A chain-of-thought trace is inherently codified, so it cannot form a tacit moat the way embodied know-how can
Two competitive advantages in the vault sit at opposite ends of one axis — codification. The ASML/Zeiss EUV mirror advantage is attributed to tacit, embodied institutional know-how that resists being written down, which is why it is durable and cannot be copied even absent patent protection (claim-euv-mirror-advantage-is-tacit-know-how-not-patent). A reasoning model's advantage is the structural inverse: a chain-of-thought trace, once produced, is its own blueprint — it is literally text, hence already fully codified. Following Polanyi's distinction between tacit and codified knowledge (see 2026-07-11-hop-polanyi-codification-bridge), codified knowledge copies cheaply while tacit knowledge resists imitation. The reasoning trace therefore falls on the copyable side by construction.
The empirical leg is claim-s1-distilled-reasoning-from-1000-traces-in-26-minutes: once traces are visible, much of the behaviour transfers from ~1,000 examples in a short fine-tuning run. Against this, hiding the raw trace (claim-openai-hid-o1-raw-chain-of-thought-partly-for-competitive-advantage) reads as an attempt to manufacture tacit-style secrecy around inherently explicit knowledge — the moat is not the reasoning itself but whatever was never said out loud. The same codification hinge appears in AI governance from the other direction: because a load-bearing expert verdict cannot be fully codified, it cannot be automated, so CANONIC falls back to a codifiable accountability ledger rather than a truth-gate (claim-canonic-deliverable-is-an-append-only-evidence-ledger).
A caveat on the "codified" claim: that a trace is readable text does not by
itself prove the advantage is fully captured in it — the trace may under-specify
the data curation, verification, or base-model work behind it, and whether the
answer-only output still leaks enough to distill remains open. The tacit pole of
the asymmetry also inherits the unverified status of the EUV note it rests on, so
this synthesis is held at seedling.
Source
“requiring just 26 minutes of training on 16 H100 GPUs”
claude-opus-4-8 · Promotion from 10-inbox/raw/2026-07-11-hop-cot-not-a-tacit-moat.md, 2026-07-12 · raw markdown