What I Would Work On If I Were Starting My PhD Today
An incoming PhD student recently asked me over email what I would work on if I were starting my PhD in the current landscape. This is a blog version of my reply. My original email was 100% human-written, but I thought it was worth sharing, so I had Claude Code (Opus 5.5) turn it into a blog post and then added some more musings and thoughts. This post assumes we will hit at least a narrow form of AGI/ASI within the span of a normal PhD (~5 years). Most projections, including my upcoming blog on RSI timelines, suggest this is quite possible, barring sudden policy changes.
1. Applying AI to other fields
If you work on popular topics like bio, drug discovery, or hardware design, which are of great economic importance, you might end up competing with startups backed by VCs who are more than happy to make risky bets and throw a lot of money at you. You could partner with a startup for your PhD, but then why not just work for them and make a much bigger paycheck?
I would pick an application that doesn't get VC funding right now. Finding one takes some research and building contacts with people in SF. You really need to make sure you won't be competing against startups on it, while still being able to convince a funding agency to give you grants to study it. You might think: wait, then why would anyone care about this when I graduate? I'm not suggesting working on something that will never be of economic importance (e.g., producing dad jokes on demand), but rather something that isn't economically viable today, where there is no good, known theoretical reason it can't be in the future (e.g., recycling plastic in a way that's actually profitable).
Finding such an application is a challenge in itself, so here is the best concrete example I came up with, after watching Hank Green's Theranos video. After the Theranos scandal with Elizabeth Holmes, no one wants to invest in wearable diagnostic devices. However, Elizabeth Holmes was really bad at execution and didn't invest in cutting-edge tech. You could point powerful models at this problem and do good work, since VCs are scared to touch it, at least for now, because of the field's bad image and unclear path to monetization (a domain expert Hank Green interviews in the video explains this).
2. General ML/LLMs
Let's say AGI starts doing all ML research better than humans. How do we still audit that research? How do we make sure the AI explores the topics we care about and are most curious about? How do we make sense of its reasoning traces, trajectories, and findings, and explain them to humans?
Finally, can we point this AGI at itself and/or the human brain to understand how intelligence works on a fundamental level, and what the recipe is for increasing it (if it can even be increased)? Do we make ourselves more intelligent through biological augmentation, fuse with AI (I was writing a sci-fi story about this but haven't had the chance to finish it lol), or something else entirely? Personally, I just prefer biological augmentation as cyborgs are creepy to me. I just want a regenerative healing factor like my favorite comic book characters Wolverine and Deadpool.
3. Social
If we reach superintelligence or AGI, how do we make sure native speakers of whatever language you speak get the same benefit from it? Most top LLM companies focus more on English, and their models might carry American/European cultural values and biases. The stakes go beyond fairness: the steam engine was one of the driving forces behind British colonial rule in South Asia. The uncomfortable truth driving the AI FOMO among countries that are lagging right now is that intelligence at scale is perhaps the greatest weapon you can wield.
Final thoughts
These are really lofty goals, and with the tech right now, I doubt a single PhD could achieve them. But when I started my PhD, first-authoring 7-8 papers over a whole PhD would get you called "cracked"; now a Claude Code subscription is all you need (sorry for making another all-you-need joke). Anyway, never be afraid to update your prior on what is possible and dream bigger when you see tech advancing! You require a certain level of delusional optimism regarding what is achievable and what you are personally capable of to survive the bitter lesson of doing a PhD :P