Short-form video content is a huge passion of mine: my 2k-word review of Instagram Reels argues it’s the greatest artform of the 21st century. Soon, plzdontkillus will bring AI safety to TikTok — and as a high-screentime (read: naturally aristocratic) zoomer myself, I have some thoughts on the subject. What follows is a (very quickly) dashed-off list of stuff I think is underserved on short-form video apps, roughly sorted from most effective in terms of impact to most exciting in terms of aesthetics.
If I boiled this post down into two words it’d say: be xitterlike. The memetic ecosystem of X is second to none when it comes to AI. Whether from tech twitter or TPOT, ambient awareness of current and future issues with the technology — and the “factual waterline” — is higher on X than any other social media platform. (Note that this is a low bar).
Accordingly, some of the highest-impact work a creator can do is that of raising the factual waterline of the short-form video apps. Their demographic skews young and ignorant, such that even just getting attention onto basic facts makes a real difference. Certainly these can be smuggled in through memes, but it’s not strictly necessary — TikTok in particular has demand for “intellectual” content, which is currently (overwhelmingly) met with pseudointellectual twaddle; there’s a pretense of serious discourse, but it’s rarely more than superficially thoughtful. Dedicated informational and intellectual content that’s actually, y’know, good/true/correct, is criminally under-served and could be very successful.
This might look like:
- “you should know” infoslide-style activism
- commonly-held mythbusting
- targetted rebuttals / “dunks”
Each of these is easy to “clickbaitify” to catch attention: the first promises something the watcher doesn’t know, the second promises secret knowledge overturning conventional wisdom, and the last promises the vicarious thrill of tearing apart someone’s argument. The apps are a field day for a debunker especially because they’re just so rife with misinformation. An example of someone in this niche is Quick Thoughts:
There are two failure modes I usually notice with this sort of content. The first is being too “old media” style; much informational content comes from established organizations who have this formal, newscasting approach to short video. This is fine, but the zoomer brain doesn’t have the inherent trust in traditional institutions of earlier generations — more often than not, it’ll be a person talking in front of their phone that lowers their guard and feels more trustworthy. It has a human face.
The second failure mode is being too complex and involved. I’ll say it three times: viewers have short attention spans, viewers have short attention spans, and viewers have short attention spans. They’re also not on Reels or TikTok or YouTube shorts to do a homework assignment or keep track of things inside their head. You have to be short, digestible, and to-the-point. Save your detailed and multifaceted takedown for your blog; highly focused, highly targeted content is the way to go. Laser in on a single point. Debunk one myth. Tell one story.
The factual waterline can also be raised indirectly. A viewer might recognize Sam Altman, and they certainly know Elon Musk, but Demis Hassibis? Dario Amodei? Ilya Sutskever? Amanda Askell? Something so simple as making memes that incidentally group the model names, logos, lab names, CEO faces, and CEO names together associates them in the viewer’s mind and raises their literacy on the topic going forward.
This can also just be as simple as meme-ing CEO quotes:
Obviously this is not a particularly evil thing Sam Altman is saying, and there are much more worrying things to focus on, but this is a good example of literally just taking a quote from an AI CEO and communicating it to the viewer so that it keeps their attention. The popular resentment towards billionaires, technologists, and artificial intelligence means that this sort of thing plays very well right now. Tapping into it can be dangerous — water usage misinformation, for example, spreads through the same channels — but the fact remains that the public cannot keep the companies and CEOs accountable if they don’t know what the companies and CEOs are saying or doing, and getting people informed of real issues is going to get people invested in real issues.
And notably, common opinion on the labs is still largely unformed. We might know that Deepmind is staffed by sterile scientists, Anthropic’s culture is (questionably) effective altruist, and OpenAI is steered by a coldblooded conniving snake — but most people on the street couldn’t tell you the difference. Characterizing the labs to the wider public is worthwhile. This might look like:
- gossip-style presentation of court documents and internal memos
- intra- and inter-lab politics (currently there are more TikToks about OpenAI being a Jewish conspiracy linked to the hexagon on the planet Saturn than there are TikToks about the OpenAI attempted coup…)
- rivalry narratives; for example, covering recent talent transfers to Anthropic
One danger in drawing comparisons between labs is devolving into what might be termed “Anthropiganda.” There was a spate of this after they refused the Pentagon, which of course deserves celebration, but to a dedicated doomer they are still only the lesser of three evils. Whether Anthropic ought to be valorized for being the most safety-pilled of the frontier labs, or castigated for being a frontier lab in the first place, I leave to the reader to decide (short-form video, of course, has no room for nuance).
The natural next step is characterizing the models themselves. Rudimentary attempts have proven popular:
Memes about LLM failure modes and campfire stories of scheming behaviour are naturally important and obvious things to do; but what impresses me most about the success of this is the positive valence. It’s downright joyous. There’s a lot of cognitive dissonance among zoomers, because of course they’re anti-AI, but it’s just so darn useful … playing into this is less immediately useful from a safetyist perspective, but from an aesthetic one I’m very interested in hopeful stories, and those centering interesting interactions between humans and AI. Did you know there’s not a single TikTok about the cyborgists? And there’s exactly one about the Claude funeral? This is cyberpunk as hell! It deserves more eyes. And who knows — perhaps, too, this stuff is seeding positive hyperstitions.
In the same vein, I’m very excited about centaur art. Most artists these days, disappointingly, have whipped themselves into a reactionary hysteria about AI. But when someone does take advantage of the full possibilities afforded by the technology, and (very importantly) they have good taste and a vision, I’m seldom unimpressed:
I expect plzdontkillus will attract the creative geniuses they’re advertising for, and it’ll be a uniquely fruitful environment for weird collaborations and weird art from creatives open to highly experimental techno-humanist jams. And surely someone there will agree with me that Tung Tung Tung Sahur is historically significant…
Lastly, one step beyond centaur art is art from the AI itself. Regardless of your philosophical position on the subject, isn’t this neat?
If I were at plzdontkillus I’d probably blow half my stipend giving Claude Code a TikTok account just to see what would happen.
As addendum, you may have observed a notable absence:
(OK I couldn’t actually find it but it was a skit about a guy in an airplane who learns the pilot is going to assemble the landing gear while they’re in flight … that old AI alignment allegory … it’s mildly funny)
Why not talk about memes and skits that are directly about AI-safety?
For one, it’s the obvious thing. For another, I think people expect this kind of thing to be more effective than it actually is; it strikes me as closer to an ingroup cheer than effective evangelism. Of course, there’s nothing wrong with keeping spirits up! In fact, a community like this one could use a little more inward-facing art. With any luck, plzdontkillus will usher in the Rationalist Renaissance on TikTok. Trust.
Of course, I have many more thoughts and prospective plans, but I’ll keep them hostage contingent on acceptance. Can’t be leaking alpha on the open internet, after all…