Diogo Almeida joins Jacob Effron to discuss Jev, a model designed for structured decisions consumed by software. Almeida argues that usefulness and reliability matter more than simply producing a new model shape or improving a benchmark scoreEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria..
Diogo Almeida and Jacob Effron examine how persistent context and key-value cachingA key-value cache stores intermediate attention information so an AI language model can reuse earlier computation during generation. shape coding-agent design. Almeida proposes ideas such as shared state and task-aware context managementContext engineering designs how relevant instructions, evidence, state and tool information are assembled for a model's task., framing them as possible future approaches rather than established capabilities.
Diogo Almeida and Jacob Effron discuss multimodal limitations, the practical value of coding agentsAn AI agent is a system that observes context, decides what to do, and takes actions through tools to pursue a goal. and AI risk. Almeida's estimate of productivity gains and his views on future agency are presented as judgments with uncertainty, not measured guarantees.
Watch on YouTube
