Andrew Ng Comments on Jensen Huang’s Departure and Explains Claude Fable 5.2, Atlas, and MEMCON
In this issue, DeepLearning.AI reviews Jensen Huang’s resignation as Nvidia CEO, Anthropic’s launch of Claude Fable 5.2 and limited release of Claude Mythos 5.2, and World Labs’ opening of the Atlas API to developers. Claude Fable 5.2 costs 20% less per token than its predecessor and supports inputs of up to 1 million tokens. Atlas can generate 3D scenes and videos with controllable viewpoints from text or images and export Gaussian splats and polygon meshes. The article also introduces MEMCON, an adaptive memory management method that improves success rates and reduces token consumption across multiple agent benchmarks.
Why read this
AI score 70
This issue places personnel changes at AI companies, tradeoffs between model capabilities and safety, exportable world models, and research on agent memory within a shared industry context.
AI Trends