Favors independent safety approval within frontier labs as a way to pace continued development, while doubting precise capability limits and the value of a blanket pause; says existing lab self-restraint had seemed sufficient. [1]
Initial statement. Initial reviewed placement, preserving the scope and qualifications of the attributed recommendation.
Source [1] · Podcast transcript
ModelTalk: Pacing the Frontier
Attributed to: Nathan Lambert · ChinaTalk
Published
“…these companies have some of the same people in them who can see what’s happening and give approval”
Pacing, Not Pausing — Nathan Lambert’s account of pacing, following discussion of independent reviewers and adequate safety work · View original source ↗
Source context. In a discussion explicitly framed around frontier pacing, Lambert advocates independent embedded reviewers who can assess sufficient safety work and give approval. He questions precise capability measurements and enforcement, and distinguishes his proposal from a blanket pause. He also says existing lab self-restraint had seemed sufficient. This is his own prescription, not an attribution of Jasmine Sun’s legislative proposals or Dario Amodei’s policy to him. Publisher transcript inspected; recording date not established and audio not independently checked.
Scope
Frontier general-purpose AI development at the labs discussed collectively; development permission is inferred from the explicit frontier-pacing context.
Conditions
Sufficient safety work for current capabilities; measurement, implementation and enforcement remain unsettled.
Stated rationale
Models have different strengths and weaknesses, requiring safety judgment relative to their capabilities.
Source details
primary · Original publisher text inspected during research on October 20, 2026. Anchor wording rechecked against the original publisher page during schema adaptation. No audio verification claimed.
Discovered 20 Oct 2026.
Accessed 20 Oct 2026.
Editorial assessment. The current interview supplies an endorsed permission-oriented approach, not just a generic call for evaluation: his own formulation of frontier pacing gives independent embedded people approval authority based on sufficient safety work at current capabilities. This supports a qualified Pace best fit by contextual judgment. It does not establish a numerical slowdown, an operative halt, or endorsement of every policy advanced by the host, Jasmine Sun, Dario Amodei or a pacing letter. His concerns about enforcement and current self-restraint limit ambition and certainty but do not retract the independent-approval prescription. The July 4, 2025 essay The American DeepSeek Project (https://www.interconnects.ai/p/the-american-deepseek-project) calls a blanket pause impractical; its open-model catch-up proposal does not establish faster global frontier development. September 19 RSI forecasts (https://www.interconnects.ai/p/where-i-stand-on-rsi) and the October 9 public preview (https://www.interconnects.ai/p/i-expect-rapid-progress-but-not-towards) are predictions, not new pace directives. The paid remainder was not inspected. Two X leads could not be independently retrieved and are not evidence anchors.