What does it take to make more Ryan Greenblatt-style thinkers / scientists? (Epistemic status: a lightly cleaned up stream of consciousness ramble, if I had more time / clearer thinking I would have made it shorter.)

I have been pretty impressed with Ryan's output for a while and I feel like it would be great to have many more clones of him. (I also personally aspire to be a very similarly shaped person.)

• In this taxonomy of AI safety talent proposed by Ryan Kidd, Ryan is clearly a Connector - someone who is great at prioritization and choosing between possible high-level directions based on original high-level insight
• To do this, such a person probably needs to be aware of + understand many different isolated things, draw connections between these things, and be very good at first-principles thinking
• They also need to have a high capacity for deep thinking about topic(s) of interest to them

Some takes on things he does especially well:

• Reasoning transparency. Generally he is very good at explaining how he arrives at his conclusions and this makes it very easy to both agree with him and disagree with him productively.
• Does a lot of experiments. Many things, e.g. "alignment faking", and "current AIs seem misaligned to me", are downstream of Ryan having spent a long time directly testing frontier models, reading transcripts, and drawing conclusions.
• High-throughput reading. From the outside view, Ryan seems to be terminally online, in that he stays extremely up to date on AI developments and usually has an opinionated take quite soon after major news breaks.
• High-throughput public engagement. Despite being presumably very busy Ryan maintains a high volume of interacting with the public through LessWrong, X, podcasts, and various other avenues. (While it's possible to go overboard here, I'd guess that most people do way less of this than they should, i.e. it's good to be advocating for your core beliefs, especially if you want action to happen based on that)

I would guess that some instrumental traits are:

• Operationalising abstract beliefs. He concretises many things, is very willing to do BOTECs / Fermi estimates and this appears to let him make non-obvious deductions with a high degree of accuracy. (E.g. he is within the top 10 of AI forecasters.)
• Willingness to share early thoughts publicly. Many people experience an aversion to writing up their thoughts because they're afraid they will be perceived as silly; resolving this seems pretty key to being able to write confidently and smoothly at a high volume (with appropriate epistemic markers)
• Intellectual breadth, e.g. having the intrinsic curiosity and willingness to engage w a huge array of topics. I'd guess he doesn't constrain himself much w.r.t topics he thinks about / does experiments on.
• Opportunistic experimenter. I get the sense that he has a big list of open questions and can often connect new observations to things he's thought about before. C.f. Richard Feynman and his 12 biggest open problems.
• Minimal trust investigations, c.f. Karnofsky. It's good to be able to re-derive things you believe from first principles as opposed to taking it on authority.