https://arunim.fyi Last 10 notes on arunim.fyi Quartz -- quartz.jzhao.xyz https://arunim.fyi/tokenomics https://arunim.fyi/tokenomics Inference BOTEC I spent about 15 minutes doing some quick back-of-the-envelope calculations to understand the economics (“tokenomics”) of inference for large transformer models ... Sat, 01 Aug 2026 18:30:11 GMT https://arunim.fyi/creating-instead-of-consuming https://arunim.fyi/creating-instead-of-consuming Relaxing by creating Chris Lakin had this great post back in the day (archived here) about relaxing by creating instead of by consuming. I haven’t read it in a while, but it still lingers on account of both my bad habit of poor quality relaxation and my related media consumption addiction. Tue, 28 Jul 2026 18:30:08 GMT https://arunim.fyi/projects https://arunim.fyi/projects A number of projects that I’ve started or thought of at some point or another. I don’t care for getting scooped so if you’re interested, please go ahead and do them! Then, do tell me about it so that I can thank you / ask you about how it went. Tue, 28 Jul 2026 18:30:08 GMT https://arunim.fyi/fog-of-war-chess https://arunim.fyi/fog-of-war-chess What is Fog of War Chess? Fog-of-war chess is by far, my favorite game in the world. Fog-of-war is a variant of chess (available on chess.com) where you are only able to see the squares your pieces are or where your pieces can move. Mon, 27 Jul 2026 18:30:07 GMT https://arunim.fyi/advice https://arunim.fyi/advice Some advice, I’ll update this occasionally. Note that I think most advice applies to everyone. Some people say all advice is bad, to them I say this isn’t advice despite being titled advice, and is instead just reminders of obvious facts about the world. Sun, 26 Jul 2026 18:30:07 GMT https://arunim.fyi/RSI https://arunim.fyi/RSI Defining RSI What is RSI?? I think it means that humans exit the loop of AI development. But like … how much? It could be the case that humans are cheaper than automated alternatives for a large variety of physical labor for some time. Fri, 24 Jul 2026 18:30:08 GMT https://arunim.fyi/Utility-Engineering https://arunim.fyi/Utility-Engineering www.emergent-values.ai/ arxiv.org/abs/2502.08640 As AIs rapidly advance and become more agentic, the risk they pose is governed not only by their capabilities but increasingly ... Fri, 24 Jul 2026 02:20:42 GMT https://arunim.fyi/ml-notes https://arunim.fyi/ml-notes todo organize these! Training ZeRO stage 1: data parallel all params on all gpus local loss, local gradients construct the global gradient on every by averaging the local gradients ... Fri, 24 Jul 2026 02:20:42 GMT https://arunim.fyi/risk-aversion-in-llms https://arunim.fyi/risk-aversion-in-llms TLDR: The (at the time of writing) nascent field of risk preferences in LLMs is currently evaluating models in a suboptimal way which too-directly analogize LLMs to humans, and result ... Fri, 24 Jul 2026 02:20:42 GMT https://arunim.fyi/Calvin-and-Hobbes https://arunim.fyi/Calvin-and-Hobbes Calvin and Hobbes is truly a top tier piece of media for me. Maybe mostly due to fortune, where a family friend gifted a box set of two decades of Calvin and Hobbes to me after occupying my room for some time. Sat, 18 Jul 2026 18:30:09 GMT