With the arrival of Claude 3.7 Sonnet and its extended thinking capabilities, we’ve created a semi-symbolic reasoning protocol using Unicode that can potentially unlock alternative (read: less probabilistic) thinking paths when working with Claude.

New research from Anthropic highlights the ways AI is actually being used to augment and automate work, based on US Department of Labor job data analyzed against real Claude usage.

MCog Core is a metacognition ontology by the Galaxy Brain team, useful for equipping AI agents with a system for thinking about thinking.

Exploring a new pattern in how LLMs structure knowledge: an experiment in letting AI naturally diffuse information outward from seed topics, creating rich networks of unexpected connections.