TL;DR

The main thing I expect busy people want when receiving an EAG cold message is:

• Enough info to roughly estimate the expected impact they’d have by helping you,
• Conciseness

My recommended message structure:

• Background (1-2 sentences)
• 1-3 pieces of evidence of your future impact/potential (1-2 sentences each)
• What you want help with: Concreteness encouraged, eg specific upcoming choices (max 1 paragraph)
  • Why they, specifically, are a good person to help with this (skip if obvious)
• A concrete question you would like to ask them (makes it way easier for them to triage)

This advice is for messaging busy people, who will get more requests than they have time for, and need to prioritise, for less established people you can be way more chill.

I’m writing this because I think cold messages can be really valuable, and I hope this gives you the confidence to reach out to busy people, and helps create a community where busy people talk to the people they can most help.

Introduction

I imagine a lot of people are currently on Swapcard for EAG London, and are writing cold messages to people they don't know asking for help (eg advice or mentorship). I have a bunch of takes about what makes for a good cold message and what makes me more likely to respond/engage, so I thought it might be useful if I wrote some of them up. This is heavily informed by my own experience, but I hope it is more broadly relevant to people who get overbooked at conferences (e.g. speakers, experienced people in a popular career path, etc), as they get more requests than they have time for, and need info to help them prioritise.

What should the message convey?

At a high level, when I receive a cold message my goal is to estimate the expected impact I can have via helping you in the conversation. This is obviously very hard to estimate on such limited info. My ideal cold message gives as much useful info as possible to help me estimate that^[1], and otherwise cuts everything not useful to that, to be concise and quick to read. I'm very happy to have chats where there’s no direct benefit to me beyond being helpful, so there’s no need to sell me on anything else – I like being helpful! But I tend to get more requests than I have time for and, you know, it’s an EA conference – what else would I be prioritising by?

Should you send a cold message?

I generally think reaching out to people is high expected value and recommend it. A norm of EAGs is that it is normal and encouraged to send cold messages to people you don’t know! Try not to waste people’s time, but IMO so long as you’re clear up front about how mutually valuable the conversation would be, it’s fine to send whatever you want. For most busy people it's pretty cheap to receive and decline a well-written message, and helpful to get a message from someone they want to meet, so please don't feel discouraged to reach out! In particular, messages that follow the tips here should be easy to skim.

I’m writing this because cold messages can be really valuable! As an example, when my co-author Jemima was an undergraduate at Durham University attending her first EAG, she reached out to an expert on covid and China, and the advice she received led her to change her plans to study abroad in China that year. (She applied for Taiwan instead, and China stayed closed.) Similarly, talking to people in the UK civil service changed her graduate plans of joining the foreign office.

Recommended Message Structure

1. Background (1-2 sentences): your job/studies, recent professional history, research experience, etc. Give them context on who you are.
   1. The more your work is close to their experience, the more specific details become useful, e.g. “I worked with researcher X” is very helpful if they work in interpretability, not if they work in molecular biology
2. Evidence of impact/potential: Share 1-3 pieces of evidence that inform how impactful helping you is likely to be. These do not have to be standard credentials!
   1. Examples: things you’ve done that you’re proud of or think are impressive, papers you’ve been involved with, winning prizes or competitions that they might have heard of, working somewhere prestigious, someone they trust saying we should chat, etc
   2. This does not need to be an essay! 1-3 sentences is fine.
   3. I’m excited to get anything that should update my probabilities of your future impact, not just standard credentials
      1. For example, if you’re early career, pointing to a bunch of work you’ve been doing to learn and skill up is useful evidence of conscientiousness, which is a valuable trait
      2. Or personal projects you feel proud of (especially if you can communicate anything re how impressed I should be by them)
      3. More generally, indicating high potential/a good trajectory is great, and my standards are obviously lower there than someone with many years of experience.

   4. This can be awkward to write and feel like bragging - I’m personally pretty fine with this and value directness^[2], but others may vary.

      1. One good framing can be beginning with “More context on my abilities/background - apologies for the directness, but I thought this would be useful info on how to prioritise meetings:”

   5. This is also an awkward thing to suggest people write, since I’m sure some of you are feeling imposter syndrome reading that (regardless of whether you actually have such evidence or not). Sorry about that!
      1. If you’re worried about this, I recommend reaching out regardless, reaching out early, and to people at a range of busyness levels, and letting them self-select. You don’t need to do anything like this level of effort for less busy people!
3. What you want help with: (1 sentence to a paragraph)
   1. How, concretely, would this conversation be valuable to you?
      1. If non-obvious, how would this help you have more impact?
   2. I’m generally most excited to help people if I can see a concrete way that advice could cache out in value, eg helping someone choose between two job offers, advising on direction for a new research project, etc.
   3. It doesn’t have to be concrete, I also think it’s a solid use of time to offer advice to promising early career people who will face many concrete choices down the line.
   4. Why they, specifically, can help: (0-2 sentences - you can skip if it’s obvious how they’d be well suited to help)
4. A concrete question you want to ask them (at least one question, more is also great)
   1. Advice chats tend to be far more productive with people who have already thought about their issue a fair bit, this is evidence of that along with whether they, specifically, are the right person to help
   2. Bonus points: If they’ve written anything relevant, indicate that you’ve read that and why you still want to chat. If a conversation just becomes “go read this blog post, it’s more eloquent than what I can ad-lib in the next 30 minutes”, that’s not a great use of time.

Stylised Examples

Example 1

Hi Neel!

I'm a final year undergraduate at University of Texas in Austin, and am interning at Microsoft over the summer.  

In my spare time, I've replicated the paper 'refusal is mediated by a single direction' and wrote up a blog post here: [link]. I did the same for a couple of other concepts, and you can find the positive and negative results on my blog.

I'd really appreciate talking to you because I'd like to go into a career in mech interp and I'm scoping about a more ambitious next project, building on your paper on empirically eliciting latent knowledge. I'd love to show you what I've done so far and hear your thoughts on this.

In particular, I’d love to get your thoughts on whether linear probes or sparse autoencoders are the right tool here.

Example 2

Hey Neel,

I’m a software engineer. I’ve worked in startups for a few years, and I want to have more of an impact by moving into AI Safety.

I don’t have much ML experience, but in my previous role I was able to set up a pipeline to process incoming requests 30% faster while preserving privacy, using LLaMA 3.3 8B, with about two week’s work and upskilling.

I’m finding the career transition a bit difficult to navigate, and I’d love to get your thoughts on how I should approach upskilling, demonstrating my skills to employers, and the best places to apply. I’m most interested in mechanistic interpretability, but happy to find wherever I can have the most impact.

In particular, I dropped out of uni to join this startup, and seem to often fail resume screens and never even get invited to interview, how can I get past this? Should I be networking more?

Final note

Everyone at EAG was once sending their first cold message. If you're unsure whether to reach out, I recommend erring on the side of doing it! The worst case is a polite 'no,' and the best case could be genuinely helpful for your journey – in my opinion, the main reasons conferences exist are to help new connections happen, and cold messages are a crucial part of that.

1. ^{^}

    Note - please include key info even if it’s also in your profile, it’s much easier to skim

2. ^{^}

    Though try to avoid undertones of “therefore I’m really impressive and you’ll obviously want to talk to me”