Pathway announces AI reasoning breakthrough (www.polskieradio.pl)

Poland-born scientist Adrian Kosowski and his Silicon Valley startup Pathway announced a claimed breakthrough they say solves “generalization over time” — the ability to learn from experience and reason about new temporal context — with a new paradigm called “Dragon Hatchling.” According to Pathway’s team (announced on a Cornell webpage and discussed in interviews), their model spontaneously forms a scale-free, brain-like structure during training that resembles neocortical organization and exhibits emergent, ongoing learning behavior. The company positions this as a milestone toward “Universal Reasoning Models” and says the approach powers LiveAI systems that think and update in real time without wholesale replatforming. If validated, the technical implications are large: a model that reliably generalizes across time could change architectures for continual learning, reduce catastrophic forgetting, and enable adaptive decision-making in production systems (finance, defense, motorsport clients are already cited). Pathway ties the work to a high-throughput data engine and an AI framework optimized for real-time training; the project is backed by researcher Lukasz Kaiser. The claim is radical and still needs peer review, benchmarks, and independent replication, but it highlights an industry push toward emergent, biologically inspired structures that support persistent, real-time learning.