Programme Thesis
The Long-Term Future Fund (LTFF) is EA Funds' vehicle for reducing global catastrophic risks, with a pronounced emphasis on technical AI safety research and field-building. It exists because potential risks from advanced AI and engineered pandemics could curtail or eliminate humanity's long-run future, and the Fund bets that small grants to high-leverage individuals and early-stage projects now can generate outsized expected-value returns. The fund explicitly weights 'expected benefit × probability of success' over certainty, making it unusually friendly to speculative but rigorous work.
Selection Criteria
• Mission fit: Direct contribution to reducing existential/global catastrophic risks — AI safety (technical or governance), biosecurity, longtermist field-building. AI safety dominates recent payout rounds.
• Expected value: Fund managers explicitly use EV framing (1% chance of $100K = 50% chance of $2K). High-risk/high-payoff proposals are acceptable.
• Counterfactual funding gap: Would this work happen without LTFF? Ungated independent researchers with no institutional salary score well here.
• Theory of change: Applicants must articulate a credible causal chain from the specific work to long-run impact — vague 'this is important science' framing is penalised.
• Track record / personal fit: Demonstrated ability to execute (publications, preprints, shipped tools). For individuals, evidence of self-direction and output under constraint matters more than credentials.
• Credibility / endorsements: Letters or documented collaboration from recognised researchers in the relevant field are weighted heavily.
• Budget realism: Requested amount must match the work scope — the fund routinely gives $5K–$120K to individuals; requests far outside that range without strong justification raise flags.
• Eligibility: No institutional requirement; independent researchers, students, and early-career individuals are routinely funded. No geographical restriction. Rolling intake — no fixed deadline.
Past Winners / Cohort Profiles
The prototypical LTFF grantee is a self-directed early-career researcher or practitioner doing AI safety work outside a major lab or university, often extending or pivoting from an existing program (e.g., post-MATS, post-ARENA). Featured grants include: Logan Smith ($40K, 2024 Q3) for LM tools aiding alignment research; Robert Miles ($121K, 2023 Q3) for AI safety video/podcast outreach; Jeffrey Ladish ($98K, 2023 Q1) to spin up a cybersecurity-alignment risk org; Sage Bergerson ($2.5K, 2022 Q4) for a compute-access policy paper. Nicky Case ($80K, 2025 Q1) received a year of stipend for accessible AI alignment explainers. Individual stipend grants dominate: of 682 LTFF grants on record, ~303 are individual stipend/salary grants ranging from $500–$400K (median ~$30K). Neuroscience-adjacent work is rare but not absent: a $120K Geneva Centre fellowship funded three fellows bridging synthetic biology, AI, and neurotechnology (2024 Q2). Bio/pandemic risk grants exist but are a small minority of the portfolio.
Ideal Candidate Fingerprint
The platonic LTFF applicant is a technically rigorous independent researcher who has already produced credible work output (preprints, benchmarks, tools), can draw a tight causal chain from their specific research to reduced AI existential risk, and faces a genuine funding gap because they sit outside a well-resourced lab or university. They are self-motivated, have endorsements from established names in the field, and request a modest stipend to extend existing work rather than seed an entirely new direction. Fund managers strongly prefer people who are 'already doing the work' and need runway, not people proposing to start from scratch.
Recommended Framing
Eniola should open with the AI alignment hook explicitly and early: reward-circuit consolidation dynamics — the mechanistic heart of CCT — is a biological substrate for value encoding, making the CCT framework directly foundational to inner alignment and value-learning theory in RL systems. This is not a rhetorical stretch: Gershman (Harvard, arXiv endorser) sits at the intersection of computational neuroscience and RL theory, and Daw (Princeton) and Mattar (NYU) study precisely how humans and animals encode reward into decision policies, which are the biological analogues of the reward models alignment researchers worry about. The pitch should be: 'CCT provides mechanistic ground truth for how reward signals become stably encoded preference structures — the exact failure mode that inner alignment theory needs empirical constraints on.' Anchor credibility on three sole-authored preprints with confirmed H1–H5, ODE/RK45 + Bayesian MCMC validation, and four top-tier endorsers — positioning this as rigorous independent computational science that LTFF has historically funded (see Logan Smith, Sage Bergerson, Nicky Case archetypes). The Theory of Change should explicitly step through: CCT model → mechanistic understanding of reward consolidation → empirical constraints on RL alignment theory → reduced risk of misaligned value learning in advanced AI.
Watch Out
1. AI alignment fit is real but requires explicit articulation: LTFF currently directs ~85% of grants to technical AI safety (interpretability, alignment, governance). A pharmacology/neuroscience framing without a clear AI safety causal chain will be screened out by fund managers looking for direct GCR relevance — the connection must be in the first paragraph, not buried.
2. No bioRxiv preprint yet: the classifier notes to 'apply once the first CCT paper is live as a preprint on bioRxiv' — posting only to OSF/Zenodo is weaker; LTFF fund managers are more likely to weight a bioRxiv preprint as credible peer-reviewed pipeline. Resolve before submitting.
3. Competitive disadvantage vs. MATS/ML alumni: the majority of funded individuals have ML-native backgrounds (transformers, interpretability, RL). Eniola's computational neuroscience + pharmacology framing is differentiated but may require more explanation to reviewers less fluent in the neuroscience literature.
4. Provisional patent in Q3 2026: if the CCT architecture is patent-pending, fund managers may question whether LTFF-funded outputs will be open and publicly beneficial — preempt this by clarifying the open-science, preregistered publication plan.
5. Amount: requesting in the $20K–$60K range for a 6-month stipend aligns with funded archetypes; requesting above $80K for a first individual grant without an established track record of LTFF collaboration would face higher scrutiny.