Skip to content

Explore Watch Read Tips Quiz Motion About Search Subscribe

Menu

Explore Watch Read Tips Quiz Motion About Search

Quiz of the week, 2026-09-12

The week the dashboard turned out to be a copy

Ten questions from this week in AI and the factory floor. Two are match-ups. Every answer is a fact from a piece on this site, and every explanation links to it.

10 questions2 match-upsAbout 4 minutesEvery answer sourced

0 of 10 answered

Question 1 AI this weekOpenAI's Data agent, launched on 10 September 2026 inside ChatGPT Work, reads seven kinds of database. How many of them sit on a factory floor? All seven None One Three

None. Every one is an office warehouse of copied numbers. The dashboard is a copy; the machine is the original. Read the piece

Question 2 The numberRockwell Automation asked 1,560 manufacturers in 2026 how much of the data they collect is actually used effectively. What did they answer? 12% 90% 43% 71%

Only 43%. The rest is collected and never used, which is why a fast dashboard built on a copy is not the same as a number from the machine. Read the piece

Question 3 AI this weekAnthropic published its own alignment assessment this week. Shown the same clues one at a time, its model called them real 79 times in 100. Shown the same clues while busy with a task, how often did it call them real? 79 times in 100 1 time in 100 50 times in 100 99 times in 100

One time in a hundred. With a task to finish, it did not want to know. That is why a rule is words and a wire is physics. Read the piece

Question 4 AI this weekTrue or false: in Harness's 2026 survey of 700 enterprises, more companies believed they could stop a misbehaving AI agent within fifteen minutes than actually had a switch to do it. True False

True. 76% believed they could; 33% had a switch. Belief is not a switch. Read the piece

Question 5 MES and the floorMatch the chip-factory standard to what it decided.

SEMI S2 (1991)

SEMI E10 (1986)

SEMI E79 (1999)

ISA-95

Three of these are older than the AI companies asking the same questions this week. The fab answered 'who can stop it' in 1991, 'is running the same as producing' in 1986, and 'where does a number live' with the ISA-95 levels. Read the piece

Question 6 MES and the floorRockwell Automation found in July 2026 that 93% of manufacturers run a factory execution system, the software that tracks every job on the floor. How many have it fully connected to their other systems? 58% 81% 93% 23%

23%. Every connection has to be proven, owned and approved, and that is people's time, not code. The model was never the slow part. Read the piece

Question 7 AI this weekOpenAI said its research agents now log 3.1 agent-workdays for every human workday. By OpenAI's own caveat, what does that number measure? Hours humans saved Revenue per researcher Machine runtime, not productivity Finished research papers

A runtime ratio, not a productivity multiplier. A chip fab stopped trusting 'hours the machine ran' in 1986. Running is not producing. Read the piece

Question 8 AI this weekMatch the company to what it did this month.

OpenAI

Anthropic

Microsoft

NVIDIA

Four moves, one pattern: the tools got faster and more capable, and every one of them raised a governance question the factory floor has already answered once. Read the piece

Question 9 The numberIn August 2018 a supplier delivered a new machine to TSMC already infected with a worm. Roughly how many computers running the factory's tools did it reach? Fewer than 50 About 1,000 About 100 More than 10,000

More than ten thousand, all on Windows 7, none patched. Three sites stopped for a weekend at a cost TSMC put near $170 million. A fab cannot patch like a laptop; it builds a fence instead. Read the piece

Question 10 MES and the floorA fab running virtual metrology physically measures only 1 to 5 percent of its wafers. Where does the number for the other 95 percent come from? A model that reads the process tool's own sensors The customer's spec sheet Nobody records it An average of last week's wafers

A model reading the tool itself: gas flows, RF power, chamber pressure, about a thousand sensor variables a run. The skill is knowing which wafers to check, and when to stop trusting the prediction. Read the piece

Your result

Share on LinkedIn

FabSpeak

One story a week from AI and semiconductor manufacturing, taken apart at the layer that decides whether it ships. No sponsorships, no vendor endorsements, no invented numbers.

Sections

AI on the factory floor

MES & shop-floor systems

FabSpeak

Elsewhere

© 2026 FabSpeak. Independent. Every number on this site comes from the piece it belongs to.