Corrections

Unsupported attributions in nine archived articles have been withdrawn. Each original URL now explains the correction and separates background reading from the original source claim.

• Agents need better handoffs before more autonomy

  The earlier article attributed a broad argument about human handoffs to Google DeepMind without identifying a supporting publication. We withdraw that attribution.

• Benchmarks are moving into the real world

  The earlier article presented a general claim about evaluation trends as a statement from Hugging Face. Its blog index did not establish that claim, so we withdraw the attribution.

• Compute is a product decision

  The earlier article attributed general claims about deployment tradeoffs to Google Cloud without a specific supporting document. We withdraw those attributions.

• Enterprise AI is becoming a workflow decision

  The earlier article described enterprise purchasing priorities as statements by Google Cloud without identifying an announcement or study. We withdraw those attributions.

• Evaluation is becoming the new interface

  The earlier article presented our interpretation of evaluation as a statement from NIST. We withdraw that attribution and the suggestion that it described a new announcement.

• How to read a model release without the benchmark theater

  The earlier article presented advice on reading model releases as statements from OpenAI Research. A research index did not substantiate those statements. We withdraw those attributions.

• Risk frameworks are becoming procurement language

  The earlier article attributed a claim about procurement practices to NIST without evidence that buyers had adopted those practices. We withdraw that attribution.

• Small models still win on the edge

  The earlier headline asserted that small models win on edge devices without identifying the task or comparison. We withdraw that broad conclusion and the unsupported attribution to the Hugging Face blog.

• The quiet economics of inference

  The earlier article attributed a general claim about tool-call costs and failures to Google Cloud without a supporting document. We withdraw that attribution.