I seem to be somewhat successful in getting opportunities (e.g., work trials, contract positions, interviews from Redwood, Bluedot, FarAI, etc.) at first contact with someone, especially from large conferences. Here is how I approach my meetings so that I can get what I want.
Much of getting opportunities in AI Safety comes down to, fortunately or unfortunately, socializing with the right people. Fellowship acceptance rates drop continuously, but getting face time with potential mentors increases your likelihood of acceptance drastically.
I will discuss:
Who to meet with
How to prepare for your meeting
What to do during your meeting
What to do after your meeting
Figuring out your deal
Before anything, you should figure out what your “deal” is: your strengths and weaknesses, what kind of things you’re looking for, what fields/industries you are considering, etc. A great exercise for this is completing a career doc such as this: Memo on my career . Afterwards, you should consult with people you trust who can advise you on whom to talk to at an upcoming conference.
Goal-Setting
In order to figure out who you want to meet with, you should first figure out what you want from the conference you are going to. I suggest you go to a conference with specific goals, such as getting a work trial at organization X, or earning mentorship from one of a group of people. Your approach to your one-on-ones will differ greatly depending on these goals. You can derive these goals from your career doc and from suggestions of those around you. If you don’t have people to talk to, consider talking to Claude or getting 80,000 Hours or EA career advising.
A realistic goal might look like: “I want to schedule one-on-ones with 10 industry professionals in subfields I am considering. I want to talk to them about their experience, advice, and how I can break into their field. I want to maintain contact with at least half.”
Tips for deciding on people to talk to
Oftentimes, conferences like EAG or other AI Safety conferences will have a ton of cracked people whom you’d want to talk to but can’t all meet. You should ideally scan all of the profiles and identify a list of people you could talk to, then narrow this down as you approach the conference date. Also, you can try to plug every participant’s profile into Claude, plug in your career doc, and ask Claude about the top 20 or so people you should talk to.
Generally, you should meet with people you can learn from but also provide value to. One-on-ones should be mutually beneficial, and you should present them as such. For example, it might not be worth it to reach out to some top researcher if you have only just entered the field. On the other hand, you should reach out to people to whom you can bring value in the form of research discussion, advice, or otherwise. That shouldn’t discourage you from shooting your shot, of course; just prioritize people you can bring mutual value to.
Actually organizing a one-on-one
Once you have a list of people you want to talk to, you should aim to schedule one-on-ones with them. Make sure your list is maybe 2-2.5x longer than the number of people you actually want to talk to so that if half decline, you still have a reasonable number of people to talk to and retain optionality. Many conferences use Swapcard. Here’s what you should do:
Write a connect message based on their interests and your interests:
“I would be interested in discussing reward-seeking, and especially your new blog post X with you. I find that Y may be incorrect and would be interested in discussing it.”
“I saw that you are working in AI Safety communications. I have recently been thinking a lot about comms and have been working on a related project, X. I’d be interested in discussing our approaches.”
Write a meeting message; it can be similar to the above. Avoid scheduling over events they’re attending, which you should be able to see on their Swapcard profile. Also, obviously, don’t choose bad times like early mornings or near the absolute end of the event.
You should find a way to talk to people you want to have a one-on-one with before you schedule a one-on-one or just before your one-on-one in general. This will give you a better idea of whether you should have a one-on-one with them, and they will have you in mind when considering who to one-on-one with so they’ll be more likely to accept.
Additionally, you should optimize the spacing of your one-on-ones. One failure mode is that you get burnt out too early and all of your subsequent one-on-ones are of lower quality. You should then space your one-on-ones accordingly such that you have time to rest. But you should approach a conference with the intention of getting incredibly exhausted because you are engaging so much, since if a pivotal moment is going to happen anywhere, it’s here.
Prep
You should be very familiar with what the person you’re meeting wants and model their theory of mind properly. For example, if you schedule a one-on-one with a senior researcher and you’re not one yourself, they might just be looking for signals of potential or to give you advice. Generally, you should (1) learn as much as you can about the person and their body of work, and (2) figure out why they want to talk to you. I have found success in reading people’s most recent technical blog posts, asking about their experimental decisions, and then letting the conversation flow from there. Similar strategies work even with non-technical people. This signals that you care and can think critically, which makes a strong impression.
The meeting
I highly recommend reading this piece from Jeremy because I agree with everything he says. You should:
Think about what the social situation calls for
Don’t press for certain topics when it would feel unnecessary or forced.
Be as helpful as you can, but also look for help. Conversations are give-and-take.
Be adaptable: don’t go in with rigid expectations; let the conversation flow, but when you see an opportunity, take it and press.
Be clear and articulate; don’t mumble.
Be overall presentable (e.g., brush your teeth, shave, wear something that makes sense).
Post-Meeting
Follow up!!! You must follow up the night of the meeting or the day after, then keep chatting with them if it makes social sense and wouldn’t be awkward! This is where you can ask further questions or claim opportunities that are on the table. A good follow-up might look like: “I enjoyed our chat yesterday! I found what you said about X interesting and thought Y about it. I was wondering if you had more thoughts or would be open to getting on a call to chat about it later.”
Conclusion
You won’t have a 100% hit rate; in fact, it will likely be quite low unless you’re really great at this. Being good at one-on-ones comes with practice. I suggest practicing one-on-ones with people you are closer to using these techniques adapted to them. Afterwards, you should ask for feedback and update accordingly.
I think people underrate the power of having great one-on-one skills. Often, getting hired is a matter of displaying competence and being known to people who matter. One-on-ones, when done right, can achieve both, which makes them very effective for landing work trials. I would imagine that anyone currently in a position you want is probably great at one-on-ones, so you should be too.