TypeSafe co-founder Diogo Almeida tells Sam Charrington why he believes models optimized for software decisions can help automate workflows that remain unreliable with text-generating systems.
Almeida describes Jev’s decision interface and the company’s work on calibrated probabilitiesConfidence calibration measures whether an AI system's stated certainty matches how often its predictions or assumptions are actually correct.. He discusses the use of open-weight modelsAn open-weight model makes its trained parameter values available so others can run or adapt it under the release's licence. and contrasts its training objective with reinforcement learning from human feedbackReinforcement learning trains an AI system to choose actions using feedback about the results of its earlier choices., while acknowledging that Jev is an early release with limitations.
Charrington challenges the assumptions behind the approach, asking about generalization, models versus harnessesAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction., and the tradeoff between flexible agents and engineered workflows. Almeida’s assessments and predictions are presented as his own.
Watch on YouTube
