Center for Applied Rationality (CFAR)

Theory of Change

CFAR has had three distinct theories of change:

2012-2020 (original): Train people in rationality techniques derived from LessWrong/Sequences. Some would recognize AI risk and work on safety. The explicit function was talent pipeline -- Open Philanthropy's grant descriptions state: "Our primary interest in these workshops is that we believe they introduce people to and/or strengthen their connections with the effective altruism community and way of thinking." Anna Salamon later admitted: "I was scared about AI risk, all the time... my notion of which participants were the coolest was mostly: who might do good work re: AI safety."

2018-2020 (pivot): Explicitly focused on AI safety recruiting, running AI-specific programs (MIRI Summer Fellows, AIRCS, CML). Impact was measured by alumni working at MIRI, DeepMind safety, CHAI, or EA orgs. Then CFAR went into hibernation from 2020-2024.

2025-present (relaunch as "aCFAR"): Salamon: "We are not backchained from 'help get the world into state X which'll be better for AI,' nor from 'help recruit people to AI safety work.'" The new model is a community of people geeking out about rationality questions, with no claimed long-term EA payoff. "If we can't make a financial go of things under our new ethos, my plan is not to revert to our past ethos, it's to fold."

What They Do

CFAR runs 4.5-day immersive workshops with ~25 participants. Curriculum includes Double Crux, TAPs (Trigger-Action Plans), Inner Simulator, Bug Lists, and new "honoring who-ness" content. They ran ~60 workshops from 2012-2020 with ~1,900 total guests, then went dormant for 5 years. Workshops resumed in 2025 with a new part-time instructor model.

Impact data (2017, most recent available): Of 894 alumni, 159 had a measurable "increase in expected impact" -- 19 at EA/AI orgs, 15 at MIRI, 28 on AI safety career paths, 32 through donations. AI-specific programs had ~3.3x the impact of mainline workshops, primarily through MIRI recruiting.

CFAR also fiscally sponsored LessWrong/Lightcone Infrastructure (2017-2024, now independent) and SPARC (ongoing youth program). SPARC has continued to receive Open Phil funding ($700K in 2022) even after CFAR lost general support.

Key People

Anna Salamon -- President and sole remaining active cofounder. Left philosophy PhD in 2008 for AI risk work. The organizational constant across all eras. Authored the most candid self-critique of any AI safety org leader, admitting CFAR's workshops were shaped more by AI risk fear than genuine rationality development. Told NBC News (Feb 2025): "in hindsight we were creating conditions for a cult."

All three other cofounders departed: Julia Galef (~2016, founded Update Project, wrote "The Scout Mindset"), Andrew Critch (cofounded BERI, SFF, now CEO of Encultured AI), Michael Smith (departed). Current team is ~7 part-time curriculum developers with no full-time staff. Duncan Sabien, former Curriculum Director and handbook author, publicly opposed the 2025 relaunch.

Money and Incentives

CRITICAL: CFAR's 990 data is misleading. The 990s show $9.1M revenue (2023) and $11.6M (2022), but this is almost entirely Lightcone Infrastructure pass-through money. CFAR's own operations were ~$1.5-2.5M/year pre-2020. Total liabilities of $23.6M (2023) reflect fiscal sponsorship obligations, not CFAR's debts.

CFAR's actual operating budget (2019): ~$1.5M total. 84% from grants (Open Phil, BERI, LTFF, SFF), 16% workshop revenue.

Current state (Dec 2025): $129K available, ~4 months runway. Seeking $125K-$200K to survive through 2026. Annual costs ~$300K+ (admin $72K, venue net ~$50K, workshops net loss ~$130K, curriculum $25K). Workshops lose money -- November 2025 workshop was -$28,400 net.

Major funding sources (historical):

Coefficient Giving/Open Phil: $4.3M total (2016-2022). General support ended with 2020 "exit grant." Last grant was for SPARC only.
SFF (Jaan Tallinn, cofounded by Andrew Critch): ~$2.2M directly to CFAR (2019-2024)
BERI (cofounded by Andrew Critch): ~$1.1M (2018)
FTX: ~$4.9M (2022, for Lightcone -- now subject to clawback lawsuit)
Workshop revenue: Always a small fraction (~$25K-$160K/year for CFAR's own programs)

Key financial dynamics:

Open Phil's decision to exit in 2020 is a strong negative signal from a sophisticated evaluator
CFAR is deliberately rejecting its old EA-donor funding model without a proven replacement
The FTX clawback lawsuit ($4.9M) is existentially threatening for an org with $129K in cash
The Bodega Bay venue is both CFAR's main asset and a significant ongoing cost (~$285K/year, offset by ~$200K rental income)
Andrew Critch cofounded both BERI and SFF, two of CFAR's major funders, creating a tight funding circle

What Others Say

Strongest critical argument: CFAR's theory of change was always circular -- it attracted people already interested in rationality/EA, trained them in techniques that were never rigorously validated, claimed credit when they went on to do AI safety work they were already inclined toward, and created a psychologically intense community that enabled several serious harms (Brent Dill's abuse, Zizian radicalization, mental health crises). Anna Salamon herself provides the strongest version of this critique: CFAR optimized for "does this cause people to look like people who will help with AI risk" rather than "are they making real research progress now."

Jessica Taylor (former MIRI researcher, 2021): Described psychotic breaks, at least 4 psychiatric hospitalizations in the MIRI/CFAR social circle, 2 suicides (Maia Pasek and Jay Winterford), and called CFAR "pretty corrupt." She wrote: "Most of what was considered bad about Leverage Research also happened around MIRI/CFAR."

Asterisk Magazine (2025): "The Sequences create the raw material for a cult." CFAR "occasionally blurred the line between rationality curriculum development, people management, and therapy." A Harvard chaplain asked: "How much more cult-like does it have to get?"

Zvi Mowshowitz (2017, pre-crisis): "CFAR does excellent and important work." Donated $4,000. But presciently warned about Goodhart's Law: measuring impact through EA/AI outcomes could distort workshops toward persuasion over genuine rationality development.

Oliver Habryka (Lightcone, formerly under CFAR, 2023): Expressed doubts about whether the entire EA/rationality ecosystem was net positive, questioning whether it had accelerated AI timelines through founding DeepMind, OpenAI, and Anthropic.

Rolling Stone, NBC, Guardian: Extensive investigative journalism covering the Zizian deaths (6 linked), FTX financial connections, and Manifest conference controversies. These have permanently shaped public perception of the rationalist community.

What's Absent

No rigorous external evaluation of workshop effectiveness after 13 years (no RCTs, no control groups, only self-report data from 2015/2017)
No public accounting of CFAR operations budget separate from Lightcone pass-through for 2021-2023
No published safety policies for the 2025 relaunch despite the Brent Dill affair and community mental health history
No systematic tracking of negative outcomes (mental health crises, disillusioned alumni)
No response to the FTX clawback lawsuit publicly
No alumni tracking data since 2017 (8 years, ~1,000 additional alumni)
No comparison with modern field-building alternatives (MATS, BlueDot, ARENA)