"A big AI-era career guide for Software Engineers and CS students" - Part 3
On which parts of programming are safer:
@jacobmbuckman: If people want safer roles, they should orient towards anything where it seems like data would be harder to collect.
Anonymous: Anything involving human interaction, and where the whole point is that it's a human, will last longer. If it's remote and computer-based involving almost no human interaction, then yeah, I agree they may be automatable mid 2030s. But I'd also add, there's very very few things that involve no human interaction. Most software engineering jobs do involve lots of meetings. Even if those meetings are done remotely, there are lots of cases where it still matters that it's a human doing it.
@finbarrtimbers: I think the things that are going to be tougher to automate are the kind of higher level things. The harder problems are providing business value and navigating the humans within the organization. The closer you are to the real world or human interaction, the safer you are.
@IvanVendrov: There are also tasks that require deep physical intuition. Like a designer at Apple or something. There will still be a lot of value provided by humans doing that kind of thing.
@RichardMCNgo: Remote job automation depends a lot on whether it's the sort of thing you can do remotely in one hour, one day, one month or one year. The shorter duration tasks will be automated more quickly.
On where there might be more jobs:
@tylercowen: Lots of new jobs to come in energy and running medical trials!
@tamaybes: The industries required for the increased compute spend I expect to see more jobs, including fabs, energy, data centers, robotics. The domains where tasks aren't automated will have wages rise, and potentially quite drastically.
@jacobmbuckman: I think there might be an increase in the sorts of careers like enterprise sales or banking or upper management where a lot of it is really about what network you bring to the table.
On how you can get good at using AI:
@jeremyphoward: Using AI tools correctly takes months and months of diligent study and practice. When you start doing it, you will be shit and you will get bad results. That's because you're shit at it, not because AI doesn't work. NB: regardless of how good or how shit at it you are, you'll over-estimate how much it's helping you, because AI does the easy/early bits fast, but makes the harder/later bits much slower/harder, unless you're very careful!
@Jsevillamol: Understand the basics of how AI works. You don't need to learn to code but it helps a lot to understand how AI is trained.
Anonymous: There's a Tyler Cowen idea that AI is like having 1000 research assistants. People who are good at having research assistants should be good at using AI. What makes someone good at having 1000 research assistants or AI assistants? Basically it's around coordination. Packaging up work. Knowing what's important and having a really good sense of what you actually want. Invest in those skills. The important thing is to have a model of what AI can and can't do. One way of getting that is to understand how AI models are trained.
On career advice for AI researchers or people who want to become AI researchers:
@jeremyphoward: Mid-career AI researchers (in fact all levels!): focus on becoming really good coders. Learn to replicate interesting research papers from scratch. Code is the medium we use to experiment, so if you're better at it, you can run more complex and creative experiments more quickly.
@jacobmbuckman: Get as close to as many gpus as you can as early as possible. Almost nothing else you could do is higher value.
@rronak_: Stop studying, build. Go one layer deeper into the infra than feels comfortable, since that's where the value is. If you're on Langchain, write the agent loop yourself; if you're on Verl, write the pytorch yourself; If you're on Megatron, write the cuda kernels yourself.
@finbarrtimbers: People need to know you exist to give you opportunities. Write about interesting ideas you have or things you are thinking about. There is an extreme hunger for “interesting lunch conversation at DeepMind” level content (not hype boi threads, not paper level technical).
@_arohan_: 1. Perspective matters more than novelty, many so-called ‘solved’ problems still hide unsolved challenges in the details. Don’t dismiss anything as trivial; breakthroughs are hidden in plain sight. 2. Don’t worry about pedigree or fitting in. Hinton bet on neural networks when the field dismissed them. The real breakthroughs come from researchers who ignore the consensus, think for themselves, and tackle hard problems—and that researcher can be you
@danielhanchen: I would definitely watch MIT, Stanford videos much much earlier - CS231N, do FastAI courses, MIT's AI course, Gilbert Strang's courses CS229
On which personal assets may increase in value:
@tylercowen: Your personal reputation, your name, who knows you, who can vouch for you. Invest in all that.
@RichardMCNgo: Try to build relationships with people you can work directly with. Really good working relationships and high trust relationships feel important.
Anonymous: The optimal allocation of tasks between humans and machines will change a lot. My view is the biggest comparative advantage of humans will be in the interpersonal elements. Dealing with other people. Anything involving human interaction, social skills, big organizations, bureaucracies, interpersonal relationships, all that stuff becomes more valuable and more important.
@kenneth0stanley: You can change your interests. Your interests are not stuck. As change becomes more rapid, opportunism probably does become increasingly important, including the ability to uncommit.
@jeremyphoward: If you're at university, try to spend a lot of time doing other things, have a lot of side hustles, have a lot of side interests.
@nearcyan: If you’re a Silicon Valley person who is capable of doing good work and getting job offers and such: 1) It's probably a good time to find allies. 2) Integrity and honesty are more important than ever. Do not compromise on it for any type of quick buck. The bucks will become easier and easier to make.
@dlbydq: I would invest in the positional goods of career capital: networks, relationships, reputation.
On which skills may increase in value:
@finbarrtimbers: The thing that's really going to matter is taste.
Anonymous: Manager skills, especially technical management skills and organizational design, are very valuable. It's a very rare skill but that's precisely what you want to have if you're orchestrating AI. Designing the machine of your organization. Designing the right machine will become a really, really valuable skill. A highly valuable person in the AI world is someone who is great at dealing with other people, and who also really understands AI.
@Afinetheorem: Anything AI can do 100% will become close to free. Anything you uniquely do as a complement to that now free factor of production becomes more valuable. You are unlikely to be able to guess what those tasks are, so you need to experiment.
@RichardMCNgo: There will be very high returns to being able to orient fast to changes.
@Jsevillamol: Creativity is going to be super great in the future. If you want to make your own movie, in five years it's going to be super easy.
@BasilHalperin: AI is a complement for being agentic. "You can just do things" is even more true when AI can help you out, and when AI can scale you.
@kenneth0stanley: Taste is a really interesting unique human capability, which is underrated. I think using taste as a compass for knowing what stepping stone to go to next, that's currently something we're way better at. Be really loyal to your interests and tastes, not just casually, but seriously going deep into what you find interesting and committing to it in some way so that you can become uniquely skilled in that area.
Thank you to the people who gave answers:
@jacobmbuckman: Manifest AI, ex-Google Brain
@kenneth0stanley: Lila Sciences, ex-OpenAI, Why Greatness Cannot Be Planned
@Jsevillamol: Epoch AI
@jeremyphoward:
@answerdotai,
@fastdotai
@neonbjb: OpenAI
@_arohan_: Anthropic, ex-Meta, ex-Google DeepMind
@Altimor: Lindy
@danielhanchen: Unsloth AI
@jeffdean: Google DeepMind, Google Research
@nearcyan: Auren, Elysian Labs
@rronak_: Google DeepMind, ex-Windsurf
@ericjang11: 1X
@BasilHalperin: Economist, University of Virginia
@Afinetheorem: Innovation Economist, University of Toronto
@dlbydq: dmodel, ex-OpenAI
@IvanVendrov: ex-Midjourney, ex-Anthropic
@tylercowen: Economist, GMU, Marginal Revolution
@RichardMCNgo: ex-OpenAI, ex-Google DeepMind
@tamaybes: Mechanize, ex-Epoch AI
@finbarrtimbers: Ai2, ex-Midjourney, ex-DeepMind
Anonymous: ex-Frontier Lab