Hi Cathy, I think you touch on the need for generalists not just to find roles within existing orgs but to actually found orgs, start projects, etc. independently. I believe this is a viable path, as there are grants willing to fund projects with a sound theory of change.
I also wonder if a lot of the talk of bottlenecks is anticipatory regarding the upcoming IPOs. Not sure!
Emily Marais
Karma: 131
Plugging the generalist gap in AI safety with freelancers
Really happy to hear this—I would love to see opportunities that don’t require relocating for months at a time.
Really good point that the field is selecting for those can withstand the uncertainty. I wonder if that’s to some degree an appropriate selection i.e. in a field that relies on funding from a relatively small pool of donors.
I’m very keen to review what you’re working on, whenever is helpful. It would be also be useful to get your thoughts on how much operations work can be projectised, and what are the barriers to contractors/freelancers being used by orgs. I’ll message you so we can set up a call.
Interesting thoughts.
On your second point, I hadn’t thought that the need for alignment was related to the risk of someone going to work on capabilities. I think it’s more about understanding the landscape, the terminology, knowing how to communicate to people in the space, etc., i.e. things that might affect your ability to do good work. Maybe alignment wasn’t the right term for this.
Thanks for your thoughts here, Mark. I have the same sense that short-term contracted work offer a win-win scenario for those transitioning and for orgs. I imagine starting by interviewing people in hiring positions within orgs and getting a sense for what is preventing them from using freelancers would be very useful.
Hi, that sounds very helpful! I’d love to see your curriculum. I have also built myself a personal tracker for all of the various tasks involved in transitioning into AI safety—we could have a call to share notes as I think we’ve taken different approaches to the same problem and I’d like to see what’s working for you.
Yes, I think you’re right—it’s not bad advice for many people.
Experience of a ‘generalist’ transitioning into AI safety in 2026
Very enlightening. I’m most interested in the following excerpts:
1: Talent, not funding, is currently the binding constraint on the AI safety field.
2. I think field-building targeted at more experienced people who can go on to start and lead competent organisations (and who have networks to convert a load of other experienced people) is much more important than more junior field-building. I think most existing AI safety fellowships act as internship programmes for the labs, which is fine, but we should be doing more than that.I’m wondering what specifically are the talents needed to start and lead competent organisations in this space? What gaps are there between existing orgs? And is that the only kind of talent that is constraining the AI safety field?
I agree that this article is written in a very confusing way. Particularly this paragraph, which introduced many new ‘things’ into the article, without anchoring them soon enough:
”Suppose that, after a tree falls, the two arguers walk into the forest together. Will one expect to see the tree fallen to the right, and the other expect to see the tree fallen to the left? Suppose that before the tree falls, the two leave a sound recorder next to the tree. Would one, playing back the recorder, expect to hear something different from the other? Suppose they attach an electroencephalograph to any brain in the world; would one expect to see a different trace than the other?”
This would have been much easier to read with an added sentence:
”Suppose that, after a tree falls, the two arguers walk into the forest together. Will they expect any different sensory experience from eachother? Will one expect to see the tree fallen to the right, and one...”
A good ancestor takes on the set of problems that they can have the greatest impact in solving. If today there is an opportunity to prevent a global technocracy from taking power, we must tackle that problem, as future generations will not be likely to have a chance to overthrow such a powerful entity. If a future generation will have an equal chance of solving a certain problem, and delaying the solving of the problem doesn’t cause immense suffering, it is not the ancestor’s most pressing problem to solve.
The section on Taking Ideas Seriously reminded me of a recent piece of writing that landed in my inbox. It’s by Henrik Karlsson, and it is about a way of processing information actively which we rarely do—actively pushing back against the ideas we form. You can read it here: https://open.substack.com/pub/escapingflatland/p/how-i-read?utm_campaign=post&utm_medium=email
Hi Pooja, thanks for sharing your experience. A 99% rejection rate sounds tough… Are you reaching out to cold connections looking for a job? If that’s the case, I think it would be more fruitful to do some voluntary work and test your fit/make connections that way.
I wish you the best in finding your fit!