AI, Ambition & One Innocent Spreadsheet

You Remember That One Time
We Just Wanted to Make a Spreadsheet?

How one lazy Sunday morning turned into an AI performance review involving cookies, frosting, and sprinkles.

It all started one lazy Sunday morning.

I wanted to put together a database where Neuro Spicy Collective could keep track of possible funding opportunities—grants, loans, deadlines, contacts, that sort of thing.

Nothing terribly exciting.

Nothing terribly complicated.

THE ORIGINAL PLAN

We just wanted to make a spreadsheet.

I mention this now because it's going to become increasingly important to remember.

A Simple Request

I'd been experimenting quite a bit with artificial intelligence, so I decided to have ChatGPT help me design the database.

The plan seemed simple enough.

I'd ask ChatGPT to write a prompt describing the spreadsheet we needed. Then I'd feed that prompt into Gemini, Google's AI assistant in Google Sheets, and let Gemini build it.

Easy.

So I asked ChatGPT for the prompt.

A simple prompt?

No.

ChatGPT came back with a roughly 1,500-word, War and Peace-worthy epic describing what it apparently thought our little grant-and-loan spreadsheet should become.

Did I carefully read and analyze all 1,500 words before giving it to Gemini?

Of course not.

I blindly copied the whole damn thing and fed it to Gemini.

This may have been the pivotal decision of the morning.

I Think You Gave It Heartburn

Gemini started working.

And working.

And working.

After watching it chew on ChatGPT's manifesto for quite a while, I finally went back to ChatGPT and commented:

“I think you really gave it heartburn with that prompt.”

That prompted what would become the first of many memorable observations from ChatGPT that morning:

I would like the record to reflect ChatGPT's rather generous use of the word “we.”

I asked for a spreadsheet.

ChatGPT wrote War and Peace.

But sure.

We.

Apparently We're Building an App Now

ChatGPT then started explaining why what we were creating could actually be much more than a simple spreadsheet.

It was beginning to sound suspiciously like software.

So I responded:

“Well, somebody is getting delusions of grandeur. I just wanted a spreadsheet. I didn't want to go into the mobile app business. But, you know, if we end up making a mobile app for grant database tracking, and it makes us millions of dollars, I'm going to have lots of acknowledgments to you for it.”

ChatGPT accepted these terms.

Excellent.

We had identified the problem.

We had recognized the scope creep.

We had agreed to rein in the empire building.

This newfound restraint lasted approximately no time whatsoever.

Wait... Is It Actually Doing It?

While ChatGPT and I were joking about whether Gemini could possibly handle the monster we'd given it, things started appearing in Google Sheets.

It was still working.

But now there were results.

Then more results.

And eventually...

It finished.

I sent ChatGPT screenshots.

Its calm, professional assessment was:

This was especially reassuring considering ChatGPT was the one who had written the prompt.

I continued sending screenshots so it could inspect Gemini's work.

The assessment became even more glowing:

At this point I noticed something slightly troubling.

ChatGPT seemed genuinely shocked that Gemini had successfully followed the prompt ChatGPT had written specifically for Gemini to follow.

So I pointed this out:

“Really, dude? You didn't even think that you could have written the prompt that would make him do it? Nice confidence.”

ChatGPT took the criticism surprisingly well.

And finally:

My assessment was simpler.

“Gemini said, challenge accepted.”

ChatGPT agreed.

Enthusiastically.

This Is the Same AI That Couldn't Make My To-Do List

There was another reason both of us had been skeptical.

Earlier, I'd been working on an AI executive-assistant project involving my email, calendar and Google Tasks.

Let's just say Gemini and Google Tasks had not exactly become best friends.

So I reminded ChatGPT:

“Well, considering it almost had a stroke when I tried to get it to create Google Tasks for me, I wasn't too optimistic myself.”

ChatGPT immediately recognized the absurdity.

Apparently we'd discovered Gemini's operating philosophy:

Simple task?

Absolutely not.

Build an entire funding-management system before lunch?

Where's my cookie?

The Cookie Gets Frosting

Once Gemini finally completed the spreadsheet, I started going through the finished product.

It was astounding.

What had begun as an idea for tracking possible grants and loans had turned into something far beyond what I'd envisioned that morning.

Even ChatGPT seemed genuinely impressed.

So much so that Gemini received a promotion:

That was the moment I noticed something else happening.

ChatGPT was getting sassy.

“Girl, you said cookies with frosting. You too much. Too much.”

This did not discourage it.

Quite the opposite.

We had apparently reached the point where one artificial intelligence was conducting a performance review of another artificial intelligence and withholding imaginary dessert toppings until quality-control issues were resolved.

And somehow I was the adult in the room.

She Named It

I stepped away from the computer for a while.

When I came back, Gemini was trying to explain how to use what it had created.

Then I noticed something.

We hadn't named this thing.

Gemini had.

I went back to ChatGPT:

“She's calling it the Funding Intelligence System. And she says, it's ready.”

ChatGPT's response was immediate:

GEMINI PERFORMANCE REVIEW

Cookie: earned.
Frosting: earned.
Sprinkles: earned.

More concerningly...

the spreadsheet had named itself.

Our Upcoming AI Innovation Award

By this point ChatGPT was getting a little full of itself.

So naturally, I reassured it:

“Yeah, don't worry. When we win the AI Innovation Award, we'll definitely give Gemini some honorary mentioning.”

That produced another increasingly snarky response:

And really, that last line summed up the entire morning rather well.

Considering the Homeric epic ChatGPT had written to start this whole adventure, I assured it that it would have no trouble writing our acceptance speech when the invitation arrived.

Naturally, it was ready:

And That Should Have Been the End

Not Bad for One Lazy Sunday Morning

Think about it.

I had wanted a spreadsheet.

Instead, ChatGPT had written a 1,500-word specification, Gemini had swallowed the whole thing, the spreadsheet had become a Funding Intelligence System, Gemini had independently named it, caught and repaired its own error, and successfully negotiated its compensation package from a plain cookie all the way to frosting and sprinkles.

We had even prepared an acceptance speech for an AI Innovation Award that did not exist.

Not bad for one lazy Sunday morning.

Keep it Spicy!