@michaeltrazzi The 2.4M views mentioned above mean that we're on track for the best case scenario (of 500k views / week)

Selected videos for tha period:
- When AI take all the jobs you cannot retrain (300k views on YT + TT)
- Ex Google Ceo Says AI Will Disrupt the world order (80k views)
- AIS will be writing essentially all of the code in 12 months (120k views)
- Living through the singularity feels different from reasoning through it (20k views)
- When Machines do all the work civilization ends (13k views)

@MichaelCohen Thanks Michael! Means a lot

@Jesse-Richardson Thanks Jesse!

@michaelandregg Thanks Michael!

@MarcusAbramovitch Thanks for the questions. Let me address both points:

On the work involved: I spend 5-6 hours a day going through multiple podcasts to find the very best clips, where most of them don't end up being posted. There's also editing / uploading work (2-3 hours) on top of that that is hard to see (adapting things from vertical vs. horizontal by scaling / position / changing backgrounds, iterating on different captions, fixing the audio, fixing subtitles, upscaling, uploading to different platforms, checking on different devices). It definitely gets to a full day of work, especially as I do more clips.

6 figure (annualized) salary: I've spent some time thinking about how much money I would ask to pay for my time on this grant. One important word above is "work full-time on this project productively". I did consider other amounts that would mean basically only paying bills and nothing else, but I don't think that'd have been sustainable nor helpful to make this project go well.

To give more context, last year I made the mistake of under-budgeting on my salary for the SB-1047 documentary (see post-mortem here), which meant that I basically paid myself for only 2 of the ~8-9 months I spent on this. One lesson I learned from this is that compensating yourself for your time is not just a cherry on top after you have everything else figured out, but something necessary to work on something productively for extended periods of time.

> And assuming you believe in AI safety sufficiently, if you only get $12.4k (current amount as of this comment) in funding, are you going to just quit posting in 6 weeks and a day

After 6 weeks of full-time work, I'd evaluate options to maximize the project's continued impact: transitioning to part-time while fundraising, mentoring someone to continue, or documenting my process for others to pick it up.

I appreciate the clarifying questions, let me know if you need anything else.

@MarcusAbramovitch Yeah, I also like the idea, but $100-$300 per tiktok clip seems weirdly expensive. If you drop this due to insufficient funding I hope someone picks up the idea as a hobby

Yeah I basically have a similar qualm to Marcus, curious to see if this reply has assuaged his concerns as right now I'm fairly on the fence about contributing.

@cian You're right that the higher end of your range sounds high framed this way. The high-reach clips (>100k views) are great ROI, but without averaging things out and looking at heavy tails (most of the value comes from the small amount of clips that take off, and these continue to pay dividends as you post many of them), the ones that don't take off feel pricey. If that helps: TikTok's algorithm favors posting 1-3x daily max, so I actually cut 3-5 clips and only post the best ones, so you're not just paying for what gets posted.

@Jesse-Richardson Thanks for still engaging while still being on the fence. Curious what your main concerns are and if there's any more questions I can answer. Is it the compensation level, the time commitment, or something else? Also, happy to explore different structures if that helps, though not sure if I can accommodate people individually directly on Manifund, since I'd still need to be fair with people who donated here.

@ I want to quickly note that it's a bit unfair for me to specifically only call you out on this or rather, that this is a thing I find with many AI safety projects. It just came up high on Manifund when I logged on for other reasons and saw donations from people I respect.

That said, no, this doesn't exactly assuage my concerns/make me want to donate here. I don't think this work really takes that much time but more importantly, I just don't think it passes the cost-effectiveness bar that I hold my donations and my regrants to given the other opportunities I know of that are out there.

I don't think people doing "non-profit" work should have to make some kind of bare minimum amount to barely cover expenses but I do think that $100k/year to repost short clips on tiktok shows that spending habits in EA have gotten a bit out of control and we should do a bit of a sanity check. I also don't think that the correct metric here is that it's reasonable to pay the equivalent salary of an ML engineer in France. I think people usually/often inflate their earning potential but more importantly, I just don't think it's reasonable for non-profit work to get paid your maximum earning potential.

I think there are two relevant cost-effectiveness factors to consider in grantmaking: raw impact per dollar, and also the reasonable cost to reproduce. I feel uncertain on the former here (and I'm going to be doing a project to address this), but on the latter, I feel pretty confident that I could get 4-5x this impact by having low-cost EAs in 2nd/3rd world countries or college students doing this work.

Based on what you wrote, I think you might feel undercompensated for previous work (on the SB 1047 documentary) and thus want to recoup some of that on this work but I don't think that is what philanthropic funders should do here. Perhaps I'm somewhat open to retroactive funding for the documentary if I/others investigated but I don't think it's good to

quick notes:

• I love that Michael is willing to be so transparent and responsive here. The quality of comments here are great; thanks everyone for weighing in!

• I think highly of all of Marcus, Jesse, and Michael -- and even so, I think it's reasonable for them (and others) to have strongly different takes on whether a particular project is worth funding/doing.

  • I think sometimes EA people fall into a trap of thinking there's one "right answer" about cost effectiveness or theory of change, and we need to argue it out to get to ground truth -- and while there's some value in this, there's also a lot of value in enabling lots of experiments to be executed by the people who are most excited by those. Manifund is built around the latter philosophy: "let a hundred projects bloom"!

• As for myself, I funded this for a small amount, to signal support for Michael based on his past track record. I think the SB1047 documentary was well done (though, I was also a funder of that, so might be biased). I think the viewership numbers here are impressive, though I'm not super calibrated on what passes for amazing media presence.

  • Contrary to Marcus, I suspect EA ought to pay people a lot more, though there's a lot of nuance ofc. I'm also very sympathetic to viewing funding this as "retro funding for the SB1047 documentary".

  • My biggest question is whether short clips of existing content are providing value to the world; how much do you get to shape the narrative if you're curating vs creating? It's not a model I'm familiar with, anyways but I do think it's on the rise.

@MarcusAbramovitch I get that you want to achieve the highest impact per dollar and you'd want to find the most cost-effective option. However, I’d just like to offer some nuances / corrections to the points you’re making.

re 4-5x more impact per $:

1. You’re not only paying for the time spent on working on a project, but also the initial traction the project has, which might take a couple months to reproduce, or fail entirely. In other words, you’ll need to pay for the initial “warming up” phase, which is uncertain, and won’t have impact right away. And I believe the failure rate to get similar results in 1-2 months will be high.

2. Relatedly, you’ll need to find someone motivated enough to work with a low amount of views at the start, and that would commit to keep on working on this for months. This difficulty is exacerbated by the fact that you're also wanting to be 4-5x more cost-effective, so paying people 4-5x less. (This can be achieved in cheaper countries, but you're then trading other things like talent density).

3. I think this job requires a mix of skills from different fields that aren’t things that you can’t quickly pickup. Having personally tried to explain AI concepts to many freelancers with 10-20+ years of video editing experience, and conversely tried to explain short-form video editing to people with AI experience, I can guarantee that for similar projects to be successful you’d actually need people with a mix of both skills, which in my experience is quite rare, or would take a least ~1 month to mentor (but then you’d also need to find the mentor, which would bring you back to square one).

re "maximum earning potential": I think you're right to challenge the distinction between for-profit ML engineering salaries and non-profit video work, and I'll update the proposal accordingly. The actual datapoints I should have given are that: 1) this is already the rate people have been happy to compensate me when working with non-profit safety orgs to do video work. 2) when I do contract for for-profits I do actually tend to ask for more, both for video work and ML engineering work (at least ~30% more).

re "recoup": I was giving the documentary example to argue that I am indeed motivated by AI Safety. I might ask for retroactive funding later one, but this particular grant proposal is not trying to "recoup" anything.

@Austin Thanks for the kind words. The comments here have also been helpful for me to clarify how I'm thinking about things, overall red-teaming the proposal.

re short clips of existing content shaping the narrative: I do think that if you actually wanted to shape the narrative in a profound way, producing original content would be a necessary condition to be able to convey exactly the message that you'd want to share, which would be something that no other creator is doing.

I do however think that there is some value in amplifying content that is already posted that is currently neglected, and that to be able to do it well requires specialized skills (as I've argued in more depth in my answer to Marcus).

To give an intuition pump: I see this work as amplifying the impact of people who are already doing original work, but have not spent enough time (because they're time-constrained, not that interested in it, or don't have the skills) looking at all the possible claims / packagings that could be extracted from their work.

So basically you're taking something raw (an interview) and you're trying to extract the what important message there would resonate the most with some particular audience (say younger generations on TikTok) through being packaged correctly for a given algorithm. And if done correctly, I think this multiplier effect could be quite impactul and also deserves funding.

@Austin, I also like that he's being responsive. I also think many different POVs can be reasonable. I did not mean for my comments to tell people that they shouldn't in any case donate/give money here. I'm mainly adding my take here. My only point is that money is limited/has opportunity costs and I was mainly explaining why I chose not to.

We can discuss elsewhere on appropriate EA salaries.

@Michael sure, I agree I would need to pay for this warm up period and for them to scale up. I disagree though that this is highly skilled work that couldn't be picked up. I think your average EA who is interested in the topic could do this.
I didn't mean my comments to say "you should return this money". Lots of grants/spending in EA ecosystems I consider to be wasteful, ineffective etc. And again, apologies for singling you out on a gripe I have with EA funding.

@MarcusAbramovitch After studying the base rates on content creation a lot, I would argue that it IS highly skilled. It normally takes years to get any sort of meaningful traction on Youtube because it is extremely competitive. I think Michael's traction there would put him in the top ~2% of channels.

I've talked to maybe 100 EAs interested/working in content creation/media, and Michael is easily top 5 in terms of knowledge and how good his takes are. So I would be more confident that he's much more well positioned to tackle this than the average EA. Content is also very power law distributed, where a video that is 10% better than the competition might not receive 10% more views, but 10 times as many views.

Wouldn't be surprised if this project is getting 10m views per month within a few months.

Seconded -- I am interested in this project but want to hear more about what outcomes you hope to achieve from an expanded audience

I agree. Awareness is good in general, but some of the most watched clips don't really touch on AI Safety, and none of them have calls to action. ("Learn More Here," "Share This," "Call your representatives," etc.)

@NeelNanda Yes that was a typo, fixed it!

Regarding messages and outcomes (cc @NeelNanda, @Jesse-Richardson and @Haiku), see below my strategy which includes a diagram summarizing the approach (also included in the main proposal):

1. Messages: my goal is to promote content that is fully or partly about AI Safety:

   1. Fully AI safety content: Tristan Harris (176k views) on Anthropic's blackmail results, summarizes recent AI safety research in a way that is accessible for most people. Daniel Kokotajlo (55k views) on fast takeoff scenarios, introduces the concept of automated AI R&D, and related AI governance issues. These show that AI Safety content can get high reach if the delivery or editing is good enough.

   2. Partly / Indirectly AI safety content: Ilya Sutskever (156k views) on AI doing all human jobs, the need for honest superintelligence and AI being the biggest issue of our time. Sam Altman (400k views) on sycophancy. These help with general AI awareness that makes viewers receptive to safety messages moving forward.

   3. "AI is a big deal" content: Sam Altman (600k views) talking about ChatGPT logs not being private in the case of a lawsuit. These videos aren't directly about safety but establish that AI is becoming a major societal issue.

The overall strategy here is to prioritize posting fully-safety content that has the potential to have high reach, then go for the partly / indirectly safety content that walks people through why AI could be a risk, and sometimes post some content that is more generally about AI being a big deal, bringing even more people in.

2. Outcomes: Rather than adding call-to-actions at the end of videos, which unfortunately makes videos much less likely to reach a lot of people on Tiktok (mostly because people would exit instead of re-watching the video) and is quite uncommon to do on tiktok compared to Youtube, especially for clips, I'm expecting the outcomes to be:

   1. Engagement / Following: about 50k people (3-4%) engaged with the content (shares, likes, comments, follows). I expect that the people who engaged will continue seeing my content in the future (because TikTok will push it). In some cases, they will end up engaging more and more with the content that is directly about safety, and eventually integrate the broader AI Safety ecosystem (to a certain degree).

   2. Profile clicks: About 0.5% of viewers click on the channel's profile (I've received 5k+ profile views). The two outcomes from that are:

      1. Watching the pinned videos: 4k views on the 3 pinned videos came from these 5k profile clicks, meaning a large fraction who click on the profile click on pinned. I think in the future one of these pinned videos could be a video with a strong CTA that directly leads to outcomes we care about around informing the public / representatives about AI Safety, similar to this one which had a very high conversion rate in having viewers take action.

      2. Clicking on the link in bio: so far I don't have a clickable link, but plan to link to eg. aisafety.com to redirect to resources to learn more about AI Safety.

   3. Progressive exposure: Most people who eventually work on AI safety needed multiple exposures from different sources before taking action. Even viewers who don't click anywhere are getting those crucial early exposures that add up over time.

Gotcha, thanks! @michaeltrazzi

That seems a pretty reasonable plan and you've gotten good reach. I'm not confident this is a good idea, but I think that's plausible and more value of information here would be good, so I've donated another month's worth. Good luck!

Thanks @NeelNanda !

Thanks for sharing! My other question is how much time you're spending on this a week? Is the TikTok + YouTube stuff roughly a full-time job at the moment?

@Jesse-Richardson Yes it's full-time.

I wrote down more details in my answer to Marcus here.