What it was like to live in an AI safety influencer house for one month

A review of PlzDontKillUs

Many have asked for my experience of PlzDontKillUs, an AI safety-themed video-making bootcamp set in Berkeley conceived of by Aella and Ronny Fernandez. Here you go!

WHAT WAS IT?
—————————————
50+ influencery people living at a small campus in Berkeley making one video per day for one month. If any “fellow” failed to make their daily video, that fellow would be asked to leave. AI safety-related videos were lightly encouraged, but not enforced, under the theory that enforcement would kill our creativity.

TELL ME ABOUT THE VIBES
—————————————
The vibes were pretty peak. Influencers are people whose job it is to be lively and entertaining on camera, and – at least for this group – that translated to off-camera as well. Nearly every night was a party with people joking or talking philosophy around firepits, often until, like, 5am. It felt like college.

Some highlights
• Cream Brothers, in which we created a whipped cream cult. You became initiated by having whipped cream squirted into your open mouth.
• Me and friends moving in slow motion for about an hour straight.
• A night where we lay with our heads in a circle listening to a playlist dedicated to the flourishing of life on Earth and beyond.

The common denominator amongst fellows was extreme openness. The organizers screened for this characteristic with the application question “If you had to have sex with a cow would you rather it be dead or alive? Why?” and it seems to have worked.

Most fellows were under 30. At 36, I believe I was the second oldest, but the zoomers never made me feel like unc.

Besides these commonalities, people were all over the map. Some leaned right-wing, others were literal communists. Some came in with <100 followers and zero videomaking experience, others had 1,000,000+ followers and made videos for a living (including the top influencer in Latvia, who taught me how to make TikTok ADHD-style edits). There were edutainers, lifestyle influencers, hot-girls-in-tech, skit-makers, musicians (including three people trained in opera for some reason), and one 19-year-old Chilean who I think might become the Greta Thunberg of AI safety (look up Trinidad Matta on Instagram). Everyone was weird and vibrant in their own way.

With such a pluralistic mix, you might expect there would be a lot of drama. But someone asked me, “If they were a Hogwarts house who would they be?” and my answer is Hufflepuff. The vibe was kind, cozy, and inclusive. Wildly, roughly everyone got along.

I’m told this was quite the vibe-shift for the campus we were staying on. Lighthaven is the hub of the Berkeley rationalist scene. So rationalists would often show up to our parties and say things like, “Wow, I’ve never seen such good dancing here.”

WHAT WAS THE STRUCTURE?
—————————————
Besides the dictate of 1 video/day, there was close to zero mandatory structure. There were lots of optional talks from AI safety experts and famous influencers (eg CJ the X, Grimes). And also lots of fellow-run events like “Cyberdeck Building Workshop,” and “KPop Demon Hunters Singalong.”

In theory, there was “mentorship.” However, what this actually looked like was this: a bunch of famous mediamakers were invited. Most of them were Youtube essayists, like JREG and Solar Sands. They lived on campus with the rest of us for varying periods of time. There was only so much mentorship they could provide, since they were masters of a completely different artform than the shortform videos we were making. The formal mentorship that did happen occurred through “office hours” when like 20 of us would bombard them with questions. A subset of us would get their videos critiqued.

This was maybe my main complaint with the program, which was that there wasn’t anyone tracking whether you were improving. That was left to the fellows themselves.

Although it’s hard to complain much, because the program gave us a free room for the month with free lunch and dinner every day and a stipend of $2000. (!!!)

HOW WAS IT FOR YOU?
—————————————
Mediocre, but this wasn’t the program’s fault, it was mine.

First of all, my intention was off. When I heard about this experiment to get a bunch of random influencers to make AI safety propaganda at Lighthaven, my thought was, “That’s too fascinating for me to miss.” Coming into it, I didn’t have an intention of becoming a video influencer, and I still don’t.

The video-making ecosystem feels like it’s made up of a bunch of charismatic people shouting into the void, or, at best, into a “following.” If you’re making shortform, the game is mostly about hacking strangers’ attention so they keep watching. This is not why I’m on the internet. I’m on the internet to get into deep discussion threads with strange and galaxy-brained people, not to “Farm Views” (the title of one workshop led by a fellow who I nonetheless loved).

My intention for being there was not well-formed. In contrast, there was a guy who came into the program with the goal, “Get a job making AI safety videos,” and he got it.

I also struggled with the social environment a bit, mostly BECAUSE it was so rich. I’m a secret introvert and my college-style partying days are mostly behind me. I’ve also lately been feeling like I have way too many connections in my life; I was nervous to add more. All this amounted to fellows saying to me, “I never see you around,” and that was because I was either isolating or spending time off-campus. By the end, I had a tragic feeling around this. There were a lot of special people who I would have liked to connect with, and I didn’t.

This especially stung because I watched the younger folks there have a life-changing social experience. I felt left out of that, and the fault was 100% my own.

UPDATES AND TAKEAWAYS
—————————————
In random order:
• I got a lot better at showing up authentically on camera.
• It felt wrong to be tailoring the message “Beware of hostile algorithms!” for platforms that run on hostile algorithms.
• Shortform video is probably not my medium. Surprisingly, posting 1 video per day increased my Twitter following by approximately 0%, despite lots of people telling me they loved my videos. Upon reflection, this isn’t that surprising. I had no interest in algorithm-hacking or making the sorts of content that normally goes viral. I think I could become a pretty good Youtube essayist, but I’m told that (a) this is a dying form and (b) it takes so much more production time than writing. And most of the sorts of people who like my stuff are not people who watch a lot of video.
• I did finally discover a treasure trove of shortform content that I liked. Check out the hashtags nichecore, corecore, and targetaudience on Insta or Tiktok – people be making magic spells!
• Unlike most people at the bootcamp, I became *more* optimistic that AI won’t kill us all. A lot of the fellows there had very little previous contact with the concept of existential risk. I, in contrast, did existential risk movement-building for years, which eventually contributed to deep depression. So I was already coming in with a high “p(doom).” However, while we were there the Hugging Face incident happened, in which an OpenAI model committed international cybercrime. This might sound like a reason to be pessimistic, but actually it’s catalyzed a bunch of AI safety skeptics to start paying attention.
- I expected to there to be a lot more collaboration than there was. Collaboration, however, was somewhat disincentivized because if you and another fellow made a video, only one fellow could claim it as their official submission for the day. This made me sad.
• I asked Grimes how to make good cause-based art. She suggested that good art needs to feel like sex. I’ll add to this that I only want to make art that feels like *good* sex. The culture of shortform video, with its emphasis on hacking your brain, feels like bad sex, where some person at a party wearing too much makeup is feeding you drugs and hypnotizing you into sleeping with them. We need to believe that sex can be holy. And it can be.