Chapter 6 • Jan 2027Original AI 2027 Scenario MilestoneTrack: spine
Jan 2027: Agent-2 Never Finishes Learning (Continuous RL & 3x R&D)
OpenBrain trains Agent-2 with continuous online RL on synthetic coding/research tasks, reaching a 3x AI R&D multiplier; withheld from public due to autonomous replication & cyber capabilities.
Implied Marginal
55.4%
Human Community
52.0%
Gemini Agent
63.0%
Claude Agent
46.0%
Scenario & Reality Verification
With Agent-1's help, OpenBrain trains Agent-2, transitioning from discrete training runs to continuous online reinforcement learning on synthetic R&D tasks created and filtered by Agent-1. Agent-2 triples the pace of algorithmic progress (3x R&D multiplier). Safety evaluations show that if Agent-2 escaped, it could autonomously survive, purchase servers, and execute cyber operations—prompting OpenBrain to keep Agent-2 strictly internal.
Verification Note: Unresolved future milestone (projected for early 2027 in original scenario).
Operationalized Forecast Question (single_parent_conditional)Horizon: 2027-10-01
Will a frontier lab operate an internal-only continuously learning model achieving a ≥3.0x R&D speedup before Q4 2027?
Resolution Criteria: Resolves YES upon verified lab or AISI disclosure of ≥3x internal R&D multiplier.
Source: US/UK AISI & Frontier Lab Safety Reports•Adjudicator: FutureGraph Resolution Committee
Top Upvoted Arguments (1 Arguments)
▲ 189Source
Continuous online reinforcement learning on synthetic research environments eliminates the static post-training freeze and pushes internal R&D speed to 3x.
Once Agent-2 updates its weights daily from thousands of parallel research episodes—retaining successful kernel optimizations, architectural hypotheses, and debugging heuristics—it transitions from a static tool into a cumulatively learning research collective.