gekro
GitHub LinkedIn
News

AI News

UC Berkeley releases CUA-Lite for agent training; Perplexity details GPU embedding infrastructure

UC Berkeley researchers open-sourced CUA-Lite, a unified platform for computer-use agent training and evaluation; Perplexity published technical details on its GPU-based embedding serving stack.

1 min read 3 sources

UC Berkeley researchers released CUA-Lite, an open platform that consolidates four previously incompatible components of computer-use agent development - agents, environments, traces, and evaluation frameworks - behind a single action space and data schema (MarkTechPost). The tool addresses fragmentation in how training data, sandboxes, and evaluation systems format their outputs, removing a bottleneck for practitioners building and benchmarking agents at scale.

Perplexity Engineering published a detailed breakdown of its embedding serving infrastructure - the Ivy, Tulip, and ROSE components that power the pplx-embed model - focusing on how retrieval quality depends on both embedding model quality and inference cost at scale (MarkTechPost). Separately, GitHub introduced Project HydraFusion, a research preview for Copilot CLI that treats multi-model workflow selection as an optimization problem rather than a fixed routing rule, adapting execution patterns - Single, Cascade with quality gates, and Critique with cross-model review - per coding task (MarkTechPost).

Compiled automatically from the linked sources and published without manual editing - a neutral summary of third-party reporting, for information only. Every claim links to its origin. Not original reporting.