowing you to prompt the AI to imagine the next few sentences by typing ‘+++’.
The experience is pretty spotty right now. (“Sometimes it outputs rubbish,” the app’s documentation admits.) Rarely in my experimentation did it produce text I could slot seamlessly into whatever it was I was writing, and it frequently produced copy that was either stylistically jarring or wandered off on useless informational tangents. But, unlike the legions of terrible AI copywriting startups which constantly appear in my Instagram ads, Lex is not selling itself as something intended to produce final copy, or to automate the writing process entirely. Instead, it is leaning on a less disruptive pitch as a friendly assistant, efficiently supplying you with ideas which you might then reformulate in your own words or do further research into. It sells itself as a productivity tool rather than automation.
Here’s how Lex suggested I continue this section following those two paragraphs:
In many ways, it feels like a kind of ‘content prompt’, with all the attendant benefits and drawbacks. On the plus side, it can help you come up with ideas when you’re staring at a blank page; on the downside, it can often produce ideas which are either too general or too specific, and which lack the nuance and context that only a human writer can provide.
Still, as a tool for writers, it has potential to be genuinely useful, particularly when used in conjunction with other writing tools such as Scrivener or Ulysses. (It also integrates with Draft, though I have yet to try this.) And as the technology improves, it’s not hard to imagine a future in which AI-assisted writing becomes the norm, rather than the exception.
First paragraph is true in a generic sense for any AI-assisted writing. Second paragraph contains a glaring factual error, but if you’re counting on the synthetic output of GPT-3 for reliably correct information, you’re not going to get very far. As is always the case with software built on large language models, Lex is great at identifying and replicating patterns from its vast corpus of learning data, but not so great at establishing context or actually mapping ideas in meaningful ways.
That said, in Lex — alongside other AI writing assistants like Sudowrite, which is marketed to writers of fiction — you can observe some pretty interesting shifts in the way writing works as a productive activity. It unbundles certain skills from the fundamental act of writing. Using Lex for some basic, uncreative writing work I was doing felt more like acting as a curator than a writer, extracting structures and passages from latent space and then editing the output for errors, structure, style and readability. The biggest barrier as a productivity tool in this regard was that I felt compelled to fact check even the most anodyne output.
Of course, there are certain core aspects of being a writer that don’t seem ready to be disrupted by language models. Lex won’t offer you a truly original idea, unless it manages to stumble on one through accidental synthesis of text strings from across it training data. Anything it spits out will by definition be a reformulation of some idea (or collection of ideas) articulated by someone else. Maybe GPT-4 will find a way to shake real insight out of the machine. But I doubt it.
Elsewhere
Are we living under technofeudalism? Malcolm Harris says that’s giving too much credit:
There is no universal architecture between nature and God — and that’s definitely not what Facebook is. Facebook is not a public utility either. Facebook is a once-ubiquitous entertainment business financed by advertising, much like the television show Friends. At a more basic level, Facebook is servers full of degrading code and a bunch of workers getting yelled at by their bosses. If Mark Zuckerberg is a wizard, he’s the of-Oz variety, as he constantly reminds us by tripping over his curtain like a dork.
Interesting feature at The Verge on BuzzFeed’s struggle to remain relevant. There’s a broader point here in the leading lights of the mid-2010s digital media ecosystem attempts to keep the lights on in the post-Facebook era.
… There used to be hundreds of small kingdoms and principalities in Europe, each ruled by its own set of inbred weirdos. There was occasional cross-pollination through marriage, but each place nonetheless retained a distinct cultural identity. Eventually these kingdoms unified, creating the Europe we know today: a few dozen countries where people’s regional pride (and conversely, their regional animus) is forced to exist within a broader national identity … Today’s internet functions in a similar way.
The internet is over, writes Sam Kriss. “You know, secretly, even if you’re pretending not to, that this thing is nearing exhaustion. There is simply nothing there online. All language has become rote, a halfarsed performance: even the outraged mobs are screaming on autopilot. Even genuine crises can’t interrupt the tedium of it all, the bad jokes and predictable thinkpieces, spat-out enzymes to digest the world.”
On Caroline Ellison, the CEO of FTX-tied hedge fund Alameda Research, one of the stranger characters to emerge from the collapse. Forbes has a story reporting that she has developed a following among esoteric right-wing types online thanks to her (now deleted) Tumblr, which went on her weird trad-slash-rationalist cyberpunk philosophy.
The Infinite Conversation, a webpage hosting a bot which spits out “an AI generated, never-ending discussion between Werner Herzog and Slavoj Žižek”.
A post summarising major critiques of modern computing, both in paradigm and execution.
On Andor, which is not only a great Star Wars show but one of the best shows on TV right now. Not sure how Disney let this one through the keeper.
Kevin Xu at Interconnected has translated the final speech of former ByteDance CEO Zhang Yiming before his resignation.
Flatpack ‘fast furniture’ is clogging landfill in the United States.
Having just started a job at start up - I can tell you two things. Your first article depressed me, but also it feels very true. Second, there is still quite a lot of VC money sloshing around...