Descriptions:
The Cognitive Revolution’s weekly AI in the AM highlights reel covers several significant developments from the frontier, anchored by a deep dive into Anthropic’s July 7, 2026 paper introducing the concept of the J-Space — a global workspace inside large language models where concepts appear to be held in mind during inference. Hosts Nathan and Pash walk through how the J-Lens probe works, noting that interventions into the J-Space produce interpretable behavior changes roughly 50–70% of the time, a result they describe as “incomprehensibly better than random” while acknowledging the remaining dark matter in model cognition.
The episode also covers field notes from the AI Engineer World’s Fair, including a segment with Dan Schwarz of Future Search on AI superforecasters. Schwarz explains the methodological challenge of evaluating AI forecasting — unlike humans, a year-old AI snapshot is nearly obsolete — and describes their “bench to the future” past-casting technique, which enabled them to evaluate Claude Fable within 24 hours of release. He argues the evidence over the past six to twelve months confirms AI forecasting is at minimum competitive with human superforecaster teams.
Additional segments touch on Zeve Farman discussing open world models at the World’s Fair, Kunlay Olukotun of SambaNova explaining why inference is fundamentally a data movement problem, and a discussion of LTX video generation. The episode is a reliable digest for practitioners tracking multiple AI subfields simultaneously.
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published July 09, 2026
🏷️ Format: Roundup
