I think the world would be a better place if there were more documentaries on AI risk on Netflix, and I’m glad that Connor and Mike are putting in the work to try to make it happen.
Making any kind of long-form films involves many costs, which can make things quite expensive. Connor says in a new proposal they raised $300k already, are looking for $327,500 more, where the overrun would go into festival route ($102k), which puts the budget for the entire project at a minimum of $730k. After talking to Mike, the cost will actually be higher since the salaries of Connor & Mike will actually be paid by private investors during post-production.
For comparison, in a 2023 survey of UK documentarians (n=74), only 13% had a budget over £400k, and 8% was above £800k, which places “Making God” at around the top 10% of UK documentaries in terms of budget (given that survey).
Given the high budget, I think it’s worth clarifying:
- what is the total amount that will be needed for the project, including money being currently raised through private investments? how much total has already been raised (private / public)?
- how much of it has already been spent?
- from the spending, how much has been spent on things personally benefitting the team (salaries, housing, purchases of equipment like cameras) vs. general expenses (paying for the crew, renting equipment, flights, hotels for shoots, etc.)? what will be the split moving forward?
- when trying to sell the documentary to distributors, will you be trying to sell it for more than the cost of the documentary? will there be a discrepancy between non-profit funders who will not get anything from the sell vs. for-profit funders / people who got equity who will get some shares from it? what’s the current non-profit vs. for-profit status of the project?
Regarding some of the details in the track records:
Worth clarifying that Mike mostly did motion graphics / editing for the SB-1047 doc, and only ~10 days was spent doing art director work.
60M+ total views in 9 months & 32M views/engagements in one month: can you do a breakdown of where these numbers come from? How much of that was organic vs. paid for. (Note: Mike told that “Your identity isn’t yours” had ~50% views coming from promotion. Tiktok’s version has low engagement so seems promoted to).
Collaboration with Max Tegmark: how much was Max involved?
Apollo Research: do you have links? How long was the collaboration?
Commercials (Starbucks, Pale Waves and Mandarin Oriental): do you have links?
In the other project you write “directed high-reach campaigns and long-form documentaries”. What long-form documentaries did Mike direct?
I think for a documentary like that to be really impactful (think the “Social Dilemma” reaching 100M+ people) you need to have a lot of criteria being met, such as a certain number of exclusive interviews (with whisteblowers giving exclusive interviews, eg. from labs) and / or very high-profile people. On top of that you’d need distribution (think Netflix), which will require the criteria above, but even then it’s not a given, and will depend on the overall quality of the documentary.
Having said that, let’s look at the interviews currently listed:
Yoshua Bengio and Geoffrey Hinton have strong AI credentials, though they’re not really “insiders” anymore. Additionally I don’t find Yoshua to be a great public speaker (eg. Ted) and Hinton has already given many long-form high-reach interviews repeating similar arguments (eg. this one), so I’d like to know how much new stuff he’s said in these interviews.
The Gary Marcus interview seems to be high-production value (based on what I’ve seen on trailers online) and Gary is a decent public speaker, representing a slightly different view. However, Gary isn’t the most respected AI researcher and often comes off as unreasonably contrarian. For instance, consider this interview with MLST where everyone in the comments just wanted Gary out of the interview.
Rose Chan Loui: seems great to have her legal expertise, considering her knowledge of the complex legal procedures happening at big AI labs such as OpenAI. However, her 80k interviews seem to have little traction (1-2k views) compared to other 80k interviews at 35k-164k views, and she doesn’t strike me as having done a lot of public speaking.
Will MacAskill: great to have him. I think he’s a good speaker, though he’s already done quite a lot of interviews and is not really an AI insider nor has specialized AI knowledge relevant to this.
I reckon they have some “whiteblow-y” interviews, such as one with Sulaji’s mom, though she was already on Tucker Carlson and (last time I checked) the actual content of what she says about OpenAI has not yet proven to be true. There’s also Daniel Kokotajlo that is planned, but at this point he’s already been on dozen podcasts and his message is well known. So would be worth if you could clarify what interviews are really key here.
Or if no specific interviews are key but it’s more about the content, it would be great to clarify what is the particular narrative here that makes this set of interviews bring a coherent original narrative. For context, I have talked to Tristan Harris & one of the editor of the social dilemma about a documentary they’re also making on AI, which I believe will be out soon and might overshadow this one.
Additionally, I’ve seen some early-stages excerpts of what seems to be one of the angles of the documentary, where you have some amount of behind the scenes of Connor & Mike going on filming people, and I felt like the behind-the-scenes distracted from the interviews and didn’t bring a lot to the table.
Regarding the post-production budget, having myself managed a budget on that front, this strikes me as something on the higher-end, especially considering Mike’s & Connor’s salary not being included here, which would make the overall budget higher:
General notes on using a production studio & motion-graphics studio: using a studio means that you’re basically trading a lot of the time that would be spent on dealing with hiring & logistics, with the studio doing a lot of the management for you. I think if you’re tight on time and have a high budget that’s something worth considering, but in that case if Mike & Connor are being paid for their time and actively working on the project, the studio route appears to be on the more pricey & “easy” side, which would significantly reduce the hours being worked but require paying a premium rate.
Post-production studio ($130k, 40% of non-salary budget) for “professional editors, sound designers, and colour graders can deliver the film quickly”. It seems that this would be paying a premium fee to get things done quickly, and I’m not sure why this is needed given the overall length of the project & the movie festival route. Worth noting that colour grading takes 1-2 weeks and sound design is also quite short, so most of that would be editing. If we say $100k is spent on editing, that means working with a high-end studio that would provide a lot of help (senior editor, assistant editors, maybe post-supervisor) at $12k / week for 8.5 weeks (given timeline). This doesn’t seem like an absurd amount but is definitely on the higher end. Especially given the large buffer ($40k) and that some things (like a post-supervisor) would usually be Connor’s or Mike’s jobs (paid elsewhere).
Archival footage ($50k) for “news reports, podcasts, and other third party content”: licensing TV news footage can be pricey, though can be negotiated down if purchasing from the same source multiple times on bulk. This essentially gives some amount of flexibility to license the key footage that they’d need. Podcasts can definitely be negotiated to much lower amounts. Would say this is mid-to-high range.
Stock B-roll footage ($40k) for “For elements that we cannot film (due to access restrictions) we will license clips, for example, the insides of data centres, political institutions, etc.”: what kind of access restrictions do you have? Can you hire someone to film that for you remotely? This seems extremely high, especially given the $50k for licensing, and ~$36.5k for motion graphics. Usually, for a high-end documentary you’d try to film these yourself or get people to film it. And even if for some reason you couldn’t possibly get someone to film it, data centers and political institutions are the most easily accessible forms of B-roll. To be more concrete, let’s say the mid-range b-roll is $80 and the pricey ones are $500. This would give you 80-500 clips depending on price. That means one mid-range to high price clip not filmed by your crew nor licensed every 15s to every minute of a 90 minute movie. With the licensing budget, that’s $90k, meaning ~27% of post-prod going to content you didn’t film.
Motion graphics studio ($30k) + graphic designer ($6.5k): similar to the post-production studio, I think this is paying for a premium to outsource a lot of the work. Having worked with motion graphics people, most of the work is in giving the motion graphics person the right things to do, but when that’s clear the work doesn’t require a lot of time (compared to editing). For comparison, a motion-graphics person at the top of the industry was asking for $3k a week, so this would cover 10 weeks of her work. And given their timeline they don’t have that amount of time for motion graphics. (Probably 4 weeks maximum). The fact that they’d be asking for 6.5k for a graphics designer on top strikes me as either duplicating work done by the motion graphics studio, and / or making the total be on the very-high end. One other datapoint would be hiring something like Rational Animations for the entire motion-graphics work, and I believe that would come at about $10k / minute of top-quality complex 2d animation, or $1k-minute of simple animation. This strikes me more as the simple type than the very complex type.
Post-production buffer ($40k): given everything is already on the higher end / handled by professional studios, having this large of a buffer (12% of budget) for things going wrong makes me think the budget is not really tight. If all of the studios end up charging you $40k more, are they actually doing their job carefully? Wouldn’t you agree on a fixed rate beforehand? Where would these $40k come from?
Composer $20k + $8k musicians: having talked to a bunch of composers, this would give you a good composer. In LA (where things are much more pricey) you can get quite experienced composers for $30k. And they usually include the musicians in their package. I expect that if you work with UK-based ones the price would be lower. You can also negotiate the price down if they end up having the rights to the music, instead of wanting to buy the full rights.
In conclusion, here are the main takeaways of this comment:
1. all-things considered, this is the budget for a high-end documentary (top 10%), with some private funding that would be worth clarifying
2. given that, a lot of the previous / future expenses are worth being documented, especially the non-profit vs. for-profit aspect, and the private / public funding
3. some of the track record could also be more detailed, with more info on what was done, and the actual projects / metric breakdowns
4. the current documentary proposal seems to lack what would make it a very impactful netflix documentary, and it’s unclear what original narrative would make it compelling
5. the post-production budget appear to be on the very high-end, with some amounts that suggest a lack of planning, especially considering the director & producers would be compensated through other sources of funding, 27% on licensing / stock footage, and the fact that they’re working with post-production and motion graphics studio at a premium
I should also say that making a high-quality feature-length documentary is very difficult and I really want this to succeed. A lot of my comments apply to what I think a high-end feature Netflix documentary would need to be successful (which they’re aiming for), but I understand that this is a very high bar, and I wouldn’t want anyone reading this to think that what they’re doing isn’t worth doing. I think this is useful work and I’m happy that Mike & Connor are doing this and getting all these interviews. We need more high-quality AI Safety documentaries!
Looking forward the answer to these questions / comments on this. I think being able to see some promotional material of the interviews already shot would also help me (and others) understand better where the project is, and its potential.
(Previously retracted comments were asking similar things but contained errors. Also crossposting to the post-production proposal)
