Happy Hump Day from a hot, sticky Asti!

Jeff C., you sent in a question ages ago. Unfortunately, I pasted it into the wrong folder and lost it until a few days ago.

Today, I answer it. It’s a great question, worthy of an entire Rude.

In today’s article, you said ChatGPT helped you in your work. After several examples in the news of completely erroneous results from (biased?) AI, why would you trust it?

-Jeff C., Decatur, GA

Is AI Biased?

I’d say ChatGPT is. I wrote a piece on it for the Morning Reckoning in March.

Here’s a hilarious example of its bias:

Credit: Daily Reckoning

Seriously?

But the following example is even more egregious.

When AI Entirely Backfired.

If you’re wondering where Jeff is getting the “erroneous results” from, here’s perhaps the most famous case.

From The Guardian:

This story begins on 27 August 2019, when Roberto Mata was a passenger on an Avianca flight 670 from El Salvador to New York, and a metal food and drink trolley allegedly injured his knee. As is the American way, Mata duly sued Avianca, and the airline responded by asking that the case be dismissed because “the statute of limitations had expired.” Mata’s lawyers argued on 25 April that the lawsuit should be continued and appended a list of over half a dozen previous court cases that apparently set precedents supporting their argument.

Avianca’s lawyers and Judge P Kevin Castel then dutifully embarked on an examination of these “precedents,” only to find that none of the decisions or the legal quotations cited and summarised in the brief existed.

Why? Because ChatGPT had made them up. Whereupon, as the New York Times report puts it, “the lawyer who created the brief, Steven A Schwartz of the firm Levidow, Levidow & Oberman, threw himself on the mercy of the court… saying in an affidavit that he had used the artificial intelligence program to do his legal research – ‘a source that has revealed itself to be unreliable’.”

This Schwartz, by the way, was no rookie straight out of law school. He has practiced law in the snakepit that is New York for three decades. But he had, apparently, never used ChatGPT before, and, “therefore, was unaware of the possibility that its content could be false.” He had even asked the program to verify that the cases were real, and it had said “yes.” Aw, shucks.

Incredible. And 100% true.

You may think I’m crazy for using it by now!

But before we proceed, let’s ensure we understand what we’re discussing.

Defining the Terms.

What is “generative artificial intelligence?”

Generative artificial intelligence is a class of AI models that generate new content from scratch. This content comes in many forms, including text, images, music, and other creative outputs.

The basis of these models is to learn patterns in data and use those patterns to produce new, unique data similar to the training data. For example, a generative AI trained in paintings could create a new picture that, while unique, embodies the styles and techniques learned from the training data.

One type of generative AI is Generative Adversarial Networks (GANs), which consist of two neural networks (a generator and a discriminator) contesting with each other in a game-theoretic framework.

Another is an autoregressive model like the GPT series, which generates outputs of one part (one word at a time for text) based on what it has already generated. (This is the one that got that lawyer in trouble.)

Do I Trust It?

In a word, “No.”

Then why do I use it?

Because it creates a taxonomy I can follow to give a fairly complete treatment of a subject in a short amount of words (like a Rude piece).

I do this to ensure I don’t miss any significant part of the subject, though there’s no guarantee I won’t.

So How Do I Use It?

Though untrustworthy, I still think learning how to use AI properly is a superpower.

Let me take you through my mornings to illustrate my point.

First, I wake, shower, and walk Micah to school. Fairly standard.

After I drop Micah off, I walk around the block to my buddy Fabrizio’s Civico Cinque café. They serve the best coffee in all of Asti.

Vittorino, Sara, and Fabrizio
Source: Civico 5, via Facebook

I’ll sit there for about an hour, researching for the day’s Rude.

I read The Journal, Zero Hedge, the Rude Mailbag, and my Twitter feed and hope to get an idea quickly.

If it’s a bigger idea than usual, I’ve got to make sure I cover at least most of the bases.

That’s where ChatGPT comes in.

I haven’t written about money laundering yet, so let’s use that as an example.

I asked Chat GPT, “How would you structure a 1,250-word article on money laundering?”

This is what it returned:

Title: Unveiling the Veil: Understanding the Complexities of Money Laundering Introduction (approx. 150 words)
I.. Understanding Money Laundering (approx. 250 words)
II. The Stages of Money Laundering (approx. 350 words)
III. Methods and Techniques (approx. 400 words)
Combating Money Laundering (approx. 300 words)
Conclusion (approx. 100 words)

 Note: The word counts provided are approximate guidelines, and the actual word count may vary based on the specific content and writing style used.

Though I didn’t print it here, it also gave me four to five points under each heading to cover.

Of course, I can also ask further questions to define each point.

What I don’t ask ChatGPT for are facts.

I learned this the hard way with my pieces on Russia’s Resources. But luckily, unlike our unfortunate lawyer above, I knew the numbers ChatGPT was spewing out were bullshit.

So I had to return to the United States Geological Survey’s numbers to ensure I was printing the correct numbers.

It was a great lesson. From ChatGPT, get the map, not the details.

Wrap Up

You can still use models if you know what’s wrong with them.

We know ChatGPT makes up stuff if it doesn’t know the answer.

So never approach it blindly. Use it for subjects you already know or, though we didn’t cover it, automation.

If you’ve got time and notice you’re doing things you don’t want to do repeatedly, I recommend you watch some great YouTube videos to learn.

Here are some I enjoyed and learned from:

How To Use ChatGPT To Learn ANY Skill Quickly (Tutorial)

How to learn to code FAST using ChatGPT (it's a game changer seriously)

I Asked ChatGPT To Make Me As Much Money As Possible

Enjoy going down that rabbit hole!