Ali Tunçgenç's Blog

Posts with tagwriting

Looped Transformers, or Recurrent Depth Models

2026-10-17
7 min read1,387 wordsai safetyartificial intelligencecomputer sciencellmsmachine learningpersonalscalingwriting

How looped transformers reuse their layers to reason in latent space, why recurrent depth is suddenly showing up at the frontier, and what it could cost us in chain-of-thought monitorability.

Thank You for Your Hard Work

2026-05-25
9 min read1,723 wordspersonaltvwriting

I've watched When Life Gives You Tangerines on Netflix in the past month and it was very moving. As I complete this sentence, I detest the word "moving" in it. Moving how? Moving what? To better phrase it, watching this show is getting bombarded by flashbacks from your own life.…

I Watched Eric Jang Rebuild AlphaGo So You Don't Have To

2026-05-25
22 min read4,255 wordsartificial intelligencecomputer sciencepersonalwriting

In April, Eric Jang spent two weeks rebuilding AlphaGo Zero from scratch, and put the result online as autogo along with a long, interactive write-up. Last week he sat down with Dwarkesh Patel for a two-and-a-half hour blackboard lecture on what he learned. The thread that holds…

How Frontier LLMs are Trained and Served

2026-05-03
23 min read4,599 wordsartificial intelligencecomputer sciencepersonalwriting

This article is based on my handwritten notes from Reiner Pope's blackboard-style interview with Dwarkesh Patel, plus cross-checks against the published transcript and Dwarkesh's flashcards made in preparation for this episode of the podcast.

The AI Ouroboros

2026-02-25
5 min read885 wordsartificial intelligenceeconomicspersonalwriting

Or: The Market Discovers That Humans Were Load-Bearing