Jeva: System One Models from First Principles
Jeva is pitched as a decision-native model design that preserves option structure instead of turning choices into serialized text.
The paper says Jeva separately encodes the request and each candidate option, then refines the request against reusable option memory. Its design targets four requirements: explicit null answers, invariance to option order, variable-size candidate sets, and cacheable option computation. The authors report near-perfect null detection, no decision flips under tested permutations, scaling to thousands of candidates, and lower latency from cross-request reuse. Source: ArXiv · AI/CL/LG's note.
The paper says Jeva separately encodes the request and each candidate option, then refines the request against reusable option memory. Its design targets four requirements: explicit null answers, invariance to option order, variable-size candidate sets, and cacheable option computation. The authors report near-perfect null detection, no decision flips under tested permutations, scaling to thousands of candidates, and lower latency from cross-request reuse. Source: ArXiv · AI/CL/LG's note.
score 5