Skip to content
Apart Research
Apart Research
Contact us
Sprint projectFeb 2, 2026San Diego, CA

The Half-Life of Compute Thresholds

Aaron Potter

Submitted to The Technical AI Governance Challenge. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: The Half-Life of Compute Thresholds

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

ompute thresholds are widely discussed as a practical governance tool, but the literature rarely specifies how quickly fixed thresholds become outdated as algorithmic efficiency improves. We develop a compute-threshold staleness model that distinguishes raw training compute from baseline-equivalent (“effective”) compute and defines staleness as the ratio between a policy threshold and the time-varying threshold that would remain aligned with a fixed risk cutoff

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. Not clear on counterfactual impact of this work; doubling time only, without these formulae, already provides "a simple model which can provide a good intuitive grasp of the problem for policy-makers." In practice, this model may be making the the explanation more opaque to policymakers.

    There could be some value in describing staleness in the intervals between doubling times; e.g how stale is the threshold at 7 months with an 8 month doubling time; but this still can be reasoned about pretty intuitively with knowledge of the doubling time, and a simple, one-off calculation.

    I would want to see more exploration or elaboration on specific use-cases for this work to better understand its impact potential. Maybe dealing with uncertainty around doubling time, or enabling very precise update thresholds under certain conditions.

  2. The model's most significant limitation is the assumption that τ is constant across domains and over time. Algorithmic efficiency likely improves at different rates for different capability types (language understanding vs. code generation vs. chemistry), which means a single threshold update cadence may be insufficient — the paper should explore domain-specific τ values and their policy implications. The backtest, while a nice touch, confirms consistency with Ho et al.'s estimates rather than providing independent validation, since the model's dynamics depend on those same estimates.

Cite this project

@misc{potter2026halflife,
  title = {{The Half-Life of Compute Thresholds}},
  author = {Aaron Potter},
  year = {2026},
  month = feb,
  note = {Submitted to The Technical AI Governance Challenge, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-halflife-of-compute-thresholds-260u}},
  url = {https://apartresearch.com/sprints/projects/the-halflife-of-compute-thresholds-260u}
}

More from The Technical AI Governance Challenge

All 47 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026