Things That Make Me Enjoy Giving Career Advice

By Neel Nanda @ 2022-06-17T20:49 (+34)

This is a linkpost to www.neelnanda.io/blog/49-mentoring

Introduction

I spend a bunch of my time having various kinds of mentoring conversations with people. I generally enjoy doing this, but some things make these conversations much more fun for me than others! This post is an attempt to compile some of these.

Obviously, writing this post is somewhat self-serving, since I hope future people I mentor might read it! But I expect there’s decent chance that at least some of this will generalise to other people. And, hopefully this is also useful to mentees! Based on my experiences trying this, approaching people for mentorship can be pretty anxiety inducing. And having a clearer model of how to make this more fun for them and more respectful of their time helps make reaching out feel easier. But note that there’s a bunch of things in this post - please take them as some tips and nice-to-haves, rather than as eg things where I’ll judge you if you miss out some of them!

Context: The central example of a mentoring chat I imagine in this post is a career coaching chat, where I talk to someone who wants to work on AI Safety, who I don’t already know, and don’t necessarily expect to talk to again. But many of these generalise to other kinds of mentoring chats, eg where I’m orienting more towards giving life or productivity advice, trying to help out a friend/someone I know better, etc.

Tips

Context Setting

Useful things to happen at the start of a call, to give me an idea of what’s going on and how I can best help:

• Making it clear what you want out of the call, and how I can be most useful
  • An easy way to do this is to explicitly state your goals and key uncertainties at the start
    • A common mistake here is trying to only state key uncertainties that sound good/rational. If your key uncertainty is a less tangible emotional crux, like “I basically buy the arguments that AI Safety matters, but feel emotionally uncomfortable with the idea of working on something with a small chance of mattering”, say! I can’t be useful if I’m focusing on stuff that isn’t relevant.
    • A good litmus test - if you changed your mind on the question we’re discussing, how would this change your actions?
  • Another angle is setting out the kind of call you want. Do you have a few concrete options you want help deciding between? Is this more of an exploratory conversation to figure out what’s out there? Do you want to dive deep into a specific question, or do you have a long list of questions you want to cover briefly?
• A sense of how you’re orienting to the call - are you seeing me as an authority figure who you want to give you answers, are you taking this as one source of advice among many, are you fairly self-confident and expect to mostly stick to your priors vs open to updating a lot, etc.
  • This is pretty awkward to explicitly say, but is fairly easy to make implicit when discussing what you want out of the call
• Your background and situation
  • To first order, the main thing I want here is the ability to roughly type-cast you - eg someone who just wants an introductory pitch, someone who’s heavily engaged and already gotten a bunch of the low-hanging fruit, etc.
    • This one is pretty annoying to figure out on my end! My main tactic is asking a series of questions about what you’re already familiar with, but this risks being patronising to engaged people and alienated to less engaged people
  • To second order, it’s also really useful to hear anything weird and maybe relevant about you, which can sometimes significantly adjust the advice I want to give.
    • Have a low bar for sharing things here, it’s hard to judge what is and is not relevant!
    • This includes less tangible traits about you, eg if you really love reading papers or crunching through code, find self-study hard vs easy, etc.
• Hearing about advice you’ve gotten from other people, and things you’ve already tried - helps me specialise into having the biggest marginal add
  • Hearing about what you’ve already tried is particularly helpful! Often a major value add is for me to try giving “obvious advice” - people’s sense of the obvious things to do can differ wildly! But it can be kind of awkward to just list a bunch of things the mentee has already tried.
    • Eg, for some people ‘cold emailing relevant people’ is an obvious thing they’ve already tried, while some people would never think of that.

Low Effort

My ideal mentoring chat is one where I don’t feel a need to drive the conversation - the other person knows what they want, can ask me things, and I can just be reactive. This feels much easier than a conversation where I feel pressure to drive it and figure out how best to be useful. Sometimes being reactive involves talking a bunch, or trying to do something specific with the conversation, but the key difference is not needing to drive it, and not having awkward-feeling gaps.

• Know what you want
  • The ideal thing is to brainstorm a list of concrete questions beforehand
    • If you feel stuck/aversion at this, set a 5 minute timer and see how many questions you can generate!
  • A common handle here is “do your homework”. I think I care less than most about whether you’ve eg read all the obvious resources, etc, but I do care about people doing their homework in the sense of knowing what they want, what they’re uncertain about, how they want to be helped, etc.
• Write up a doc with context and your uncertainties/questions
  • I find zero-prep calls much lower effort, and so ideally I could just turn up to the call and live-read and comment
    • Note: 2 1hr zero-prep calls feel much lower effort than to me than a 1hr call with 1hr of prep - needing to make time for prep feels much more effortful than just turning up to a call! I expect this preference generalises less though.
• Keep the conversation going, have questions queued up that you ask at natural pauses
• It’s much easier to be reactive than active - if you give me a bunch of information, I can react to it, give relevant points, correct parts, etc all pretty easily
  • Eg, describe your current plan, given your existing knowledge and goals
  • Eg, describe your current best-guess understanding of a topic you’re confused about
  • Eg, give your best guess for a key uncertainty
  • It’s easy to feel insecure here - if you don’t have a clear picture of something, it can feel embarrassing to try describing it and risk messing up in front of someone who knows more than you! But I think doing this can be really worthwhile for helping me figure out how best to help and where you’re currently at. I recommend prefixing what you say with eg “I expect I’m missing some important details, but my current best guess is…”
• A sub-point of being reactive is paraphrasing - after I’ve said something, try repeating it back to me in your own words. This makes it super easy to correct any misunderstandings.
  • If you feel insecure here, my advice is to prefix it with “OK, so if I understood you correctly, what you’re saying is…”
  • Paraphrasing is great. I expect with fairly high confidence that someone will misunderstand some details of anything I say. Paraphrasing helps me identify and target the things that didn’t come across, and helps me build a better model of you.

Make it more useful to you

A common mistake is to focus too much on helping the mentor feel good about themselves and trying to optimise for what you think they want - eg nodding along to things that didn’t quite make sense to you, because you don’t want to say you’re confused. 

• The main way of doing this is by directly giving feedback in the call:
  • If you already know what I’m saying, say so!
  • If I’m going too fast or too slowly, say so!
  • If I’ve misunderstood what you said, say so!
  • If I’m talking about stuff you don’t care about, or failing to engage with your crux, say so!
  • If something I’m doing in the call isn’t working for you, say so!
    • This includes things like “actually, I’m really tired right now and this is majorly preventing me from getting value from this call. Can we reschedule?”
• In general, the way I orient to mentoring chats is that you are the expert on your life, not me. I have some general knowledge and priors you may not have, but you are much better placed to judge what is and is not applicable to you, and everyone needs subtly different advice. This means I will say a bunch of stuff, and not all of it will apply, and some will need subtle nuances - the only way I can figure this out is if you pushback and correct me on things!
  • You don’t need to have well articulated pushback to do this - just saying something like “hmm, I’m not convinced by what you just said, but find it hard to articulate why” is great - we can try to figure out what went wrong together.
• Prioritisation - in general, I expect 90% of a mentoring chat to be kinda useless, and 10% to be particularly high value. You likely have some idea of what kinds of topics could lead to high value advice vs not. And this differs between differently people. Try to target the conversation onto those! We’ll never have time to talk about everything relevant, and need to triage!
• The reason I have mentoring chats with people is because I want to actually add value and be helpful, not just think I did! This means I want you to directly optimise for what would make this useful to you.

Minimise my anxiety

One thing that can make mentoring chats much more draining is things that cause anxiety, where I’m eg afraid of doing a bad job, hurting the other person, or otherwise giving harmful advice. This is a messy section and hard to properly ameliorate, but I think some things can help!

• Give a sense of how receptive you are to bluntness/discouragement/negative feedback
  • I often talk to people who have ambitious goals and want to get into competitive paths. This has a high chance of failing! Both due to maybe not being good enough, and due to dumb random noise. I think it’s important to go into this kind of thing with well-calibrated expectations, and to seriously consider not doing it if eg you’re unusually non-robust to failure. And thus, I think an important role I can play is setting expectations, calling out flawed or overly ambitious plans, and giving blunt feedback. But this feels bad, and can really hurt some people, which makes this pretty hard!
    • This comes up a lot because often the best way to maximise your impact is by pursuing high-risk, high-reward strategies. And high-risk strategies are, well, high-risk. In particular, most EA careers and most careers to do with AI are competitive as fuck. Often, pursuing a competitive path is the best decision, since there’s a lot of randomness and uncertainty and it could go super well, but this can be a difficult thing to advise someone to do!
  • Please only signal that you’re OK receiving blunt negative feedback if you actually are!!
• Don’t take me too seriously
  • My ideal way for someone to orient to a conversation with me is “I will try to give useful thoughts, some will be right, some will be wrong. I only vaguely know what I am doing. You have way more context on your situation than me, and it’s up to you to filter my advice into what is and is not good for you”
  • Sometimes, people have a way higher opinion of me than is deserved, and put way too much stock in what I say. I think this is both empirically incorrect, and makes things feel way higher stakes for me!
    • In general, giving advice is hard and noisy. If you ask someone else, they’ll likely disagree on a bunch of points!
  • I tend to present socially as being very confident, which makes this something I’m particularly concerned about
• Unrealistically high expectations
  • There’s a bunch of pretty hard and thorny questions that go along with giving advice. What should you do with your life? How to choose between wildly different options?
    • This is particularly bad with AI Alignment related things: Which AI alignment agenda is most promising? Is AI Alignment more valuable than AI Governance? What are AGI timelines?
  • I’m pretty happy to give hot takes on questions like this. But these are really hard and I don’t have good or confident answers to them! And it can feel hard and frustrating to confront this kind of confusion, and to want someone else to give you answers. Which is totally understandable! But it feels pretty bad to chat to someone who I think expects me to be able to give them answers to this.
    • A decent chunk of this may just be in my head though, *shrug*
    • A concrete thing that could help with this is suffixing a hard question with “obviously this is a hard question, and I don’t expect you to have a confident or perfect answer to this, but I’d love to hear your takes!”

Appreciation

Part of why I do mentoring chats is the warm fuzzy feeling of actually adding value. And having chats where I feel this more strongly is great! But just expressing thanks doesn’t work great, because this says more about your personality than about whether this was a great vs decent vs meh chat. The key things that help this for me are being sincere, being specific, and being concrete.

• Mention specific takeaways you found valuable
• Mention specific things I did in the chat that were helpful
• Give feedback on things in the chat that could have been more helpful - on an object level this is just useful, and on a meta level the willingness to give negative feedback makes me trust positive feedback way more!
• Long-run follow-up
  • Hearing about concrete things you did differently, or how you’re doing a few months on feels great! This is a much better proxy for actually adding value than how someone feels at the end of the call
    • I particularly like getting people’s estimate of the counter-factual value add (it’s fine if this is ‘probably not that counter-factual, sorry’)