Corrections
We correct factual errors and identify material changes in the affected article. A correction explains what was wrong and what changed; it does not silently replace a published fact.
Point to the passage, explain the issue and include a source when possible. Your email is optional; include it if you would like a reply. This form also accepts requests about personal data.
Correction history
- · Ai2 introduces ToyBlocks to study how robot skills transfer
Correction: ToyBlocks also includes physical evaluation with a robot arm and cameras; its results are not limited to simulation.
- · CWE-bench v2 adds 100 repository security repair tasks
Correction: CWE-Bench scoring is programmatic and deterministic. Model panels helped curate checks; they did not judge benchmark runs.
- · IBM updates Granite Switch speech and decision components
Correction: the cited update concerns the Decision Adapter in granitelib-core. Jev Decision Index is the test suite used to evaluate performance, not an updated component.
- · LeRobot adds dataset streaming and live policy instructions
Clarification: /vqa and /autosteer depend on text support in the policy or an external planner, according to the documented configuration.