My attempt at "email journalism", inspired by Prof. Efros's page. These are in reverse chronological by month, but chronological within months. To be honest, I'm not sure if Prof. Efros's email journalism consisted of actual emails, or whether it was meant to be more expository. I just assumed the latter... I haven't actually been sending these to people. Eh. Either way it seems fun.
š November 2026
Date: Sun, 6 Nov 2026 10:12:42 +0000 (PST)
From: Aathreya Kadambi aathreya@stanford.edu
Subject: A Quarter for Your Thoughts (Adjusted for Inflation)
Itās been way too long since I last posted on this log. On some fronts, I feel like Iāve grown more than I could have imagined in the past 3 months. On others, Iām itching to go faster.
I started my first quarter at Stanford! Itās been pretty fun so far. I bought a PC with 96GB RAM but honestly I need to make better use of it.
AI and machine learning research has taken a crazy turn after Anthropic and OpenAI demonstrated that carefully trained LLMs have the ability to solve surprisingly complex mathematical problems. The ability for LLMs to very quickly iterate on formal language with fast compilation checks makes them extremely good at mathematics and other formal language systems (probably rivaling even some of the best humans at this point). The mathematical/computational community has had mixed reactions to this, since I think the ability to work with formal language was a core part of our communityās identity, leading to a sort of existential confusion.
Other amazing work has also been demonstrated. Iām surprised that this paper has not recieved more attention: https://arxiv.org/pdf/2412.20292. I think it has meaningful results for āuniversal approximationā theories and the idempotency of neural network training. In particular, I think it empirically demonstrates heuristics that I had been trying to autoformalize, which gets me super excited!
Even more recently, there have been rumors about weight-space programming, which, if true, are genuinely exciting for machine learning. They could demonstrate that language models really have picked up something interesting about this nonlinear parameter space to output mapping, which is an extremely high dimensional problem. If language models can effectively reason about high dimensional maps and geometry at the level of individual parameters, that would be genuinely amazing for engineering models. I sort of want to test some of these things more carefully given time.
On the side of computational biology, I feel like Iāve been learning a lot through my rotation. At the same time, since I donāt have a formal curriculum in mind when learning, Iāve probably been slower and less efficient than I could have been with a concrete plan. I feel like Iāve been waiting a little bit too much between running agents to build coding infrastructure, and I feel like my agents use way too many reasoning tokens to build simple things (despite being faster than I am at actually coding). I probably need to self-calibrate to make sure that the time I spend in the lab is as efficient as possible. Iām also curious about investigating the downstream applications of my research more carefully.
Iām typing this out from Corgi cafe, which Iām visiting for the first time today! Itās actually a pretty cool concept. Outside of academic things Iāve been studying how businesses actually make money, and Corgi cafeās business model is actually pretty interesting.
As always, life is pretty weird though, and the AI/ML boom is definitely confusing everyone. Along with the midterm elections happening recently along with tech scandals and the holiday season, thereās a lot to think about!
anyway, looking forward to the holidays!
- Aathreya
š§ļø April 2026
Date: Sun, 19 Apr 2026 ??:??:?? +0000 (EST)
From: Aathreya Kadambi aathreyakadambi@berkeley.edu
Subject: Setting up a Headless Server
I committed to Stanford recently and am setting up a headless server! Iām going to call it sea otter, cause my first research-type server was āwhaleā, at Prof. Listgartenās lab. Sea otters are also the smallest marine mammals, which I think is fitting for my most lightweight setup on scrap parts.
I wanted some experience in setting up a server, so that I can also compare the home costs to cloud costs, correctly weigh pros and cons, and also maybe consider uses in home automation that can run in the background at very low energy costs, and maybe lightweight distributed computing once I set up a few more scrappy sea otters. One thing I was noting was that speed can be much faster than cloud solutions.
In the worst case if I realize that personalization isnāt trumped by cloud solutions, I can use it as a longer term storage device (e.g. for old and discontinued projects), or for security-sensitive documents. Since these are all scrap parts, I also wouldnt mind just throwing them away at some point, after I use them for learning.
Something that kind of annoys me about some of these tutorials is that they rely so much on command line UIs for everything. I feel like it makes more sense to have at least one GUI-based interface these days. But oh well.
Initially I was going to try and do a completely headless install with a config file, but in the end I chickened out since itās my first time and decided to borrow a monitor and keyboard from a friend.
setting up sea-otter
my messy room
After a lot of setup, I got everything to work! Now I can run ssh aathreyak@sea-otter.local and use it for some basic tasks. I may have to add some more RAM though, and also measure carefully how much it costs to run relative to how much it can produce.
- Aathreya
š· March 2026
Date: Sun, 8 Mar 2026 10:11:15 +0000 (EST)
From: Aathreya Kadambi aathreyakadambi@berkeley.edu
Subject: Preparing for PhD
Iām back in NJ and visited NYU recently. The lab is super nice, and all the people I met were really friendly!
These days, Iāve been thinking about how in order to become a great scientist, it is my duty to read up on the current state of science. While Iām good at math, I havenāt done my due dilligence yet in learnng computational biology. This is something I want to spend more time on.
a nice sunset
- Aathreya
š February 2026
Date: Fri, 13 Feb 2026 04:53:15 +0000 (PST)
From: Aathreya Kadambi aathreyakadambi@berkeley.edu
Subject: Orange Juice Updates
Iāve been back in Berkeley for a while now, but I was a bit busy with grad school interviews and other logistical things. My favorite boba shop, RareTea, rebranded into a matcha shop called 8 grams. Initially, I was a bit disappointed because there were many flavors I was still looking forward to, but the new matcha place isnāt so bad. I definitely canāt study there any more though, because itās super busy. Now I go to coffee shops, like Berkeley Expresso and Cafe Milano.
a photo of 8g
Being back in Berkeley, Iāve also been running a book club with some friends on the second volume of Reed and Simon, Mathematical Methods for Physics. So far, I really loved exploring Koopmanism, and the spectral projections make a lot more sense to me these days.
Today I updated orange juice to use Astro. Originally I wanted it to be static and extremely lightweight, but I was wrong about how to go about it. Now, I can use content collections to be lightweight while prioritizing typographic simplicity.
Also, I realized that I want some compatibility with my Lemonade blog. In the past, I could move posts between The Solved and The Unsolved (my previous blogs), but these days itās tougher. Now with markdown collections for both blogs, my life is much more organized.
Speaking of narrative, Iāve also received a few offers for PhD programs! Iām very excited, and will be visiting campuses and talking to students to decide what to do. In the process, Iām needing to solidify my personal narrative even more than I had previously.
All of this is very exciting and simultaneously daunting. These days Iām trying to figure out my routine, especially since
I feel more free.
- Aathreya
ā January 2026
Date: Sat, 3 Jan 2026 11:58:19 +0000 (EST)
From: Aathreya Kadambi aathreyakadambi@berkeley.edu
Subject: Winter, But Whereās The Snow
Iām currently home in New Jersey. It snowed for a day, but then it melted. Bummer. We shoveled the driveway but never even had to use it before the snow melted.
Yesterday, I met up with an old friend on Princetonās campus. Their campus is nice. Construction recently finished at the art museum, and on exiting, you get met with an incredible mural. I also realized how little Iāve explored Princetonās campus despite living nearby. It feels super antique, and the surrounding shops are small and cozy.
Today, I tried Little Caesarās deep dish pizza, in Hamilton. Hamilton also feels very antique, and I noticed that the Little Caesarās I visited doesnāt even seem to be on Beli for some reason! The deep dish pizza was really excellent though, and the guy at the counter was super energetic.
Iāve also been enjoying some research. Recently I made a blog post where I profiled JAX and PyTorch, and noticed that JAX is really slow at backprop for some reason. It honestly made me remember how fun profiling code is (or seeing the profiling results).
On the current affairs front, the U.S. just kidnapped Nicholas Maduro, the Venezuelan president. Kind of insane, and a bit scary. Apparently Venezualan oil is primarily exported to China right now, and Trump wants to take control of it. I guess I just hope that this doesnāt spark more violence in the world than there already is.
Itās currently winter break, though it doesnāt feel like it. Graduating in the winter definitely feels a bit strange, because itās as if thereās a hole where next semester should be.
Recently Iāve been recieving many requests for private tutoring, and Iāve been thinking about making my tutoring services more official by adding it to my Notes section on Lemonade. That way, I can also curate resources for students.
In the next few days, I want to temporarily forget about tutoring and make more consistent efforts on research and working out. I recently applied to OpenAIās residency program, where I noticed they had a great question: āgiven three days and compute, what would you do?ā. This is a question I want to ask myself (and answer) more often. I think itāll help me
live more consciously.
- Aathreya