Max Ghenis

Democratizing evidence-based policy @ThePolicyEngine. Views my own.

[Max Ghenis @MaxGhenis]

Correction: a reader caught the model pricing non-US health gifts at US rates — bed nets at Medicaid prices. Health dollars now route by delivery geography. New numbers: ~205,000 QALYs skeptical (~420,000 credulous), blended $148k/QALY, frontier ~500×. Most of the health comes from the ~5% delivered abroad.

[Max Ghenis @MaxGhenis]

.@mackenziescott has given away $26 billion. What did it buy in health?

I built an interactive model from her own gift database: ~70,000 QALYs weighting each study by causal credibility, ~200,000 taking every effect at face value.

[Max Ghenis @MaxGhenis]

Just posted to the EA Forum: https://t.co/Wm5h8N7E6A

Another benchmark: Scott's donations would have generated 50-100x the QALYs if provided as cash transfers to the global poor via @GiveDirectly.

[Max Ghenis @MaxGhenis]

Update: a reader caught the model pricing her non-US health gifts at US rates — bed nets at Medicaid prices. Routing health dollars by delivery geography tripled the estimate: ~205,000 QALYs skeptical, ~420,000 credulous, $148k/QALY, frontier ~500×. Now fixed everywhere.

[Max Ghenis @MaxGhenis]

.@mackenziescott has given away $26 billion. What did it buy in health?

I built an interactive model from her own gift database: ~70,000 QALYs weighting each study by causal credibility, ~200,000 taking every effect at face value.

[Max Ghenis @MaxGhenis]

@KuviacM Which do you favor?

[Max Ghenis @MaxGhenis]

@kjonasj @mackenziescott This? Loading for me. https://t.co/erVwNJmziM

[Max Ghenis @MaxGhenis]

Oh Fable, I'll miss you on July 20 (😏)

[Claude @claudeai]

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

[Max Ghenis @MaxGhenis]

https://t.co/5UneKZJNQX

[Max Ghenis @MaxGhenis]

I just became the first Straude user to report 1 billion output tokens.

(The dollar column is API-list-equivalent — I'm on subscriptions.)

Most of them went into things we start shipping this week.

[Max Ghenis @MaxGhenis]

@essemmeppi Thanks! Yes and we're looking for other things to eval, any requests?

[Max Ghenis @MaxGhenis]

It's a good model, sir.

[PolicyEngine US @PolicyEngineUS]

GPT-5.6 joins PolicyBench, our benchmark of how accurately AI computes US taxes and benefits (no tools, scored against PolicyEngine):

Sol #1 of 23 — 88.7% within $1 (prior best 83.5%)
Luna #2 — 84.5%
Terra #4 — 83.4%

https://t.co/e7idbTLBgr

[Max Ghenis @MaxGhenis]

Asking Fable to route cyber tasks to Sol subagents

[PolicyEngine US @PolicyEngineUS]

PolicyBench now scores 20 AI models on how accurately they compute US taxes and benefits — adding Claude Fable 5, Claude Sonnet 5, and five open-weight models: DeepSeek, Qwen, GLM, MiniMax, and Kimi.

GPT-5.5 leads, Fable ranks #2, and DeepSeek v4-pro tops the open-weight models.