Bretton Goods is becoming Speculative Decoding

I started Bretton Goods as an endeavour in understanding the causes of the variations in the wealth of nations. Why are some countries rich enough to spend billions on sending men to the moon, while others struggle to pay for basic necessities like food and water? A meandering journey which included book reviews, analyses of financial crises and criticisms of the metrics we use to measure economic freedom led me to the answer that economic development happens because the elites of a country find it in their interest to do so. Some countries are stuck in unfortunate traps where pursuing development means they risk their power and don’t do so. A few lucky countries have leaders with the gumption to do so, or their leaders have nothing to lose after years of war and destruction. And so they take the risk, while the vast majority of countries don’t. Your economic circumstances are determined by the country you are born in, and its inscrutable intra-elite politics, and for the vast majority of people there is not much they can do about it. As dismal as this conclusion is, it is one I have come to terms with.

By the standards of a side project I intended to run in the gap between high school and military service, I have been moderately successful. The blog has over 2,000 subscribers, and has been featured in Marginal Revolution and the Diff. I’ve got an Emergent Ventures grant for the blog, and it has played a part in my acceptance to college, and other academic achievements. But on the other hand, my writing on this blog has been inconsistent. I wrote five posts in June 2021. I wrote only eight in the whole of 2022 and seven in the whole of 2023. As I started college in mid 2023, the blog slipped out of my attention span and I didn’t write anything for myself till the start of 2025, till I started posting on my personal website.

What has changed between now and then? Personally, I’ve a very different approach to my career. I thought I’d work in development policy, or a macroeconomics hedge fund. Circumstances made it so that this didn’t happen, and I started a degree in Computer Science at NUS. I’ve spent the better part of two years learning about monads and MIPS, instead of macroeconomics and mechanism design. In the world at large, the largest perceptible change has been the rise of language models, to the extent that they can now program and write in ways that are eerily human. As a person trying his best to be homo economicus I responded to this by getting an internship in GovTech working on making chatbots hallucinate less, and now am at Elicit working on AI evaluations to measure and improve Elicit’s products.

What then for this blog? This question has been answered by my other Substack Speculative Decoding. The name (much like Bretton Goods was of Bretton Woods) is a pun on the algorithm of the same name. The blog is about my new interests: AI evaluations (my field of work), being a scratchpad for my research ideas in AI safety, explainers, policy arguments and whatever catches my fancy. Much like Bretton Goods asked what makes human societies rich or poor, Speculative Decoding asks how do we know what AI systems are actually doing? The stakes of the problem — definitely bigger than the internet and possibly the largest technological change in human history — demand sense-makers who can dispassionately follow the evidence.

Given my abysmal track record at consistently writing on this blog, I’m not going to promise Marginal Revolution style daily updates. What I can commit to, however, is one post a month, on average over the next year, that answers questions that matter to the future of AI.

You don’t need to do anything. I’ll be archiving this substack, and porting all the subscribers over to Speculative Decoding by the end of the year. Please share Speculative Decoding with everyone interested in AI evaluations, safety, policy, and the future of AI more broadly.