ii. writing
Writing.
Long-form on multilingual AI, alignment, language-learning, and the strange shape products take when they grow.
Published here first. The Loop on LinkedIn mirrors it.
2026
Jul 2026 Jun 2026 Jun 2026 Feb 2026 Feb 2026 Jan 2026
LLM Grooming: Aiming Propaganda at the Machine
Russia is flooding the open web with propaganda engineered to be scraped into AI training data. The tactic is real and inexpensive; whether it actually changes what the models tell you is a harder and still unsettled question.
Who Decides What AI Calls True
The fear is about who can switch the models off. The real power is quieter, and harder to take back: who decides which version of the truth your AI serves you first.
I asked nine AIs to map Middle Eastern politics
Western left-right doesn't fit the region. So I built Tayyar — a panel of language models that scores MENA parties on sixteen axes — then used it to dig into Israel's fault lines and a live Knesset coalition simulator. The data says the Middle East barely splits on the economy.
How AI killed the soul of the world's favorite language-learning app
A look at Duolingo's user-(retention)-centered approach.
Why AI thinks in English and forgets in Arabic
An audit of the architectural friction and reasoning bottlenecks across Semitic languages.
The sanitization of intelligence: how the alignment tax is killing the LLM frontier
On the alignment tax, corporate risk-aversion, and what "safe" AI gives up on its way to being safe.
2025
The Loop goes out roughly monthly. Subscribe via RSS for the canonical version, or follow on LinkedIn for the mirror.