I have noticed a change in my writing - as I started writing more blogs for PauseAI, and posting them to my substack, I feel like I have lost a bit of my writing style / personality in my writing. So, I thought I would write about my life in the last month, as I think it is a bit more personal, and I am able to write in a more ‘flowy’ fashion this way, which hopefully means more feeling will be portrayed in this post. I will probably only use substack for more personal blogs from now on.
In May this year, after getting more involved in Effective Altruism, I went to EAG London. I initially got into Effective Altruism due to a combination of liking the anti-animal torture views and wanting to get involved in AI safety. Disappointing as it is, I think I mostly wanted to get into AI safety because my quant aspirations were not materialising as much as I had hoped, and my underlying motivation of doing the most prestigious thing possible made me consider AI safety - thanks to my brother, I had somewhat of an ‘in’, as I had access to knowledge of the pipeline to follow. Thankfully, over the next four or so months, exposure to more of the earnest arguments for being concerned about the catastrophic risk of AI gradually shifted my motivation from prestige to wanting to help. I don’t necessarily think I am completely good-focused right now - I think I still definitely place a significant value on chasing prestige - but I do think my main drive is now wanting to do good. This is kind of an intuition, but maybe some evidence is that I have zero interest in doing finance any more.
Anyway, I came out of EAG London hungry to have some impact, but I lacked the necessary degree of agency. I had been in this limbo for a long time at this point - I had done my first AI-related project over a year prior, and was in a cycle of alternating between high motivation bursts and barely maintaining commitments to what I wanted to do. I had made a few useful connections, though, as I generally hung out with my brother and met his friends. I think that this experience was both motivating and demotivating - the people I met in my dream roles, or at a similar stage to me, seemed significantly higher-context and just generally quicker than me. Although I tend to be good at telling myself I am clever, crossing the gulf in skill between me and some of the people I spoke to seemed over-ambitious and unrealistic. I think I was also aware of how much more they were doing than me. I was planning on pushing to do some work over summer, but was aware that I would probably lose motivation, or be busy, or something might go wrong.
A few weeks after EAG, I got a text from my brother - his friend offered for me to stay in his house over August in London, so I could gain context quickly. I said yes, and also was lucky enough to get funding for my stay. I think this month’s stay in London was the most meaningfully productive month I have had in my life, super fun, purposeful, and maybe even slightly impactful. This is an approximately chronological log of what I got up to. I am also including non-work stuff, just because it was fun.
Started volunteering for PauseAI UK, writing their first blogs at pauseai.uk/blog
Bumped into Gymskin in Dalston
Wrote 20+ google docs on different papers, ideas, and blogs
Met an author of some of my favourite interpretability papers
Got rejected from ML4GOOD
Interviewed for SPAR
Saw the solar eclipse and had a picnic with onigiri
Wrote my first LessWrong post on Monitor Robustness
Worked in LISA ~2 times a week
Asked an Anthropic fellow a silly question about LoRAs and got a weird look
Met one of Aldo’s friends
Made a bedframe
Learned how deception probes work
Did a cool experiment where I elicited reward hacking with a probe and KL regularisation active
Ran twice (it’s a start)
Went to my first shabbat dinner
Went to a Pivotal party
DJ’d at the housewarming party with Bilal and Jay
Went to Notting Hill Carnival
Had anniversary pizza with my girlfriend
Started working on setting up an AI safety org for upskilling in the midlands
Saw the Belair Lip Bombs live!
Made pizza for the house twice
Made friends with a cat called Poppins
Carried a picnic bench and a shelf through some London streets
Learnt how to be more agentic! I highly recommend reading this post by Neel Nanda.
I made ~7 new friends, meaning I would comfortable asking them for a favour / hanging out
I met ~6 friends that were in London at the same time
Met a lot of really cool people
Worked hard every week!
Overall, I feel like my foundations are significantly stronger than before, although I am not super confident on how little I knew before the month (it’s been a month of constant learning). I was also pretty busy socially, which means that overall it felt like a maximally fulfilling and exciting month! I think my main takeaway was learning to be agentic: I think going forward I want to keep the drive I felt in London, and generally do reps of my ‘agency muscle’. This means hosting dinners (hopefully regularly) in my uni house next year, staying ambitious, finding short feedback loops for ideas, and not getting lost in goal proxies. In regards to the gulf I sensed between me and others in the field, I noticed that as I spent time learning foundational knowledge, building and testing opinions with actual researchers, and publishing my output, I started being able to keep up with the conversations that I couldn’t understand a few weeks prior. This was a really validating experience, and I think it has helped me believe that I can actually have some impact.
I really hope I don’t eat my words, and actually keep this momentum. Hopefully one day I can repeat this London experience, with more impact under my belt, sometime when the world hasn’t ended yet.
Thanks for reading! I’m also excited to talk about things in general and I feel like I should increase my presence on substack, so feel free to leave a comment or message me or something!