Story

On air. The free wire runs 15 minutes behind live. See pricing

World, crypto, finance and tech news, scored for impact.

CryptoTechFlowChinese, translatedFiled 4 h ago

From AGI to ASI: Four Paths to Superintelligence

Original headline从 AGI 到 ASI:通往超级智能的四条路

Open the sourceReported by TechFlow, techflowpost.com, 20 November 2026

Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.

Google DeepMind research scientist Tim Genewein outlined four technical pathways from human-level AGI to artificial superintelligence (ASI) in a long-form analysis, framing intelligence as a continuous spectrum rather than a single milestone. The article identifies scaling of models, data and compute; paradigm shifts beyond autoregressive Transformers toward search-based planning, neuro-symbolic hybrids and neuromorphic architectures; coordination among millions of specialized AI agents; and recursive self-improvement as the complementary routes to ASI. It argues that "cognitive industrialization" is compounding effective compute through three multiplying drivers: roughly 1.5x annual gains in hardware performance per dollar, 3-6x yearly declines in the compute needed to train a given capability plus about 3x annual inference efficiency gains, and around 2.5x annual growth in AI infrastructure investment. Genewein contends that even without further architectural breakthroughs, these forces could drive the marginal cost of AGI-level capability toward near zero, enabling millions of digital experts to work in parallel on scientific and engineering problems. The piece stresses that the four paths are not mutually exclusive but reinforce one another, with scaling underpinning new paradigms and agent coordination feeding back into faster progress. What to watch next is whether recursive self-improvement closes the loop, letting AGI systems design and test their own successors and potentially trigger an intelligence explosion.

Muhabbit's AI wrote this summary in its own words from TechFlow's report and translated it from Chinese. It can get details wrong, so check the source.

Named in this story

ASI, Google DeepMind, AGI, Tim Genewein