Neel Somani's Blog

The Endgame for Mechanistic Interpretability

The endgame for mechanistic interpretability is formal methods.

Jan 5 • Neel Somani

November 2025

Intro to Routing: Mixture-of-Experts and Expert Choice

I derive MoE and EC from first-principles.

Nov 14, 2025 • Neel Somani

1

October 2025

A Minimal Route to Transformer Attention

Is attention inevitable?

Oct 30, 2025 • Neel Somani

Killing the GIL: How To Use Python 3.14's Free-Threading Upgrade

The global interpreter lock (GIL) has been interfering with true parallelism in Python. That ends with Python 3.14.

Oct 14, 2025 • Neel Somani

What You Didn't Learn in Berkeley CS 188 — Part 4

Is GRPO broken?

Oct 11, 2025 • Neel Somani

What You Didn't Learn in Berkeley CS 188 — Part 3

Off-policy methods, for better sample efficiency and scalability.

Oct 9, 2025 • Neel Somani

What You Didn’t Learn in Berkeley CS 188 — Part 2

Implementing the policy gradient methods: REINFORCE, A2C, TRPO, PPO.

Oct 7, 2025 • Neel Somani

What You Didn’t Learn in Berkeley CS 188 — Part 1

Why isn’t there model-free policy iteration?

Oct 4, 2025 • Neel Somani

July 2025

A Free Market for Eyeballs

Many takes on the attention economy, with very few attention economists.

Jul 29, 2025 • Neel Somani

February 2025

The BLAST Playbook

I argue that the $500B-1T in annual software investments should reallocate to BLAST: assets that monetize boredom, loneliness, and scarcity.

Feb 24, 2025 • Neel Somani

January 2025

Privatize the FDA

I advocate for privatizing the FDA by replacing its drug approval monopoly with a competitive market to reduce delays, lower costs, and foster…

Jan 11, 2025 • Neel Somani

October 2024

The New Economy

We have massively overallocated our youth to roles like software engineering and medicine - which will be shortly replaced en masse by AI. I propose how…

Oct 25, 2024 • Neel Somani