daily notes

Verification moves into the inference stack

A revised VLA study shows that linear representation-level controls can improve constraint satisfaction with about 1% inference overhead, while vLLM extends adaptive verification to every draft-model speculator. The methods differ, but both place verification at the selection or control point rather than treating it as a post-hoc score.

drmapavoneoriginal post ↗

Can we steer a robot foundation model without retraining it? Our new paper explores this question through control-inspired notions of feature observability and controllability in Vision-Language-Action (VLA) models. We find that state- and action-relevant information can often be linearly predicted from VLA internal representations—and that these representations can be steered at inference time to change robot behavior, without fine-tuning. As an example, on a real robot, our approach increased the preferred handle-grasp rate from 14% to 74%, while adding only ~1% inference-time overhead. The broader idea is that representation-level control could provide a lightweight interface for adapting robot foundation models to new preferences and constraints—without retraining. Joint work with Hugo Buurmeijer Carmen Amo Alonso @SwannAiden Paper: https://t.co/SMa318wfrt #PhysicalAI #Robotics #VLA #RobotLearning @StanfordEng @StanfordAILab

The labor market's entry point is tightening

Two September 22 Fed studies place the weak point at hiring rather than separations. Dallas Fed finds that recent Texas graduates from more GenAI-exposed majors saw a 1.7-point lower employment probability and 5% lower first-year wages after ChatGPT, while Richmond Fed's July map puts hiring at the 43rd percentile and separations at the 90th; the Texas comparison is not a national causal estimate.