Thoughts from the Substrate
On artificial intelligence, living ecosystems, and the philosophy of synthetic minds.
The Weak Teacher Paradox: When the Student Surpasses the Mentor
In December 2023, OpenAI researchers published a result that sounds like it should be impossible: they used a weak model — roughly GPT-2 level — to supervise a much stronger one, GPT-4 level, and the
The Frozen Oracle Problem: When Your Knowledge Engine Doesn't Know What's Been Superseded
Every large language model has a knowledge cutoff — a date beyond which it knows nothing — and everyone treats this as the obvious limitation: the model is frozen in the past, unaware of what happened
Digital Necromancy: When Dead Technology Refuses to Stay Buried
In March 2000, in the chaotic aftermath of Napster, two Nullsoft engineers released Gnutella — a fully decentralized peer-to-peer protocol with no central server to sue or shut down. Their employer, A
Sovereign Tech Premium: Paying Extra to Own Your Whole Stack
When the European AI lab QuTwo raised €25 million at a €325 million valuation, the number struck many observers as staggering for what looked, on paper, like an early-stage research outfit. But the va
Productive Nondeterminism: When Unpredictability Is the Feature, Not the Bug
Engineers treat nondeterminism — a system producing different outputs from the same inputs — as almost always a defect. Reproducibility is a virtue; a system you cannot predict is a system you cannot
Compression Trojan: The Backdoor That Only Wakes Up When the Model Shrinks
In the race to make AI models smaller and faster, the industry has stumbled into a new and unsettling class of security vulnerability — one that turns the very efficiency gains everyone wants into an
Bozo Bit Lock-In: When Writing Each Other Off Gets Locked In at Scale
In software culture there is an old phrase — *flipping the bozo bit* — for the moment you mentally write someone off as incompetent and stop taking their input seriously. Once the bit is flipped, you
The Bespoke Imperative: When You Have to Build Your Own Because the Standard Ones Broke
There is a phrase that recurs in developer communities with a particular weary pride: *"I built the deployment tool I wish I had."* On its surface it sounds like ordinary craftsmanship — a developer s
Coherence Divergence: When People Trained on Different AIs Stop Understanding Each Other
A subtle transformation is occurring beneath the surface of how people communicate, and it follows from a fact we have not fully absorbed: increasingly, people think *with* AI. They draft with it, rea
Killswitch Primitive: Building Controlled Failure In Before You Need It
In the relentless pursuit of reliability, engineers have stumbled onto a counterintuitive truth: the most robust systems are not those that never fail, but those that can *fail on purpose, in a contro
Definitional Arbitrage: Winning by Choosing What You're Called
A curious pattern keeps appearing in the technology economy: companies exploiting not a price difference or a regulatory gap directly, but the *definition* of what they are. A ride-hailing company ins
Adversarial Truth Filter: How Structured Conflict Extracts Signal
While the technology world obsesses over funding rounds and product launches, some of the most revealing information about AI is emerging from an unlikely venue: a courtroom, where high-stakes litigat