Book Overflow Carl Brown Discusses All Things AI
Mar 5, 2026
Carl Brown, YouTuber and software engineer who offers skeptical, evidence-first takes on AI and software trends. He discusses trust and bootstrapping for toolchains, practical defenses against hidden or compromised code, limits of current models for reliable complex coding, risks from synthetic training data and agent prompt injection, and how engineers can realistically adapt and use AI for prototyping.
AI Snips
Chapters
Books
Transcript
Episode notes
You Can't Blindly Trust Code You Didn't Write
- Ken Thompson's Reflections on Trusting Trust warns you cannot fully trust code you didn't write yourself.
- Carl Brown relates this to modern AI by highlighting hidden, self-modifying behaviors in binaries and compilers as long-standing security risks.
Cuckoo's Egg Shows Real Forensics Steps Missing From Fiction
- Carl Brown recounts reading The Cuckoo's Egg and contrasts practical incident-response best practices missing from the Coding Machines story.
- He names checksums, cross-compilers, switching OSes, and tracing origins as concrete steps the characters did not take.
Skepticism Should Follow Evidence Not Hype
- Carl frames his AI skepticism around scientific method and evidence rather than ideology.
- He tracks AI improvements empirically (tests, regressions) and asks what the true limiting resource is for continuing progress.
