NVIDIA-Cambridge 'Red Queen Gödel Machine' self-evolving AI paper revives 2028 ASI debate

NVIDIA and the University of Cambridge published a paper introducing the 'Red Queen Gödel Machine,' a self-evolving system that writes and empirically tests its own code, iteratively improving at code generation and mathematical proofs. The framing — a machine that recursively improves itself with verification gates — deliberately echoes the theoretical Gödel machine and triggered fresh debate over whether artificial superintelligence could arrive on a 2028 timeline.

The mechanism is the interesting part and the basis for skepticism: 'self-evolution' here means the system proposes code modifications and keeps only those that pass empirical tests, a constrained search rather than open-ended recursive self-improvement. Critics will note that bounded, benchmark-driven code optimization is a long way from general self-improvement, and that flashy 'ASI by 2028' extrapolations outrun what the paper demonstrates.

The research lands amid a busy NVIDIA week that underscores its dominance and its risks. NVIDIA committed $119 billion in supply, added $80 billion in buyback authorization, and struck an AI-access deal with Australia's Firmus Technologies; its technology powers 81% of TOP500 systems and 90% of systems new to the list. Yet the same week saw Google tout TPU cost advantages and Amazon, Apple and others pass through chip-cost pressure — reinforcing the customer push to build custom silicon and diversify away from NVIDIA's 'software trap.'

For readers, the signal is less the ASI headline than the strategic positioning: NVIDIA is publishing frontier-research credibility while locking in supply, buybacks and geographic access deals. Watch whether the Red Queen results reproduce independently and whether self-evolving code methods make it into production tooling, versus remaining a research provocation that fuels timeline arguments on r/singularity.