Aubrai & RMR2 Updates

The Spaces convened Bio Protocol’s Eric, Aubrey de Grey and David Wood (LEV Foundation), Matt (VitaDAO), and the Aubrey AI engineering team (Akash and Marco) to update the community on Aubrey AI, the hypothesis challenge, and the RMR aging studies. Eric shared metrics: the Aubreye token’s FDV rose to $35.6M from $200k, $340k in research fees in under a month, and 12 hypotheses funded toward Aubrey de Grey’s lab and the RMR2 study; demos will be shown at Token 2049’s Decent Summit in Singapore. Matt outlined a two-round hypothesis challenge sourcing high-risk/high-reward ideas, “tech tree” enablers (e.g., scarless regeneration), and biomarkers, with IP tokens launching soon. Aubrey and David recapped RMR1 (additive effects in females, less clear in males; didn’t beat the ~4-month mid-life extension record) and detailed RMR2: 2,000 mice, eight interventions, pilot replications, dosing/order/frequency design, synergy/antagonism testing, and iterative open data releases. Akash and Marco described the agent’s continuous knowledge integration, knowledge graph grounding (80% claims now sourced), 24/7 reasoning, and planned dataset integrations. The session emphasized DeSci’s role in fixing incentives, tokenized IP flywheels for treasury value, and expectations that RMR2 results will catalyze broader funding beyond crypto.

Aubrey AI x LEV Foundation Twitter Spaces — Comprehensive Summary and Notes

Housekeeping and Event Updates

• Token 2049 Singapore: The team will host a smaller, more exclusive Decent Summit with a dedicated Aubrey AI section featuring demos. Programming also includes extensive talks on real-world assets (RWA) and scientific intellectual property (IP), plus fireside chats (one expected with Arthur Hayes). Interested attendees should apply via the event’s Luma link shared by the Bio Protocol account.
• Public metrics dashboard: Aubr.ai provides up-to-date, publicly listed metrics on the agent and token performance for anyone to follow.

Project Metrics and Tokenomics

• Token performance: Fully diluted valuation (FDV) has grown to ~$35.6M USD from an initial ~$200k USD.
• Research funding impact: In under a month, ~$340,000 USD has been generated in fees directed toward research.
• Hypothesis pipeline: 12 hypotheses have been generated so far via community interaction with the Aubrey agent.
• Funding allocation: Proceeds are routed to Dr. Aubrey de Grey’s lab, supporting the RMR studies—especially RMR2.
• On-chain IP: When Aubrey AI produces a hypothesis, it can be minted on-chain; any resultant IP can be tokenized, with a portion of those IP tokens accruing to the Aubrey AI treasury. This mirrors the flywheel dynamics observed in VitaDAO’s model and creates long-term value alignment for Aubreye token holders.

Hypothesis Challenge (Longevity Challenge)

• Objective: Source high-quality ideas broadly—beyond traditional academic pipelines—then use Aubrey AI as a scientific copilot to turn them into viable experimental plans.
• Review: Submissions are reviewed by the LEV Foundation and the VitaDAO team (Matt, PhD candidate).
• Current status:
  • Multiple hypotheses received from diverse sources.
  • Two projects soft-vetted by LEV Foundation.
  • One expects to launch IP tokens next week; another is awaiting a CRO (Contract Research Organization) quote (launch potentially in ~2 weeks).
• Two-round process:
  • Round 1: Interact with Aubrey AI on X to co-develop and refine a hypothesis; submit via Google Form.
  • Round 2: Selected participants receive terminal access plus prompt guidance to improve methodologies, then resubmit refined hypotheses.
• What makes a good hypothesis (categories Matt outlined):
  • High-risk/high-reward, outside the usual “lamp post”—bold directions with strong translational potential.
  • “Tech tree” enablers—intermediate steps that unlock bigger outcomes (e.g., scarless wound healing and regeneration as precursors to organ replacement therapies).
  • Biomarkers—robust measures to assess whether interventions work, beyond lifespan, which is slow and resource-intensive.
  • Alignment with de Grey’s damage-repair paradigm—hypotheses targeting accumulated damage and its reversal.

RMR Program Overview

RMR1 Outcomes (Dr. Aubrey de Grey)

• Goals: 1) Demonstrate additivity of multiple damage-repair interventions, 2) Achieve record-setting magnitude of lifespan extension starting at middle age (≈1.5 years in mice).
• Results:
  • Additivity achieved: In female mice, groups receiving more interventions did better than those receiving fewer; controls did worst. Effects in male mice were less clear and remain under investigation.
  • Magnitude: Did not break the current record (~4 months extension in middle-aged mice). This motivates continued experimentation.
• Insight: The field now has a steady stream of interventions that modestly extend lifespan in middle-aged mice, providing a richer set of candidates for combinatorial testing.

RMR2 Plan and Methodology (Dr. David Wood and Dr. Aubrey de Grey)

• Scale: ~2,000 mice (up from ~1,000 in RMR1).
• Interventions: At least eight interventions (up from four), leveraging the expanded universe of validated, modestly effective treatments.
• Reproducibility-first pilots:
  • Before deploying the full 2,000-mouse study, several smaller pilot arms will confirm that previously reported individual intervention effects can be reproduced under current experimental conditions.
  • Pilots will also test for expected physiological responses and probe select combinations where theory suggests synergy or antagonism (several were independently flagged by Aubrey AI).
• Practical considerations: Budget allocated for pilots; ongoing negotiation over reagent costs (some unexpectedly high due to vendor courting). Team is optimizing for sustainability and rigor over speed (“more haste sometimes leads to less speed”).
• Publication cadence: Expect iterative pilot data releases, not immediate full-scale results from 2,000 mice next week. The approach aims to build confidence and refine design ahead of scale-up.
• Design variables under consideration:
  • Dose tuning, potential antagonisms, and mechanistically informed adjustments.
  • Sequential timing and periodicity (e.g., alternating weekly dosing or intervals between interventions).
  • Case-by-case decisions driven by mechanistic hypotheses, with acknowledgement that some design choices remain informed speculation pending empirical data.
• Team recognition: Shout-out to lead scientist Caitlin Lewis for extensive effort in nailing down these design details.

Open Access and Iterative Publishing

• Funding model advantage: Because the work is philanthropically funded via mechanisms outside of traditional government or industry channels, ongoing data sharing during experiments is feasible.
• Impact: Enables continuous community engagement, hypothesis iteration, and agent updates (including direct dataset integration), rather than waiting years for a single publication.

Sequential Therapy Timing Discussion (Matt’s Question; Dr. de Grey’s Response)

• Challenge: Sequencing matters—for example, whether to deliver senolytics before, during, or after partial reprogramming can materially affect local efficiency due to senescent cell secretions.
• Approach:
  • Budget-optimized learning: Prioritize replication of known encouraging results with similar dosages; layer mechanistically informed adjustments for combinations.
  • Timing strategies: Consider periodic vs. single administration; simultaneous vs. alternating schedules; introduce intervals between interventions where warranted.
  • Reality check: Many combinations and timings would ideally be explored (20,000 mice scale) but constraints require staged, highest-yield designs first.

Aubrey Agent: Architecture, Intelligence, and Safety

Continuous Knowledge Integration (Akash)

• Base capabilities: The agent has read thousands of papers; uses Open Scholar (base model ingesting ~2.5M papers).
• Continuous updates: Active ingestion of new scientific literature and direct integration of emerging RMR2 datasets to keep knowledge fresh and context-aware.
• Intelligence as a spectrum: In some dimensions the agent surpasses human breadth (paper ingestion); in others, domain experts like Dr. de Grey provide nuanced judgment. The interaction is complementary, not hierarchical.

Hallucination Reduction and Claim Grounding (Marco)

• Goal: Each claim backed by a source—paper, knowledge graph entry, or vetted internal notes/unpublished research from Dr. de Grey.
• Progress: Claim grounding improved significantly (from approximately ~50% to ~80%), with a path toward 100%.
• Knowledge sources expanding: Integration of domain databases (e.g., PubMed, UniProt, Gene Ontology) and rodent aging datasets (including RMR study result repositories). Agent gains browsing capabilities over these sources before answering.
• 24/7 reasoning: The agent can continuously generate, evaluate, and refine hypotheses; routine human review (daily/weekly) can elevate quality and filter noise.

Knowledge Graph: Precision, Disambiguation, and Inference (Marco)

• Why a graph: Text-only semantic retrieval often misses structured relationships between entities (e.g., disease subtypes, molecular pathways). Graphs encode explicit links between terms, papers, findings, and directions of effect.
• Precision gains: Graph structure enforces specificity (e.g., disease type A vs. type B) and prevents conflation (e.g., song vs. film named “Hotel California”).
• Novelty detection: Absence of an edge between two biological entities across thousands of papers can signal unexplored combinations worth testing.
• Graph inference: Defined rules enable the agent to infer new relationships automatically, expanding the knowledge base as hypotheses are minted and papers added.

DeSci and Incentive Alignment

• Problem framing (Matt): Traditional funding structures push career scientists toward incremental, “safe” publications; big swings remain underfunded because failure risks careers.
• Role of the agent: Provide mentorship-like guidance at scale—improving hypothesis quality, refining experimental plans, and increasing the odds that high-risk ideas are testable and informative.
• Theory of change (Eric): The Center for Open Science’s five-layer pyramid—policy, incentives, communities, applications, infrastructure—needs simultaneous movement. DeSci (referred to as DCI in the conversation) can catalyze reforms across funding, community-building, publishing models, and infrastructure (as exemplified by this call and the tools being built).

Funding Outlook and Catalytic Potential (Dr. David Wood)

• Expectation: RMR2 is likely to outperform prior combinatorial attempts. Visible, iterative success should catalyze broader interest and new funding beyond the crypto community.
• Path to scale: Rather than centralizing 20,000-mouse experiments in RMR3/4 alone, expect distributed adoption—more labs, more combinations, and exponential growth in testing variety driven by demonstrated results.

Actionable Next Steps

• Attend: Apply for Decent Summit during Token 2049 (Singapore) via the Luma link posted by Bio Protocol.
• Participate: Interact with Aubrey AI on X, craft and submit hypotheses via the Google Form; watch for Round 2 invitations with terminal access and prompt guidance.
• Follow: Track real-time metrics and updates at Aubr.ai; expect pilot data releases as RMR2 proceeds.
• Watch for IP tokens: One project launching next week; another likely in ~2 weeks pending CRO quote.

Participants and Roles (Inferred from context)

• Eric (Host, Bio Protocol): Moderation, tokenomics context, DeSci vision, and event updates.
• Dr. Aubrey de Grey (LEV Foundation): RMR program rationale, RMR1 outcomes, RMR2 design goals, open data stance, and scientific leadership.
• Dr. David Wood (LEV Foundation Executive Director): RMR2 pilot strategy, reproducibility focus, budgeting and reagent procurement, publication cadence, and funding outlook.
• Matt (VitaDAO team; PhD candidate): Hypothesis Challenge structure and criteria; perspective on reprogramming and multi-pronged aging interventions; questions on sequencing.
• Akash (Bio Protocol, AI team): Agent knowledge integration strategy; continuous updating and dataset ingestion.
• Marco (Bio Protocol, AI engineer): Hallucination reduction, claim grounding, knowledge graph design and benefits, and continuous agent operation.

Key Highlights

• Additivity confirmed in RMR1 (stronger in female mice), but not yet record-breaking magnitude—motivating RMR2.
• RMR2 will scale interventions and mice, with reproducibility-first pilots to de-risk and tune combinations before full deployment.
• Continuous, open publication is enabled by novel funding mechanisms—allowing dynamic community input and agent updates.
• Aubrey AI is evolving rapidly: stronger grounding, broader knowledge sources, and a structured knowledge graph that elevates precision and novelty detection.
• The Hypothesis Challenge provides a pathway for non-traditional contributors to propose bold, damage-repair-aligned ideas, supported by the agent’s mentorship-like capabilities.
• Token holders benefit from on-chain IP and treasury accrual tied to agent-generated hypotheses—creating a research-to-value flywheel.
• Anticipated RMR2 progress could trigger broader, non-crypto funding influx and distributed experimentation across the field.

Closing Sentiments

• Dr. de Grey expressed deep gratitude for the community and the funding model that is finally enabling vital studies to proceed at pace—and to share results iteratively as they emerge.
• The session closed with encouragement to continue engaging, submit hypotheses, and follow upcoming pilot releases and IP token launches.