In this post, I outline some advice (and links to other advice) that I find myself giving increasingly frequently, despite having less and less time to do so. My hope is that in future I will be able to redirect people to this post. I also hope that, if nothing else, the resource list at the end of the post (which is the most comprehensive such list that I know of) will be useful.

For those who prefer watching to reading, I gave a closely related short talk at the Cooperative AI Summer School earlier this year. The talk is slightly less targetted towards EAs (i.e. EAs will find some of the advice very familiar) and the examples I give are about cooperative AI, but the key content is the same.

Introduction

As a mere PhD student myself (and not even one of the most successful PhD students I know) you might well ask what qualifies me to proffer such advice.

First, one of the more difficult things about doing a PhD is that there is very little deliberate instruction on how to do it, and much of the instruction one does receive is banal, bureaucratic, or both. Most people will have probably heard most of this advice before they finish their PhD, but I for one would have benefitted from hearing it much earlier.

Second, while it is easy to dismiss the advice as simple or obvious in retrospect, the important thing is not being aware of it but deeply internalising it. I frequently find myself in positions where: a) I am not following this advice; and b) I would be happier and more successful if I did follow this advice. Having it written down somewhere means I end up in these positions less frequently.

Before beginning, some caveats:

• I expect this advice to be most useful for people doing – or interested in doing – a PhD, an AI PhD, and an AI safety PhD, in increasing order. It is also tailored to me; your mileage may vary.
• Most of the advice is selected from the resources listed further below – written by people much smarter and more successful than me – mostly without attribution. I initially compiled the advice for myself,^[1] much of whose advice and resources I copied into my own advice document and going through everything to find the correct attribution didn't seem worthwhile.
• This advice is on how to do a PhD, not whether to do a PhD.^[2] It is also not especially about what to do after a PhD, though working back from that goal is an important part of how to do a PhD. Much of the advice is relevant to independent researchers too, but not all of it.
• The post ended up longer than I expected, but it's possible to skim and/or only read certain sub-sections. If you only have a couple of minutes I would suggest reading the 'Highlights' section and skimming the 'Resources' list at the end to find other things that you might want to return to later.

I used to update the document from which this post is distilled relatively often. Recently these updates have been fewer, but I may still update this post in future. At the very least, I will add new advice resources to the list at the end of the post. If you have additional advice that you think is missing, feel free to leave it in the comments below.

Advice

The primary activity of your PhD is research, which can be split into doing and disseminating. This second part is vitally important and should not be underestimated. Other key skills and more mundane advice is covered towards the end. If short on time, prioritise the highlights and look through the list of resources further below.

Please also note that simply reading this advice is not enough to benefit from it. You must deliberately take the time to stop, zoom out, and proactively do it. For example:

• Coming up with your own research agenda is not something you do in your head while in the shower, you should block out at least a day to write a first draft, then come back to it again and again;
• You cannot prioritise longer-term development just by thinking about it half-heartedly: make a list of what skills or knowledge you're lacking and brainstorm concrete ways to improve (like working through a textbook with a friend or helping a co-author code something up that you're unfamiliar with);
• Don't just think about what you want from a meeting or event once you're already there – take 5-30 minutes the day before to sit down and write yourself an actual checklist.

Highlights

If you can successfully etch these points into your mind such that they become second nature (and sincerely follow them), you will be well on your way to becoming the most impactful researcher you can be. Note that the points can be thought of as mere taglines – I go into more detail on all of them further below.

Do

• Work hard (on the right problems, at the right time, the right way), there is no substitute for it.
• Focus mostly on solving important/big problems (be goal-driven, not idea-driven), even if you end up taking small steps towards a solution.
• Try to create your own personal research agenda that is different to that of your supervisor(s).
• Think critically and efficiently when reading papers, attending events, taking part in meetings, etc. – know what you want, prepare in advance where necessary, and get to the point.
• Run occasional self-assessments/reviews about your progress and make updates to your projects or working patterns accordingly.
• Optimise transaction costs (lower those for good practices, increase those for bad, and minimise opportunity cost).

Don't

• Work in crowded areas or play it safe – be bold, take risks!
• Spend time on things that aren’t valuable, your time is precious and short and you should manage it consciously.
• Work on too many projects at once – a good aim is to have one main (first author) project and one side-project at any time.
• Forget or forsake long-term development and skilling up for short-term rewards.
• Be co-opted by standard academic incentives – doing the most important thing needn’t result in more papers or improvements along other standard metrics.
• Forget to look after yourself and others around you.

Doing Research

Unlike one's prior education, research cannot be boiled down to mere problem-solving. Much of the challenge of research is about asking the right question, rather than answering it. I therefore focus below on finding and prioritising problems, with only a few brief comments on solving them. Finally, I highlight the importance of efficient reading and effective collaboration.

Finding Problems

• To get important results you need to ask important questions.^[3]
• Don’t invent problems; if you can’t think of at least three examples of your problem, then it’s probably not a real problem.
• If you can’t do something (e.g. not enough compute), why not? Turn getting around that problem into your research question.
• Finding a new, different lens on a problem (adding it to a ‘portfolio of lenses’) can be very useful and form a significant contribution.
• Identifying messes can be a good way to figure out where there might be opportunities for simpler, novel perspectives.
• Find senior researchers whose work you admire and ask them to suggest problems they think people should work on, as well as for feedback on your own ideas.
• A common failure mode in PhD students that are far smarter than I am is that they fail to review previous literature adequately.
  • At least when starting out, your null hypothesis for most research questions or methods should be that it has already been done – later on you will get a better sense for the state of the art.
  • At the same time, it is important to avoid the paralysis induced by the feeling that all the (interesting) tractable problems have been posed or solved already – this is not true.

Prioritising Problems

• It is an extremely useful exercise to try to create (and continually refine) your own, personal research agenda, which should explain:
  • How and why this research matters (i.e. a theory of change for helping to reduce existential risks from AI);
  • How this agenda fits into what other people are working on;
  • How the projects you are doing (or want to do) fit into that agenda.
• Recall Hamming’s famous quote: “If you’re not working on an important problem, and it’s not likely to become important, then why are you working on it?”
• Try to prioritise goal-driven (this is where I want to get to, now how do I get there) as opposed to idea-driven (follow some literature, come up with a new idea that hasn’t yet been tried) research where possible – the difference is between making something work for the first time vs. making something work better.
• ‘Incremental work’ is the worst adjective possible in academia, avoid ideas that seem like an obvious next step that is waiting to be done.
• Difficulty is not necessarily a good indicator of the importance of a problem, think instead about the new areas opened up, connections uncovered, questions asked, etc.
• Work on problems that are suited to you, think about your comparative advantage. What is your ‘secret weapon’? What ‘unfair advantage’ do you have over everyone else?
• Always ask yourself: how does this fit in with my long-term plan? Is this likely to be the most valuable use of my time?
• Some additional safety perspectives:
  • Before you’re actually ready to make real contributions to the field, try to avoid rationalizing doing things because “they might help with safety”; instead, do things because “they might help me personally to understand safety better, in ways that might be idiosyncratic to me and my own learning process.”
  • Publishing good papers is not the problem, deluding yourself is. Doing things you don’t see as a priority but which other people are excited about is fine. You can view it as kind of a trade: you work on something the research community cares about, and the research community is more likely to listen on (and work on) things you care about in the future. But to make a difference you do eventually need to work on things that you find impactful, so you don’t want to pollute your own research taste by implicitly absorbing incentives or others opinions unquestioningly.

Solving Problems

• Remain open to changing your approach, progress requires change. Put another way: it is extremely unlikely that your favourite method or approach will be right for all of the important problems you might want to solve.
• Aim for crisp, easily communicable insights – form an explicit hypothesis, even if it is one that you later refine.
• Be realistic: it is unlikely that as a very junior researcher you will be able to easily solve important, difficult problems; you should balance work spent on such problems with smaller, more easily achievable results.
• When working on a project with multiple uncertain components of varying difficulty, do the components in order from most informative per unit time to least informative per unit time (not easiest to hardest, or vice versa), measured by either of:
  • Expected time saved (if an earlier component fails, how much time will you have saved by not working on the other components?);
  • Failure rate (assuming the chance of failing occurs independently at any time point, what is the chance of failing per unit time for each component?).
• Research as a stochastic decision process: you want to maximise the probability of success while minimising the amount of time spent.
• It is far better to do an outstanding job on a moderately sized project than an average job on a large project.
• Ruling out ideas and general approaches towards a problem is valuable, and spending time on it will often be worthwhile in the long run.
• Before pitching or starting a project, attempt to answer the Heilmeier Catechism. The answers can also form a useful basis for disseminating your work later:
  • What are you trying to do? Articulate your objectives using absolutely no jargon.
  • How is it done today, and what are the limits of current practice?
  • What is new in your approach and why do you think it will be successful?
  • Who cares? If you are successful, what difference will it make?
  • What are the risks?
  • How much will it cost?
  • How long will it take?
  • What are the mid-term and final “exams” to check for success?

Reading

• Make (rough) notes on every paper you read, and store these notes directly in your reference manager so that everything is in one place.
• Read (chapters of) textbooks, not just papers, especially when coming to a big topic for the first time.
• Know when to stop reading: you do not need to have read every paper on a topic before you start working on it. Instead you should:
  • Give yourself a rough grounding in the topic;
  • Start work, and return to the literature as and when you need to (and you will need to).
• Sometimes it’s better to read a follow-up paper rather than the original paper if you just want an overview of an idea – the follow-up paper will cleanly and concisely give a summary of the original paper in a few paragraphs, and may also point out its limitations.
• Work out what you want from a paper beforehand, then read it accordingly.
• Consider using a three pass system, proceeding to the next pass only if necessary:
  • First: abstract, intro, headings, conclusions (10 minutes);
  • Second: whole paper, minus proofs, code, etc. (1 hour);
  • Third: in depth, comb through things step by step (1-5 hours).

Collaboration

• In academia, cold emailing people can get you a long way (as long as you demonstrate a strong understanding and interest, and tailor your communications accordingly).
• Research visits and internships are are a great way to build connections, work on something different, and open up possible future job opportunities – make the most of them.
• Try to make advisors and other more senior academics want to help/work with you (e.g. by showing a very strong work ethic and/or framing things to their tastes).
• Make an agenda for every meeting where the other person is ‘doing you a favour’ (e.g. supervisions, academic superiors, extremely busy people). Consider sharing this with them in advance, or at least sharing your notes with them afterwards.
• Keeping your door (mind) open won’t make you more capable, but it may help you to know which problems to work on, and to find new ways of working on them.
• Play to your advisor’s strengths, you want their help and their knowledge – in an ideal world they will be mentioning your research when they give talks about their own work.
• At conferences:
  • Think about the purpose of any particular meeting/event;
  • Make sure to structure your learning and participation;
  • Talk to people, and prepare to talk to people (moreover, prioritise this above presentations – they are recorded nowadays anyway);
  • Take notes, but rank interesting things to follow up on by importance and relevance to your own research.

Disseminating Research

There is little point to doing good research if no-one ever hears about it. Moreover, unless you happen to have a famous supervisor or co-author, people will not hear of your work without sustained effort on your part. A viral tweet or a popular blog post can be the difference between a paper with hundreds of citations and one that fades into obscurity.

Papers

• Read Jakob Foerster’s guide for writing ML papers.
• Keep an eye on AI conference deadlines, locations, and ‘rankings’:
  • AI Conference Deadlines;
  • Top AI venues on Google Scholar.
• A paper is a single significant idea, identify this before you start writing. It can help to draft an abstract first (instead of last) and continue referring back to it, and adjusting it as necessary.
• When writing up, the paper should be a compelling story, including a strong narrative ‘flow’ and a clear, logical structure.
• Most people will only skim your paper, therefore (and also to potentially entice them further) the abstract, introduction, headings, and conclusion should be crisp and strong. Pay attention to the opening sentence(s); if possible try to concisely indicate the objective of the work and why it is important.
• Optimise for a few excellent papers, rather than a large quantity of papers. In slogan form: “quality trumps quantity”.
  • Your job talk and interviews will likely matter more than your CV, and having a couple of big hits for the job talk will be really helpful.
  • No-one is impressed by publications in some random conference/journal, take the time to get a high quality paper into a venue that matters.
• Finish your paper (at least) a week before the deadline.
• Ask colleagues and friends to read your paper and give feedback (and offer to do the same for their papers).
• Provide an intuitive description or informal explanation for any proof or theoretical result.
• When writing rebuttals:
  • Summarise the reviews and main issues up front (this is primarily for the area/program chair);
  • Structure your rebuttal carefully to clearly address these issues;
  • Use a warm but firm tone;
  • If pointing out flawed reviews, attack the review, not the reviewer;
  • Be prepared to make concessions;
  • Add revisions to your paper in a different colour.
• When deciding on a title:
  • Shorter is better – use the most minimal title you can without being too imprecise or grandiose;
  • Joke/pun titles are almost always incredibly lame and no-one else will find your joke as funny as you – avoid them;
  • The ‘<Catchy Quip>: <Actual Paper Title>’ formula is trite and mostly unhelpful (exceptions include using the name of your method or model, followed by the title).
• The biggest stylistic mistakes (independently of the quality of the content) I see from people who are new to writing papers are:
  • The content and tone is too pedagogical (too many basic things are explained, too many worked examples are given, etc.);
  • The literature review is incomplete and/or too informal (missing large sub-fields, hyperlinks to blogs or Wikipedia pages, poorly formatted references, etc.);
  • The structure and flow isn’t clear enough (not enough sign-posting, not enough thought about ordering and breaking up sections, etc.);
  • The latter point also (separately) applies to proofs, which often look like the answers to questions on problem sheets, rather than neat mathematics and clean explanations.

Presentations

• Read Matt Might’s tips.
• Give presentations as often as possible (practice makes perfect). While you might think that the median presentation is (almost by definition) averagely good, the median presentation is, in fact, objectively quite bad – you should be aiming much higher than this.
• Always rehearse your talk.
• Tell a story:
  • Good stories are thoughtful, engaging, clear, and concise;
  • Break up the talk into short ‘episodes’ and leave time for questions between them.
• When creating slides:
  • People love (relevant, informative) pictures, use them;
  • Don’t have more than a few sentences on any slide – they should not be serving as your presenter notes;
  • If using a Mac, consider using KeyNote (which supports LaTeX equations);
  • Avoid:
    • The standard LaTeX ‘beamer’ template;
    • The same cliche AI images (e.g. white humanoid robots on a blue background).
  • Consider making your own personal slides template – this is a small upfront cost that will have more than paid off by the end of your PhD.
• If you don’t have time to cover something, skip it entirely (or better yet, don’t put it in the talk in the first place). In general, don’t cram in material and plan to end slightly early (you’ll overrun anyway and no-one ever complained about having a few spare minutes due to a shorter presentation).
• When watching, always try to evaluate the style of a presentation as well as the content – what can you adopt to improve your own presentations?
• Don't be afraid to move the talk on if the audience gets bogged down in discussion mid-way through.
• Giving a talk is only partly about telling your audience about your paper/recent work, more importantly you want to be getting them excited about the problem you’re working in, and to teach them something (ideally including your solution).
• Maintain a smooth gradient between simple introduction and more complicated later sections, don’t let there be a big jump between the two.
• Your null hypothesis should be that the majority of the audience don’t really care about what you do, and will not read your paper. Assume the question the audience has in their head is "so what?", and tell them the answer explicitly.
• Ask the audience questions and encourage interaction where possible.
• Prepare a final conclusion statement in case there’s time left but no more questions – it’s a good way to end the talk on a high note.

Posters

• Err on the side of less detail over more detail – your poster should function as slides for a ~5 minute talk (see advice for presentations above) – and keep text and images clearly legible from several feet away.
• Make sure you bring:
  • A printed version of the paper;
  • A bottle of water;
  • Chewing gum or mints.
• Know your audience: make sure you understand the background of the person you're talking to and be prepared to update your pitch and summary in response.
• As with slides, consider producing a personal template the first time you make a poster, which you can then re-use.
• Add a QR code for your paper.
• See more tips for good posters here from Aaditya Ramdas.

Thesis

• As with your overall PhD, your thesis should essentially comprise around two or three major research contributions. Even if your university does not allow for thesis by publication, you should still view the chapters of your thesis as roughly corresponding to papers.
• While your thesis does not have to be that coherent (especially if doing a thesis by publication), the ideal outcome should be that people view you as "the person who did X" and your thesis is a good opportunity to solidify this perception.
• Allan Newell’s thesis types:
  • Opens up new area;
  • Provides unifying framework;
  • Resolves long-standing question;
  • Thoroughly explores an area;
  • Contradicts existing knowledge;
  • Experimentally validates theory;
  • Produces an ambitious system;
  • Provides empirical data;
  • Derives superior algorithms;
  • Develops new methodology;
  • Develops a new tool;
  • Produces a negative result.

Blogs and Social Media

• Ideally, every paper you write (and at least those you are a first author on) should have:
  • A summary tweet thread;
  • A short (5-10 minute) blog post.
• When writing tweet threads:
  • Have an image in your first post and get the hook/point of the paper across within the character limit;
  • Ask more senior collaborators, advisors, and colleagues to retweet your tweet;
  • Focus on boosting one tweet thread instead of multiple people giving different versions (if they want to do this they should do so by quoting the main thread);
  • Include a link to the full paper at the end of the thread, as well as an accompanying blog post and/or a video of you giving a talk about the paper;
  • Include pictures and graphics to get your point across (again: people love pictures);
  • Keep it short (ideally the thread should contain no more than 10 posts).
• When writing blog posts:
  • Your aim is not to compress a full paper into something that can be read in 5-10 minutes, but to get the most important message and findings of the paper across (and to encourage interested readers to look at the full paper);
  • Cross-post to other fora;
  • Know your audience (for more technical posts, a good yardstick is to aim for is a master's student or a first year PhD student who works in AI but in a totally different subject);
  • Refer to the advice for presentations above.

Other

A PhD is unlike many other jobs in its lack of structure and how little explicit instruction you will likely receive. Creating your own working habits is one of the more freeing but potentially challenging aspects of being a PhD student.

Planning

• While it may not seem this way when you start out, your PhD is short (assuming you want to achieve a lot). Never forget this.
• Have a plan for your PhD, and the time during which you are completing it – know what you want. Concretely:
  • Have a checklist for your PhD (collaborations, internships, papers, events, experiences, etc.);
  • Make time for a quarterly or biannual retrospective (i.e., a self-assessment/performance review) in order to assess whether you are on track to achieve your long-term goals.
• Keep a regular (though suitably flexible) schedule:
  • Use your good time for deep thinking (e.g. mornings);
  • Use your bad time for meetings and mundane/routine tasks (e.g. afternoons);
  • Block out a few (ideally, consecutive) days a week solely for research, with no meetings and minimal admin;
  • Check emails/messages/etc. only a small number of times per day, and respond in bulk;
  • Consider blocking out some additional time (perhaps half a day a week) for reading or other self-improvement.
• Keep a unified to-do list across all of your projects. Similarly, sync all of your calendars and email accounts to be accessible from one app.
• Use version control for everything (you can easily use git with OverLeaf as well).
• Eliminate time-wasting distractions and opportunities for wasteful procrastination wherever possible:
  • Consider tracking your time with something like Toggl if you want to manage/maintain your working time across different projects;
  • Put timers on social media apps or distracting websites, and get a friend to set the password so you won't be tempted to override the timer.

Software

• I use/recommend:
  • VSCode for coding and writing LaTeX;
    • You can sync your local LaTeX files to OverLeaf using git;
    • You can get Copilot for free as a student and it will change your life.
  • Jabref for a minimal but powerful reference manager built on top of your .bib files;
    • Zotero is also good.
  • Obsidian for notes and keeping records of projects;
    • Notion is also quite good;
    • Roam is not to my taste but your mileage may vary.
  • Reclaim for managing to do lists;
    • Tools like Motion, Reclaim, and Sunsama automatically schedule your tasks, though this can be overkill for some;
    • Todoist and similar tools are also fine.
• Maintain a LaTeX template with your macros, comment commands, and preferred formatting. I also suggest reading Adam Gleave's LaTeX advice.

Websites

• Use Google Scholar to make sure your papers are listed online accurately and subscribe to relevant researchers to get alerts every time they publish a new paper.^[4]
• Consider using Twitter to follow relevant researchers and hear about the latest developments in AI (though note it can be distracting).
• Create a (somewhat professional) personal website, e.g. using GitHub sites, to provide a landing page for you and your work, and offer contact details – this is more important than you might think.
• Connected Papers and Elicit can be useful for uncovering relevant research that you might not find just using a regular search engine.
• Subscribe to the newsletters of the main AI companies, as well as the following overview newsletters (ordered roughly by priority):
  • Transformer;^[5]
  • Center for AI Safety;
  • Center for AI Policy;
  • Daniel Paleka’s Newsletter;
  • Import AI.

Teaching and Mentoring

• There are four good reasons to teach at your university:
  • Wanting an academic job (as hiring committees will look for experience);
  • Needing the money (though this is not usually the most lucrative opportunity that will be available to you);
  • Improving your communication and presentation skills (this is not to be underestimated);
  • Enjoying it (no, really!).
• You can get most of these benefits from teaching one or two courses early in your PhD, doing more than this is unlikely to be the best use of your time.
  • This applies somewhat, but less so, to supervising bachelor’s or master’s projects, as they might contribute to your own research, and/or lead to publication.
• Make an effort – it really doesn’t cost very much at all to be substantially more engaging and helpful than the median tutor or mentor.
• There are tonnes of mentoring opportunities out there nowadays – if you are a senior PhD student you are probably capable of being a good mentor.
• When supervising/mentoring projects:
  • Scope the project carefully, and pick the difficulty level carefully – things will take longer and be harder for people than you expect;
  • For more junior researchers, try to avoid overly 'conceptual' projects;
  • Be generous with your time and responsive, but make it clear what the best way to interact with you is (e.g. asking people to send written notes at least 24 hours in advance of a meeting for comments) and set expectations accordingly;
  • Be encouraging and inclusive – part of your goal should be to get your mentee/student excited about the project and progress.

Reviewing

• Summarise the paper first, even if you’re not required to – this is a good exercise in making sure you have understood the main point of the work.
• Consider:
  • Relevance – Is the paper relevant to the conference and the state of the field?
  • Significance – Does this paper make a good contribution? Is it too incremental?
  • Novelty – Does the paper demonstrate new techniques or ideas?
  • Technical Quality – Does the paper make sense? Are the proofs/reasoning valid?
  • Writing Quality – Is it clear? As many writers do not have English as their first language, be generous.
• Respond to rebuttals and consider updating your scores accordingly. Moreover, when submitting your original scores, think in advance about what – concretely – it would take for you to increase them (and signal this in your review).
• Always be constructive: if something was bad, how could it be improved? Be concrete and helpful, imagine you are the authors’ research collaborator.

Resources

As noted further above, most the advice above is selected from the resources linked below (the list is in no particular order):

• Rohin Shah’s FAQ for Alignment Researchers;
• Andrew Critch’s:
  • Career advice for junior AI-x-risk-concerned researchers;
  • Deliberate Grad School.
• Andrej Karpathy’s PhD Survival Guide;
• Richard Hamming’s You and Your Research;
• Jakob Foerster’s How to ML Paper;
• Neel Nanda’s:
  • Concrete advice for forming inside views on AI safety;
  • How I Think About My Research Process sequence;
• Matt Might’s posts on:
  • 10 easy ways to fail a Ph.D.;
  • 3 qualities of successful Ph.D. students: Perseverance, tenacity and cogency;
• John Schulman’s Opinionated Guide to ML Research;
• Toby Shevlane’s How to succeed as an early-stage researcher: the “lean startup” approach;
• Michael Nielsen’s Principles of Effective Research;
• Jacob Steinhardt’s Research as a Stochastic Decision Process;
• Ethan Perez’s:
  • Tips for Empirical Alignment Research;
  • Tips and Code for Empirical Research Workflows (with John Hughes);
  • How I Select Alignment Research Projects;
• Mark Xu’s How to do theoretical research, a personal perspective;
• Chris Olah’s Research Taste Exercises;
• Aaditya Ramdas’:
  • Checklists for Research;
  • Poster checklist.
• H. T. Kung’s Useful Thoughts about Research;
• Dorsa Amir’s Modest Advice for New Graduate Students;
• Michael Bernstein’s slides on:
  • Velocity;
  • Vectoring.
• Gabe Mukobi’s ML Safety Research Advice;
• Marius Hobbhahn’s Some advice on independent research;
• Adam Gleave’s Writing Beautifully in LaTeX;
• Michael Aird’s:
  • Building a Theory of Change for Your Research;
  • Tips and readings on getting useful input from busy people;
• Alex Turner’s Looking Back on My Alignment PhD;
• James Chua’s and John Hughes’ Tips On Empirical Research Slides (with Ethan Perez and Owain Evans).

1. ^{^}

   I was inspired to do so by the wonderful Mrinank, much of whose advice and resources I copied into my own advice document.

2. ^{^}

   For those interested in this question, I recommend reading Adam Gleave's (outdated, but still very good) Careers in Beneficial AI Research document as well as relevant posts from 80,000 Hours.

3. ^{^}

   This is near-tautological but it bears repeating.

^{^}

Neel Nanda has a good tweet thread about this here.

^{^}

I have the honour of being the source of the newsletter's primary marketing line – "The only AI newsletter I read all the way through" – and it is no exaggeration.

Show all footnotes