Rising-λ AI Futures Model

The AI Futures Model with one λ for the past and another for the future that can rise over time, running entirely in your browser. The dashed line is the preset at λ 0.5, the solid line the two-λ variant, and the dotted line the original model with the same λ everywhere.

Start from
Parallel coding labor
λ rising over time
η, the coding automation efficiency improvement, was estimated at λ 0.5. The variant rescales it for coding labor growth after today; the original model's power form is offered for comparison.
Starting…

When each milestone arrives

Preset λ Two-λ Original, one λ

Effective capability over time

Preset λ Two-λ Original, one λ

log10 effective training compute. Dotted lines are the milestone thresholds.

Milestone table

How to read this

  • λ is the return to parallel coding labor: doubling it multiplies serial labor by 2λ. The original model uses one λ for historical human teams and future AI swarms alike.
  • With one λ, raising it also changes two calibrations of the past: today's measured coding uplift is read as less automation, and returns to software research (r) are refit to 2024 progress. Both lengthen takeoff and can outweigh the direct benefit.
  • The two-λ variant keeps both calibrations at λ 0.5, so today's automation and r are unchanged, and applies the forward λ only to growth in coding labor after today.
  • λ rising over time moves the forward λ from its value today to a higher one over the chosen period. A rise applies to all coding labor growth since today, so better coordination also helps the agents already running. η is rescaled for today's forward λ only. With no rise this is the fixed forward λ. The variant line uses the schedule; the dotted original-model line uses the forward λ alone.
  • AC is the automated coder and ASI artificial superintelligence, as defined by the AI Futures Model. Takeoff is the time from AC to ASI. Coding Uplift anchor.
  • The original model accepts λ up to 1. Everything is solved in your browser with Pyodide. Source and method: aifm-rising-lambda, built on aifm-lambda-fixes.