Right now, circular financing and AI bubble takes are a dime a dozen. These may prove true, but be warned:

“Markets can remain irrational longer than you can remain solvent.”
― John Maynard Keynes

I think another kind of circle is forming: The Weaponized Spam Circle.

Thesis⌗

LLMs are great at commoditized tasks because it lowers the cost to produce them. People love this so much they leverage them for non-commoditized tasks, in doing so, they shift the burden of actually thinking to the reader. A reader will be overwhelmed by the quantity of generated text and require LLM assistance to ingest/summarize it, resorting to LLM assistance to respond.

It’s Brandolini’s Law but instead of refuting the bullshit, the bullshit is shot back at the bullshitter and the cycle begins anew.

In adding another layer to how we communicate with one another, we shell out big $$$ to afford the matrix multiplication to do so.

But what is a commodity?⌗

Let’s do the wikipedia cliche:

“In economics, a commodity is an economic good, usually a resource, that specifically has full or substantial fungibility: that is, the market treats instances of the good as equivalent or nearly so with no regard to who produced them.”

“With no regard for who produced them…” will be the thread I hold throughout the rest of this piece.

Computers (neural networks trained on millions of positions) are better than humans at Chess. Humans prefer to watch humans play chess and regard using a computer as cheating.

  • This means playing chess IS NOT a commodity, we CARE who plays the chess

Do people check where the gas is coming from… do they deep dive the exact source of the energy being used to charge their Tesla? Pretty much no… if you want to get pissed here and tell me how you’d never use Saudi oil and you check at every gas station be my guest.

  • This means filling up/charging our cars IS a commodity, we DON’T CARE how it happens… we want it as cheaply/quickly as possible.

But writing... is it a commodity?

When Characters are a Commodity⌗

Not all writing is thoughtful. There’s tedious writing marked by “character limits”, extrinsically imposed deadlines, random boiler plate, documenting a new feature, emails to insurance companies, and the gut wrenching feeling of having to start putting words down for a book report.

In these moments, an LLM can shoot through character limits and fake decorum to produce words that felt far away. With a bit of editing, large writing burdens are reduced to a slight inconvenience.

But we’ve already made a big mistake… if you’re nodding your head at the two paragraphs above, read them again.

Expand when you've formed an opinion

I threw “book report” in there with the rest of those “examples”. A book report is not a commodity. If you’re writing a book report you’re most likely learning the basics of writing. You’re learning the fundamentals of how to take a position on something and defend it. If that is a commodity, then I don’t know what is. The thinking is the important piece and the author wielding their brain is what matters, the physical words are a by-product used for other humans to interpret your thoughts.

Also what about: “documentation” and “extrinsically imposed deadlines”. I can spot issues there as well.

Sometimes it’s not easy to determine what IS and what IS NOT a commodity when it comes to writing.

Sometimes it’s too easy to label some writing as a commodity and shovel it off to an LLM.

Sometimes it’s not easy to think, thinking is hard, new ideas are hard.

When Characters are NOT a commodity⌗

“Writing is thinking. To write well is to think clearly. That’s why it’s so hard." ― David McCullough

At Amazon, there is the notorious six pager.

  • “When you have to write your ideas out in complete sentences… in complete paragraphs it forces a deeper clarity of thinking.”

  • “The author who has put a tremendous amount of work into writing the memo gets to see everyone read the memo.”

A six pager is NOT a commodity. The moment it becomes a commodity is the moment it becomes no better than a powerpoint. The people who wrote the document should care about it, should have thought deeply about its contents, and those reading the document digest it out of that same thoughtfulness.

Yet LLMs have quickly crept into this space, because writing a six pager is hard, it requires thinking, and plopping that burden on something else is just a few key strokes away. The writing is on the wall:

  • Python 3.7 on Lambda (LLM Knowledge cut off)
  • “its not just a ___ its a ___”
  • cascades of bullet points
  • Hallucinated information that looks ok but is wrong after a bit of thought

Wait, are they treating this document like a commodity?

Did they take the time think through the key ideas?

Did this person even read what they wrote?

Do they even care about my time?

…I’ll respond in kind.

That “six pager” can be done so FAST because nobody is THINKING ABOUT IT. The LLM takes a unique piece of articulate thinking and turns it into a commodity. How do we respond to a commodity? Drive down the price! If they chose not to think why should I chose to think. And……… we have our circle.

Output as a result of input⌗

I could care less about the output of the hallucinated + verbose document. Show me the input, show me the meat of what you are trying to convey. If there was no or sparse input to begin with… then there was no or sparse thought.

Is adding fluff to things holding us back? Is this why people reach to LLMs to puff up their ideas? If there is an important point to make then lets start there, but we don’t need to play telephone back and forth via LLM.

Why do we teach assembly⌗

Many undergraduate curriculums teach assembly. Most undergraduate students will never earn a $ or spend a minute of their free time writing assembly. Why teach it? Because it helps provide insight into how computers work.

Why take the time to develop an idea and write it down? Because it forces you to think critically about the opinions and about the facts.

Doing this because we can’t help ourselves⌗

There are many practical use cases for LLMs. It is 100% clear that they are beneficial, especially if you take the time to give them the tools, context, and guidelines they need to succeed. Just like a human, if you give an LLM limited context and tell it to do something it will probably suck at it. LLMs are NOT humans, it’s math, and if the necessary inputs aren’t provided the equation won’t be solvable.

Just like with any beneficial thing, humans are going to exploit it for their gain. It’s fundamental to how we operate. This massive boon in LLMs feels like a free lunch… it feels like something to be exploited, yet it remains a balancing act.

  • Sure you can generate a bunch of text and throw at someone, but they can do the same to you… we are back to square one.

  • Sure you can generate 100,000 lines of code, but now you need to maintain, refactor, and change one hundred thousand lines of code.

  • Sure you can automate that PMs flow where they aggregate a bunch of stuff from Asana, Excel and your ticketing system into a general summary, but why the fuck are we tracking things across 6 different application in the first place.

  • Sure you can automate your Linkedin post, but it’s just getting gobbled up by bots rolling around in the absolute slop trough of their own creation. If you want a peek behind the curtain and see what the dead internet will look like, scroll around there for 5 minutes.

Other Random Thoughts⌗

Do we really want people growing up and learning how to read/write via LLM? The pattern’s they’ll learn will converge us towards the style of LLM writing that people find so obviously annoying today. Will the humans have the ability to recognize that and AVOID that style?

You have a brain, you can still think, brains are powerful… use it.

Writing without thought distills an entire method of thinking into something transactional, input output. Don’t ask how it works it just does.

For kids, writing is the new math.

In regard to math:

  • When will I actually use this when I grow up?
  • I have a calculator I don’t need to learn this.

Now, in regard to reading/writing:

  • I have an LLM I don’t need to write this.
  • What’s CliffsNotes? I just use an LLM

For adults, the same super powers you can use to generate a lot of text can be used against you. Most adults will be using their single LLM chat session to compete against an entire layer of corporate LLMs. Let the words be fired back and forth and to the datacenter owner’s, chip manufactures, and model trainers go the spoils.

Whenever I leverage an LLM to write I ask a flavor of these questions:

  • Would the reader or any of the reader(s) consuming this content care if it was produced by an LLM?
  • Would I personally care if the content under my name was authored by an LLM?
  • Have I provided the proper steering, guidelines, and context for the model to be successful?
  • Have I read what the model has produced?