The Large Language Model Threatening Public Education Today

Good day, CharterFolk.

There’s lots of talk happening right now about this sign-of-the-times article that was published by the New York Times this week.

Publicly funded schools with as little regulation as we have seen in generations …

Anyone can open a microschool, although more than two-thirds of founders are current or formerly licensed teachers. And these schools can teach anything they like, including biblical versions of science and history. Facilities may not be inspected; staff member background checks are sometimes unnecessary. And while many microschool founders say they cater to students with disabilities, the programs do not have to follow federal disability law, and most do not provide the therapies and counseling that are often available in public schools.

… are growing like crazy.

The article was timed to coincide with EdChoice’s announcement this week that a million students across the country are now accessing private school options through publicly funded vouchers and ESAs.

Meanwhile, another sign-of-the-times piece came out this week at CNN.

Interviewed by Fareed Zakaria, Sal Khan maintained his generally upbeat assessment of AI’s ability to make education better.

Now we can use AI to get … that much closer to a real tutor for every student and a teacher assistant … [for every teacher].

Zakaria was decidedly less sanguine.

How do you deal with the problem that everybody I know in education is worrying about, which is: will students actually do the work if the AI can do the work for them? How do you devise tests in a world where the students can just get the AI to write the paper for them or give the answer?

It’s a sentiment that is gaining currency across the world:

The idea that Large Language Models present a threat to education as we know it.

The article about microschools at the top of this column leads me to concur.

Large Language Models are, indeed, an immense threat.

But not in the ChatGPT sense.

In fact, ChatGPT helped me uncover the true threat to public education coming from Large Language Models.

This week when I asked ChatGPT to estimate the number of words in War and Peace, it had no problem.

“War and Peace” by Leo Tolstoy has approximately 587,000 words, making it one of the longest novels ever written.

Nor did it have a problem with the multiple volumes of Das Kapital.

The word count for “Das Kapital” by Karl Marx varies depending on the edition and translation. The first volume alone typically contains around 230,000 to 250,000 words. If you consider all three volumes, the total word count can exceed 700,000 words.

Or the multiple sections of a typical Sunday edition of the New York Times from the 1980’s.

A Sunday edition likely had around 250,000 to 300,000 words, considering multiple sections such as news, editorials, features, and ads.

Even when estimates exceed a million words, ChatGPT has no problem.

“Remembrance of Things Past,” also known as “In Search of Lost Time” by Marcel Proust, is one of the longest novels ever written. It contains approximately 1.2 million words across its seven volumes.

Or, fact, tens of millions of words.

The Federal Register varies significantly in length, but it is generally very extensive. On average, the Federal Register publishes around 80,000 to 90,000 pages annually. Given that a typical page contains about 500 words, the annual word count is roughly:

80,000 pages × 500 words/page = 40,000,000 words 90,000 pages × 500 words/page = 45,000,000 words

So, annually, the Federal Register could contain between 40 to 45 million words.

But when I asked for an estimate of the number of words in the California Education Code, what did ChatGPT reveal?

That it had a problem.

First there was evasion.

To estimate the number of words in the California Education Code, you can:

Use Online Text Analyzers: Copy and paste sections of the code into an online word counter tool.
Download the Document: If available as a PDF or text file, you can use document processing software to get a word count.
Use Programming Tools: Write a script in Python or another programming language to scrape and count the words from the online text.

Then there was buck-passing.

For a detailed section, you can manually copy parts into a word counter tool for a rough estimate.

And ultimately, it was total surrender.

For specific word counts or details, please visit Casetext – California Education Code, Title 1.

Apparently, the task is just too vast.

Too sprawling.

Too split up across multiple repositories.

Such that, in the eyes of ChatGPT anyway …

… yes, CharterFolk …

… the size of California Ed Code is literally inestimable.

And that doesn’t take into account the words contained in California regulations.

From a national standpoint, neither does it contemplate what we find in other states’ education codes and regulations. Never mind federal statute and regulation.

The number of words contained in those documents would equate to … what?

“Inestimable squared?”

This week, some of you may have seen that a federal judge blocked the implementation …

… of the new Title IX regulations that the Biden Administration issued earlier this spring.

I asked ChatGPT if it could estimate the number of words in the new regulations, but I only got back more surrender.

So I followed its instructions for estimating the inestimable.

I downloaded the pdf and then converted it into a word document. And then I used the word count function.

What came back?

For those of you who struggle with fine print:

That’s right. Over 460,000 words.

Just a couple chapters short of War and Peace.

For regulations covering one relatively small section of federal statute pertaining to public education.

Promulgated at a moment when what parents want …

… is virtually none.

CharterFolk.

There is definitely a “Large Language Model” that threatens all of public education today.

It is the model that imposes so large an amount of restrictive language on schools that they can’t get done the simple things that parents want.

And it’s feeding the abandonment of public education we are seeing unfold before our very eyes.

So if anyone asks you what is the harm in having a Large Language Model of regulating public education in our country today, you tell them:

It is inestimable.