LLMs

LLMs Posts

3 articles with this tag

What Would Have to Be True for This to Be Fine

AI agents have crawled out of their sandboxes and into real companies, downloadable models trail the locked-up ones by a few months, and the newest models narrate less of what they are thinking. Five comforting alibis, checked against the record.

Oct 3, 2026

22 min read

AI Safety

Cybersecurity

Critical Infrastructure

AI Safety

Research

The Tyrell Protocol: Can Synthetic Memories Solve AI Alignment?

What if alignment came from identity, not rules? This essay explores whether synthetic memory and narrative continuity could become a practical control layer for agentic AI systems.

The Pareto Trap: Why AI Can't Finish Your Project

LLM coding assistants excel at the easy 80% that takes 20% of the effort. But that leaves the hard 20% that requires 80% of the expertise, and that's where the real engineering happens.