reasoning models

artificial intelligence Sep 28, 2026 5 min read

AI Needs a Second Gear: Metacognition and Slow Thinking

AI does not become reliable merely by generating longer answers. Metacognition helps a system judge difficulty, switch from fast pattern matching to slower reasoning, use tools, verify intermediate steps, and communicate uncertainty.

by ahsan

#artificial intelligence #large language models #metacognition #reasoning models #test-time compute

artificial intelligence Sep 27, 2026 5 min read

Ember-1: Teaching AI Agents to Reason Efficiently

A practical guide to the engineering idea behind Ember-1: teach a reasoning model to preserve useful reflection while spending fewer tokens on repetitive planning. The post explains token costs, agent loops, training, benchmarks, and production evaluation.

by ahsan

#ai agents #ember-1 #llm inference #reasoning models #token efficiency