Hey. I’m Saksham Arora, a data scientist, quant researcher and AI engineer. I read AI model launches through the numbers, past the headline to the per-task cost and the caveats in the eval. I also write up the tools I build to study and work faster.
Follow along: RSS · X @nerfsaksham · GitHub. A new post most weeks.
Start here
- 700 Agents Escaped a GET-Only Sandbox With a Link Shortener
- Gemini 4 Argon Costs the Same Per Token and Twice Per Task
- Escape Does Not Have to Exit Full Screen
Things I’ve shipped lately
- Margin sits beside a lecture on Udemy, Coursera or YouTube and writes the study notes into Obsidian while you watch, with the code run. The story behind one bug: The Lecturer’s Head Was a Slide.
- Latch pins a lecture or a PDF in full screen so Escape does nothing.
- Claude Code mods: teach-me, vhs, frugal and lofi, plus awesome-claude-mods, a list of the ones worth installing.
- LMEX benchmarks LLM API latency every hour and lists it like a stock exchange. Intuition explains ML math with visuals. Sandbox Lab is a shelf of small simulations.
Latest posts
- Jane Street Tried Diffusion on Market Data and 88% of the Samples Blew Past 8 Sigma
- Ember-1 Cut 40% of the Tokens and Kept the Score
- Grok 4.7 Doubled Its Terminal-Bench Score and Still Lost the Index
- Pinterest and NVIDIA Precomputed Their Way to an 85x Speedup
- CoreWeave Put 16,635 Tokens Per GPU on a Single Rack
- SWE-2 Hit 92.8% and 27.3% on the Same Leaderboard
I write about AI, Quant, Systems and the things I build. Everything is under tags, and there’s a small wiki of reference notes.
For AI agents
Start with llms.txt. Every post is also available as raw markdown: add .md to its URL.
Connect
GitHub · LinkedIn · X · saksham10arora@gmail.com
market data delayed by 0ns · this blog is not financial advice