Tag: agents
All posts tagged "agents".
Turning Occasional Success into Reliable Capability
4 min readFind, select, and train on paths a model only occasionally discovers to understand how demonstrations, exploration, rewards, and feedback reshape behavior.
Put What the Model Cannot Yet Do into the Harness
4 min readDomain knowledge, practical experience, specialized tools, and workflows bring general models into real work while supplying material for further learning.
Treat an Agent as a Long-Running Asynchronous Task
5 min readLong runtimes, external resources, and unpredictable combinations of actions explain why agents need scheduling, isolation, durable state, and recovery.
As Models Become More Capable, What Remains in the Harness?
5 min readAs developing model capabilities becomes a repeatable process, a harness supports both task execution and a continually moving gap in capability.
When Language Models Learn the Physics of a Text World
5 min readFiles, commands, and tools connect text to a changing environment. As models learn the consequences of actions, context becomes their interface for ongoing interaction.
How Can a Prompt Elicit a Model's Capabilities?
4 min readStart with the relationships learned through next-token prediction to understand how an instruction changes a model's behavior and why prompting works.