Aman Gupta

Two papers accepted at ICML 2025! 1. AlphaPO -- Reward shape matters for LLM alignment - arxiv.org/abs/2501.03884 More details on the 2nd paper will be released soon! Excited to visit Vancouver for the conference!

People don't do business with the smartest person in the room. They do business with the person they like, trust, and respect.

We are in the singularity

moreover the anticipated path to AI was “geniuses figure out the deep structure of cognition and intelligence” and the actual historical path was “lol throw the internet into a giant blender and make gradient descent go brrr once moore’s law gives us enough compute”

The KLS constant is O(log^1/4 n)! arxiv.org/pdf/2607.24164. Speaking informally, the KLS conjecture says that the best way to cut a convex shape into two equal volume halves so as to minimize the newly exposed surface area is more or less just a straight cut. The KLS constant measures the gap between the best arbitrary cut and the best straight line cut. Kannan, Lovász, and Simonovits conjectured that this term is O(1). About two weeks ago I had a breakthrough on some adjacently related work, and I had noticed that it would be quite fruitful to study the KLS conjecture through something in convex geometry literature known as a moment measure after I had discovered an interesting formula. This moment measure satisfies a second order PDE known as a Monge–Ampère equation. I used ChatGPT 5.6 Pro to differentiate this equation, and then after a lot of experimentation, ChatGPT 5.6 Pro produced a striking result, which after a some extra work gave what is now Theorem 2.5 in the paper. At the time of uploading this paper, it seems that independently Yuansi Chen and Boaz Klartag had used ChatGPT 5.6 Pro to produce a slightly weaker version of Theorem 2.5, which still is of course the main breakthrough, see here: weizmann.ac.il/math/klartag/s…. Quite remarkable!

I already feel like 50% papers on arXiv use LLMs heavily for finding original arguments and examples. 90% probably use AI for literature check and quick review. Science is undergoing the same process coding went through last year.

Asked a nerd friend why he suddenly looked upset and he said, choking back tears, “we’ll never know what Von Neumann could have done with this level of compute”

I have officially reached the age where I am bothered by lights being on, doors left opened, loud noises & people

My friend Zhengqing used GPT-5.6 to solve a beautiful well-known conjecture in probability. Feige's 1/e conjecture: for independent nonnegative X_1,…,X_n with E[X_i] ≤ 1 and S_n = X_1+⋯+X_n, we have P(S_n ≤ E[S_n] + 1) ≥ 1/e. Exciting to see a problem we used to kick around over lunch and dinner get solved! 1/2

“To a disciple who was forever complaining about others, the Master said, ‘If it is peace you want, seek to change yourself, not other people. It is easier to protect your feet with slippers than to carpet the whole of the earth.’”​ — Anthony de Mello

No good deed goes unpunished

Tim Gowers on a possible AI-dominated future of mathematics: ... because of [AI] people are no longer motivated to put in the years of effort it takes to reach the level of expertise that a typical research mathematician has now. After a decade or two, we might arrive at a situation where the mathematical literature has, in some form, been vastly expanded, but there is no corresponding community of human experts who have a shared understanding of parts of it. gowers.wordpress.com/2026/07/26/tho…

I don’t think you understand what ChatGPT Voice unlocks I went on a 4 hour hike yesterday. Through the northern California redwoods. Got more work done in those 4 hours talking to Voice through my AirPods then I do in 8 hours at my desk Work can be done ANYWHERE now 45 minute drive back using self driving? Now 45 minutes I can talk to Voice and work on side projects 20 minutes in the cafe? I can sit and talk to Voice and when I get home I have drafts on my computer for my newsletter for the next month Being able to just use your voice to talk to a super intelligence that controls your computer is unlike anything we’ve ever experienced before I don’t even care that the voice sounds human like, or you can interrupt it, or any of the wild advancements they made to the tech It’s the fact that it can control my computer, meaning all I need to get incredible work done is talk through my AirPods anywhere in the world that is the major game changer for me I don’t think people truly understand the implications of this yet Why even have a desk or monitor anymore. Work can be done ANYWHERE now The key is your set up: 1. Choose one device (preferably an always on desktop) as your main “headquarters” device all your work gets done on 2. Get the ChatGPT app on all your other devices (iPhone, iPad, laptops, Mac Minis) 3. Set up ‘connections’ in your settings so that all your ‘node’ devices can control your ‘headquarters’ device 4. Go legit anywhere in the world. Turn on Voice. Ask it to brief you on your projects, give you a recommended next step, then spin up new threads to do work I really don’t think work looks the same moving forward.

What a fantastic read.

@pulkitbe love it !

@aman2304 "if equal love cannot be, then let the more loving one be me" :-)

"But the glory of life is to love, not to be loved; to give, not to get; to serve, not to be served. It may not be our fault that we cannot respond to the offer of friendship or love, but it is our misfortune. The secret is revealed to the other, and hid from us. The gain is to the other, and the loss is to us. The miracle is the love, and to the lover comes the wonder of it, and the joy."

"You will be forgotten either way. So live in such a manner that, while you remain, you belong to yourself." — Marcus Aurelius

Airports should have a fast lane for people who know how to take their laptops out.

crazy how 45 min of cardio and six hours of sleep is the difference between feeling like projects are going nowhere and being deeply grateful for the opportunities before you

OpenAI Names BNY, Nubank CEOs to Board Ahead of IPO bloomberg.com/news/articles/…