Dario Amodei published “We Must Pace the Frontier” today, committing Anthropic to embedded third-party evaluators and asking governments to require every other frontier lab to match. Sam Altman reportedly agreed within hours. Jake Gold’s reply accepts the stated goal — slowing AI progress — and argues the law Anthropic is asking for will never deliver it.

Gold’s case against that kind of regulation:

  • Rules like embedded evaluators, compute thresholds, and industry coordination with antitrust waivers get written with help from the current frontier labs, because nobody outside them understands the technical details well enough to draft them.
  • Once on the books, rules only accumulate. Every incident adds one; none are ever removed.
  • Big labs can afford the compliance teams and lawyers. Each new requirement raises the cost of catching up, so the regulation ends up protecting the incumbent’s position and profits instead of restraining it.

His alternative is a single rule with no moving parts: any model a company offers to the public has to be released as open weights — the model’s actual numerical parameters published so anyone can download and run it, rather than reached only through the vendor’s API.

  • It would not cover every model a lab trains, only the ones it makes publicly available. Internal and research models would be unaffected, and governments could still get access to unreleased models.
  • The mechanism is financial rather than regulatory. Frontier development is funded by valuations that assume the weights stay proprietary. Remove that assumption and less money flows to future training runs, slowing every lab at once “without a regulator deciding anything.”
  • Openness is directly observable, which is his answer to the usual objection that any rule can be dodged.

He asks Amodei specifically because of a record of taking costly positions: leaving OpenAI in 2020 when the safe career move was to stay, backing export controls that shrank Anthropic’s own market, paying $1.5 billion to settle with authors rather than fight to the end, and running as a Public Benefit Corporation — a corporate form that legally permits putting the mission ahead of short-term profit.

What the thread adds

The 86-comment thread on Hacker News is largely hostile, and to the mechanism rather than only the conclusion. It is also where the author answers his critics directly.

  • jacobgold — Gold replies in the thread to both standard objections. On the claim that a lab would simply stop selling publicly, he answers that enterprises count as the public: “Enterprises are part of the public in this context, so that’s not a loophole which would exist.” On the worry that frontier models would then never be released at all, he concedes the shape: “That’s a legitimate possibility but it’s also much slower, more risky, and harder to raise money for than what they’re doing now.”
  • Aurornis — the sharpest attack on the logic. The argument requires “every lab in the world agreeing to self-destruct their valuations and stop competing,” which “doesn’t even follow,” and it ignores new labs starting up from the open weights: “a much more impossible ask than anything Dario is asking for.”
  • floatrock — the best steelman, and it argues the letter is more coherent than it looks. A normal regulator loses to relocation (“your headquarters has just moved the Cayman Islands, Ireland, or Switzerland”), so the only enforceable lever is the return on capital: “If the only winning move is not to play, how do you ensure everyone stops playing? Simple: you must destroy the game.”
  • zozbot234 — independent corroboration of the mechanism, from an unexpected source: the “open weights are decelerationist” argument “is straight from Dean Ball on Twitter,” which they note cuts differently depending on whether you fear a frontier arms race or doubt that AGI is close.
  • sgammon — the practitioner’s cost check on the promise that open weights help ordinary people: it “doesn’t really help ’the little guy’ when it costs ~$50K per H200 and 8 of them aren’t enough to run a single node.” nijave accepts the hardware numbers but argues hosting providers still compete on price and get pushed to serve models more efficiently.
  • notcodingtoday vs Marha01 — the safety objection from the other side, read as a marketing piece for the author’s company: jailbroken frontier models mean “constant hacking leading to public desertification of the web (best case).” Marha01 offers the counter-theory that a model able to find a vulnerability can patch it, and there are finitely many, so the status quo repairs itself until a stronger model arrives.

The cluster, and the question the thread keeps asking

Four commenters reach the same destination in four different wordings — a mandate to open the weights produces labs that simply stop releasing models publicly, concentrating capability instead of distributing it. tristanj notes Anthropic “gets most of its revenue from enterprise agreements anyway”; jsnell spells out the endgame, with public models stagnating while internal ones “eventually eat the whole economy”; wyager compresses it to “no models released to the public, complete concentration of power”; jrflo says there would be no incentive to develop a new non-distilled model at all. That is one objection with four authors, and it is the strongest thing in the thread.

The other repeated ask is about enforcement, which the letter never addresses. aarondong reconstructs the whole chain of reasoning sympathetically and then notes it “hinges on strict enforcement of this policy, when AI model routing can already be opaque.” baby asks why the harder options — banning publication of strong open-weight models internationally, or nationalizing the labs — are not discussed at all: “if we are being serious why are we not discussing these options?” Nobody in the thread answers for the author, and the letter does not either.

That gap is the honest shape of the disagreement. Gold’s proposal is a designed sacrifice that depends on universal participation — either through law or through everyone agreeing the money should dry up — and the thread’s reply is that the labs’ alternative sacrifice, quietly not releasing anything, is cheaper and legal today.

A note on reading comments as evidence: HN handles are pseudonymous and the site publishes no per-comment scores, so ordering reflects HN’s own ranking rather than a vote. This is a slice of the thread, not a consensus. Reported claims — Altman’s agreement, the settlement figure, the Anthropic models named — come from the letter’s own links and are quoted, not independently verified.