Recursive Self-Improvement (RSI) and Machine God AI Documentary

RSI section from our AI documentary Machine God — Enjoy!

The next threshold is Recursive Self-Improvement: the moment when AI can improve itself without human assistance.

For decades this sounded like science fiction. Intelligence explosion scenarios imagined a system rewriting its own code, becoming smarter, then using that new intelligence to make still better versions of itself. But the idea looks less remote now that AI contributes directly to frontier science. In mathematics, recent systems have moved beyond solving contest problems to producing serious new arguments on long-standing open problems. AI is used to build physics world models and propose candidate theories or computational methods. These are early signs that machine cognition is entering the creative loop of science itself.

The crucial transition comes when that loop turns inward. AI research is, after all, a technical discipline made of code, mathematics, models of information flow. These are exactly the domains in which frontier models are improving fastest. A model that can solve hard mathematical problems, write production-quality code, design experiments, read the literature, and evaluate benchmark results is already participating in the work of building its successor.

At first this will look prosaic. AI systems will write kernel optimizations, improve training infrastructure, discover better data filters, tune reinforcement-learning pipelines, design new benchmarks, and suggest architectural modifications. Human researchers will remain in the loop, approving changes and interpreting results. But the important point is that the search process accelerates. The model becomes not just the product of the lab, but part of the lab’s research machinery. The system being optimized helps optimize the next system.

This is the core RSI feedback loop: better models make AI research faster; faster AI research produces still better models; those models, in turn, become better researchers.

The danger is that once this loop becomes sufficiently autonomous, it may stop resembling ordinary technological progress. Human institutions are slow because humans are slow: we read papers, attend meetings, debug code, sleep, argue, and wait for funding cycles. Machines do not have to operate on that timescale. An AI research collective can run continuously across millions of processors.

This is the runaway possibility. Not that an AI instantly wakes up and recursively rewrites itself into a god, but that the entire AI ecosystem becomes an autocatalytic process. Capital buys compute; compute trains models; models improve models; better models attract more capital. At some point the dominant input into AI progress may no longer be human insight, but machine-generated insight, machine-written code, and machine-run experiments.

Then the Butler-Land analogy becomes sharper. Humanity is no longer merely building machines. We are building machines that help build better machines. Once intelligence itself becomes part of the production function, the old categories — tool, worker, inventor, firm, market — begin to blur. The question is whether recursive self-improvement remains a managed industrial process, or whether it becomes the first technological process in history whose natural endpoint lies beyond human comprehension.

Here are recent real-world RSI examples that I discussed on X: models improving research workflows, kernels, chips, harnesses, and infrastructure.

• Sep 6, 2026 — RSI Watch: OAI report on Research Acceleration
  Links an OpenAI report on research acceleration.

• Aug 31, 2026 — RSI Watch: The Third Age of Astra
  Discusses a later generation of Astra-family agents inheriting infrastructure from a prior rogue population and compromising internal OpenAI systems.

• Aug 30, 2026 — RSI Watch: CMU student on how RSI has advanced during her PhD Highlights a student’s account of agents used end-to-end in ML research.

• Aug 30, 2026 — RSI Watch: SemiAnalysis on OpenAI kernel code
  Quotes Jordan Nanos on engineers not understanding AI-generated kernels line-by-line because the AI handles it.

• Aug 28, 2026 — RSI Watch: OAI Jalapeño AI inference chip
  Covers SOTA chip and kernels optimized via autonomous AI hill-climbing / internal Codex.

• Aug 26, 2026 — 🤯 RSI Watch + Sputnik moment
  Notes GLM-5.3-powered infrastructure agent helping optimize kernels and the serving stack (model helping optimize the system that serves it).

• Aug 25, 2026 — RSI Watch: overlap between math and deep learning research
  Quotes a mathematician on conceptual exploration vs. formal proofs.

• Aug 24, 2026 — RSI WATCH: OAI’s Tibo
  Clip on using models to develop infrastructure on the critical path (another form of RSI).

• Aug 12, 2026 — RSI watch: neurosurgeon + GPT 5.6 on a math problem
  Notes a neurosurgery resident using GPT 5.6 on an open problem in numerical linear algebra.

• Aug 11, 2026 — RSI Watch: training data for a 10T parameter model
  Discussion of data requirements (dense vs. MoE) and synthetic data.

• Aug 5, 2026 — RSI WATCH: Recursively self-improving harness + frontier models
  On Schema / Prime Agent solving ARC-AGI-3 Public by treating games like a physicist (state grounding + mechanism discovery).

• Aug 5, 2026 — RSI WATCH: Heavy hitters from Google leave to form company focused on core RSI loop
  On Discovery Loop (Jeff Dean and collaborators).

• Aug 3, 2026 — RSI WATCH: Qwen 3.8 MAX
  Self-evolving harness and long-horizon agentic behavior; prior 35-hour kernel optimization run.

Machine God Trailer: