Hi Reader,

Welcome to the Future Proof newsletter, your Monday morning espresso, in AI form! (Formerly the Google Sheets Tips newsletter.)

One of the stories that jumped out to me in the past week came from an interview with a tearful Cornell mathematics professor, asking "what's the point?" when AI surpasses human ability and steals the meaning (and joy) from mathematical research. (Which is a calling, not just a job to many.)

One line stuck with me, and I'm paraphrasing here, but basically, "in the same way that I can't teach linear algebra to my dog, what if AI becomes so advanced that we become the dog."

Whether you care for mathematics or not—they're the canary in the coal mine here—it asks us some uncomfortable questions since the same transformation will eventually reach every knowledge job (and let's not even mention robotics today!).

So the question becomes, how do we stay useful? And find meaning in our work?

I wish I had a pithy answer for you, but the truth is, nobody knows.

I think all we can do is adapt to these changes and remind ourselves that moving from what's theoretically possible to what actually works in the real world is incredibly messy and takes a long time. (Evidence: the many reports questioning the return on corporate AI spending!)

Anyway, let's come back down to earth. Today we're looking at a simple technique to improve the accuracy of your results from AI tools. But first a roundup of news, including an announcement from me...

→ News

I.
NEW COURSE COMING VERY SOON!

I've just finished recording my latest course:

⭐ From Idea to App: Gemini for Builders ⭐

• It's designed to teach you best practices for building software with AI tools, aka "vibe coding".
• It picks up where the 28 Days to Gemini Mastery course left off (although that's not a prerequisite).

It will be available next week or early November!

II.
​An update on the launch of skills in the Gemini app and Workspace.

TLDR; Skills in Gemini app are rolling out between now and mid-November. Skills in Workspace are delayed until further notice.

III.
​New logic and search steps in Workspace Studio help expand automation capabilities.

The Google team continue to expand the capabilities of Workspace Studio with another feature announcement.

(Workspace Studio is Google's version of Zapier. It connects Workspace apps and lets you build automations "if this happens, do that". It's only available to paying Workspace customers.)

→ Future Proof Tip #6: Use Embedded Quality Assurance Checks To Improve Accuracy

Embedded Quality Assurance (QA) means building a check into your prompt so that the model verifies its output before returning the results.

It can dramatically improve the precision of results in certain scenarios.

Of course, not every prompt suddenly requires a QA clause. In fact, most don't.

But if you're dealing with data, calculations, or lots of assumptions, then it's a useful technique to save you time running extra prompts to check or fix these things.

Here's an example from my own work, when I needed a photo-realistic receipt for teaching purposes.

Notice the QA check in the second paragraph:

"Ensure that all line items, subtotals, calculated taxes (if any, specified with a percentage), and totals are accurate, mathematically correct, and consistent throughout the text of the generated receipt."

This receipt was used to show how an AI tool can extract the data from an image and put it into a Google Sheet. So it was important that the receipt be mathematically accurate otherwise the demonstration would fall apart.

How To Add A QA Check

For situations involving data analysis, calculations, or factual claims, add a QA clause to the end of your prompt.

A QA clause is one or two extra sentences that ask the model to verify its figures and assumptions.

Be sure to include specific criteria and instructions rather than general vague statements like "make sure it works".

Example QA Checks

• Ensure that all line items, totals, and calculations are accurate, mathematically correct, and consistent throughout.
• For every claim, include the page or paragraph it came from. If you can’t find support for a claim, leave it out and flag it.
• Mark each statement as confident, likely or uncertain, and say what you’d need to verify the uncertain ones.
• Review your draft against these criteria: under 150 words, one clear ask, no jargon. Fix anything that fails and give me only the final version.

Pro Tip: If you're struggling to come up with a QA check, ask Gemini to add one to your prompt.

For example:

Before I run this prompt, add a short QA check line at the end.

My prompt: "Our revenue went from £100k to £200k in year one, then fell back to £100k in year two. What was the average annual growth rate?"

Gemini will then revise your prompt with a QA line included, which you can then use in a new chat:

"Our revenue went from £100k to £200k in year one, then fell back to £100k in year two. What was the average annual growth rate?

QA Check: Please show the step-by-step mathematical calculations for both the Arithmetic Mean and the Compound Annual Growth Rate (CAGR) before providing the final answer to verify that the math is correct."

Notes

• This technique works great in any other LLM like Claude or ChatGPT too.
• You do not need to add a QA check to every prompt! Only ones that would benefit from additional accuracy check.

If you enjoyed this newsletter, please forward it to a friend who might enjoy it.

Have a great week!

Cheers,
Ben

P.S. We're all ironing boards.​