Agenda

Five Things That Actually Mattered in Tech This Week

AI alignment building censorship infrastructure, Saudi Arabia rerouting energy logistics, and a dead-simple question nobody is asking about LLMs in science.

Speed:

Dario Keskin

Aug 14 · 3 min read

Key takeaways

AI alignment methods are, architecturally, also censorship infrastructure — whoever controls the classifier controls the output, full stop.
LLMs deployed as co-scientists carry unresolved liability questions that Gulf research funders should resolve before procurement, not after.
Saudi Arabia's Red Sea rerouting is an engineering story and a startup opportunity; every disrupted shipping corridor accelerates logistics-tech investment.
The UAE-Iran situation this week illustrates that GCC geopolitical complexity is not background risk — it is the operating environment for every tech company in the region.
If reasoning in AI is a learnable rule-based process, structured reasoning companies have a real technical moat — provided they can outlast the next foundation-model price compression.

It is Friday. Here is what deserved your attention and probably did not get it.

— 1. THE ALIGNMENT PEOPLE ARE BUILDING THE CENSOR'S TOOLKIT —

A position paper dropped on arXiv this week with a thesis that should make anyone in AI policy uncomfortable: the methods researchers are using to align large language models — to keep them from saying harmful things — are, structurally, a censorship architecture. The argument is not that alignment is bad. It is that the tooling being built to enforce al…

#Geopolitics○ ...