Data quality means the examples a model learns from set the habits it picks up. A model copies what it is shown, the careful parts and the sloppy parts alike.
A short explainer video, under a minute. The engineer and the robot are an illustration.
At a glance
What Data Quality Is
Her fix was simple: keep only the examples that cite their claims.
Which Examples to Keep
The video sets two choices side by side. Keep everything, and the pile is huge but the robot learns the sloppy habits.
Why It Matters
The video turns this into one question: what does a good example look like in our work?
Picture an engineer who teaches a robot to write reports. She gives it a big pile of examples.
The robot copies them all, even the sloppy ones, where a claim comes with no source. So its own reports do the same.
What Data Quality Is
Her fix was simple: keep only the examples that cite their claims. That is data quality.
The idea in the video is plain. A model copies what it is shown, so the choice of examples is a choice about its habits.
Which Examples to Keep
The video sets two choices side by side. Keep everything, and the pile is huge but the robot learns the sloppy habits. Keep only the cited ones, and there are far fewer examples, but it learns the careful ones.
The board names a risk on each side. The first copies the noise. The second drops some good examples.
The Idea on One Board
Which examples to keep
Keep everything
Examples
a huge pile
Habits it copies
the sloppy ones
Risk
copies the noise
Keep the cited ones
Examples
far fewer
Habits it copies
the careful ones
Risk
drops some good ones
The honest limit
✓ A filter drops good examples too
✓ Check what it removes
The Honest Limit
A filter can throw good examples away too. The video’s advice is to check what it removes.
A cleaner set is not automatically a complete one.
Why It Matters
The video turns this into one question: what does a good example look like in our work?
Its takeaway is short. A model copies its examples, so check the examples first.
The key insight: A model copies its examples. So check the examples first.
Data quality is where one open lab says it found its biggest gain. Our report on Ai2’s AstaBrief says Ai2 credits its largest training improvement to a plain filter that keeps only reports that cite their claims, and says more elaborate filtering did not help. Starting from a pool of 90K research queries, Ai2 says, it kept 47K usable training examples after quality filtering. These are Ai2’s claims, and our piece verified none of them independently.
The Business Engineer Lens
Where this idea sits in the Business Engineer library:
The Business Engineer framework maps where each part of the AI stack sits and who holds the leverage. These pills are short lessons from the same library.
Data Quality builds on Data Labeling. Both say a model copies what it is shown: labels set the ceiling, and sloppy examples teach sloppy habits.
Data Quality constrains Synthetic Data. Invented complaints come out tidy and polite while real ones are not, a quality gap the model then copies.
Data Quality constrains Few-Shot Prompting. Few-shot models copy the flaws of the examples too, the data-quality lesson’s point applied at prompt time.
Data Quality contrasts with Unlearning(lesson unlocks soon). Filtering examples before training is something you can check; taking one back out afterwards means retraining or hoping.
Data Flywheel(lesson unlocks soon) builds on Data Quality. The flywheel feeds user corrections back into the product, and the data-quality lesson says a model copies what it is shown.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
