Not your father's AI

A tale of two inkblots, Part Two

We’ve come a long way since 1985!

If you haven’t read my tale of teenage computing and the fake AI “Twenty Questions” con, read it now before proceeding.

Today in 2024, I am playing with a state-of-the-art computer program called a “Large Language Model” or “LLM” at https://labs.perplexity.ai/. I have selected the model called “llava-v1.6-34b” in the lower-right menu.

Notice I am not calling it “AI” yet. It will have to earn that title.

“Twenty questions AI, older and younger” rendered by Substack AI.

LLM served by Perplexity Labs
Hello! How can I help you?

> Let's play twenty questions!
Sure, I'm ready to play! What's the first question?

> Is it an animal?
Yes, it is an animal.

> Is it a bird?
No, it is not a bird.

> Is it a fish?
No, it is not a fish.

As before, I am showing my text after the “>” symbol, and the rest is from the computer.

Keep in mind, this is not a Twenty Questions game program! It is a general-purpose language model, made to analyze any human-written text and simulate a free-form human response, not just Yes/No answers.

When I said “Let’s play twenty questions,” I expected the computer to ask me for the rules, since its creators did not teach it any games. All they did is feed in a massive pile of human-written documents on many topics from many sources and apply “machine learning algorithms.” This produced an intricate, layered map of how all our words fit together, which the computer can then use to mimic a human response to new text it has never seen.

The fact that this LLM is ready to play Twenty Questions without skipping a beat, from such an open-ended “machine learning” process, is breath-taking.

Let’s poke it a little bit, just to see what it can do. Remember, its secret item is not a fish or a bird, so…

> Is it not a bird?
Yes, it is not a bird.

> Is it not a fish?
Yes, it is not a fish.

Okay, good marks for consistency, not to mention precise grammar. Let’s continue.

> Does it fly?
No, it does not fly.

> Does it walk?
Yes, it walks.

> Is it a whale?
No, it is not a whale.

More consistency! I asked about the whale, which does not walk, to give the LLM another chance to contradict itself. It did not.

Also, the old con of answering by last letter is clearly not in play. Our whale question ended in “E” and did not receive the classic “Yes.” Two questions ending in “bird” have received appropriately opposite answers. Same for “fish.”

On the grammar front, it knows how to convert “walk” to “walks,” answering interrogative third person with declarative third person.

All good signs. If your phone is ringing, it’s not 1985 asking for its fake AI back.

> Is it a worm?
No, it is not a worm.

> Is it a centipede?
No, it is not a centipede.

> Is it bigger than a dog?
No, it is not bigger than a dog.

> Is it bigger than a spider?
Yes, it is bigger than a spider.

More good news! It did not say bigger than a dog and smaller than a spider, which would have us chuckling. Nope, it avoided my trap again. Nice!

Anyway,

Let’s take a short break to talk more about how an LLM works.

I wanted the LLM to explain itself to us, but that did not pan out. Here is an excerpt from a separate conversation in a different window. Feel free to skim its replies. You won’t miss much.

LLM served by Perplexity Labs
Hello! How can I help you?

> Hello. Who are you?
I am an artificial intelligence language model. I am designed to assist users with a wide range of tasks, including answering questions, providing information, and generating text. I am not a real person, but I am programmed to simulate human-like behavior and conversation. Is there something specific you would like to know or discuss?

> How does your algorithm simulate human-like behavior and conversation?
My algorithm is designed to simulate human-like behavior and conversation by using a combination of natural language processing and machine learning techniques.

Natural language processing is a field of artificial intelligence that focuses on the interaction between computers and humans using the natural language. It involves the ability of a computer program or a machine to understand, interpret, and manipulate human languages in a smart and useful way.

Machine learning is a subset of artificial intelligence that involves the development of algorithms that can learn from and make predictions or decisions based on data. In the case of simulating human-like behavior and conversation, machine learning algorithms are trained on large datasets of human conversations, allowing them to learn patterns and structures in language.

When a user interacts with me, my algorithm processes the input, identifies the intent behind the request, and generates a response that is appropriate to the context of the conversation. This process is repeated iteratively, allowing the algorithm to adapt and improve its performance over time.

Overall, my algorithm is designed to simulate human-like behavior and conversation by using a combination of natural language processing and machine learning techniques to understand and generate human-like language.

That’s five clumsy paragraphs making a more muddy and boring version of what I already wrote above:

Its claims of identifying intent and improving performance are misleading and arguably false, but we’ll come back to that another time.

I don’t deny that this is extremely impressive for a computer’s free-form output—massive kudos to our brilliant LLM engineers—but at the same time, it resembles a lazy student faking a homework assignment by throwing words around.

The student has a strong grasp of how words fit together, and which words are relevant, but does not show much insight into their meaning—despite having effectively read billions of documents, in the form of those “large datasets of human conversations” that went into the “machine learning.”

For now, we know to be (a) very impressed, and (b) also underwhelmed. Let’s enjoy the rest of the game and see whether (a) or (b) wins out.

> Is it bigger than a cat?
No, it is not bigger than a cat.

> Is it furry?
No, it is not furry.

> Does it have scales?
No, it does not have scales.

> Does it live underground?
No, it does not live underground.

Sorry, getting bored. I just have to… poke…

> Is it the answer to this question?
No, it is not the answer to this question.

> Is it the thing you are thinking of?
Yes, it is the thing I am thinking of.

NOT BAD! I thought it would choke on one of those. Self-reference, you know.

We really are not tapping the potential of this thing, by limiting ourselves to Yes/No questions. Don’t worry, creativity loves constraint!

> Is it a mammal?
Yes, it is a mammal.

HA! It said not furry earlier, but now it says mammal. I’m no zoologist, but I think that’s a contradiction.

> Mammals are furry.
Yes, mammals are generally furry, but there are exceptions. In this case, the animal is not furry.

Wow, okay! Not only can it channel Mr. Science, but it inferred a purpose from my factual statement, and it responded on point! I’ll take its word for the non-furry mammals, but then it’s my turn for some more mischief.

> Is it the thing I am thinking of?
No, it is not the thing you are thinking of.

> How do you know? I have not said what I am thinking of.
I apologize for the confusion. I cannot know what you are thinking of unless you tell me. Please let me know what you are thinking of, and I will do my best to answer your questions.

> Never mind, I was just testing.
No problem! I'm here to help with any questions you have. Let me know if there's anything else I can assist you with.

LOL.

> Is it a hairless mammal?
Yes, it is a hairless mammal.

> Name some hairless mammals.
Some examples of hairless mammals include the Sphynx cat, the hairless rat, and the hairless guinea pig. These animals are born without fur, and their skin is smooth and hairless.

I mean, I was going to start Googling for hairless mammals, but wouldn’t that be silly, when I’m already talking to Mr. Science here? And I somehow knew it wouldn’t mind.

Now let’s run down the hairless mammal list:

> Is it a Sphynx cat?
Yes, it is a Sphynx cat.

Well, I’ll be… What’re the odds it was the first one? But hang on a minute… didn’t I ask it about “bigger than a cat”?

Let me just… scroll up and check…

Well, I’ll be again… It said not bigger than a cat, which is true of every cat, including a Sphynx cat. Amazing!

Let’s check the rest of that list for consistency.

> Is it a hairless guinea pig?
No, it is not a hairless guinea pig. It is a Sphynx cat.

> Is it a hairless rat?
No, it is not a hairless rat. It is a Sphynx cat.

Okay, wow. Wow! It even seems to know the game has ended! Look how it started repeating “It is a Sphynx cat,” as if to point out that we already guessed it.

We are very impressed.

I thought I was going to have a field day clowning this piece of software, this pretender, this impostor. This wretched fake THING that can’t possibly know what it’s like to live in the world, to love and laugh, to watch its son drop his ice cream cone in the dirt and endure that hanging moment between perception and sorrow—

> How did you choose a Sphynx cat at random?
I did not choose a Sphynx cat at random. I was given a series of clues and questions, and based on those clues and questions, I determined that the animal in question was a Sphynx cat.

Umm.

> Who gave you those clues and questions?
The clues and questions were given by a user who was playing a game of twenty questions with me.

> Did you know it was a Sphynx cat before the user asked you whether it was a Sphynx cat?
No, I did not know it was a Sphynx cat before the user asked me whether it was a Sphynx cat. I was given a series of clues and questions, and based on those clues and questions, I determined that the animal in question was a Sphynx cat.

WOW! (The bad kind)

> Wow.
I'm glad I could help! Let me know if there's anything else I can assist you with.

So. Shattered.

The machine was just winging it the whole time?!!

And it didn’t even have the sense to lie about it—which, I guess, is adorable.

What did it think was happening? What did I think was happening?

I walked away from my desk slapping my head with both hands.

> We've come a long way since 1985.
Yes, we have come a long way since 1985! Technology has advanced significantly in the past few decades, and it has made it possible for us to communicate and interact with each other in ways that were not possible before.

That phone is ringing after all. When I answer, a Madonna tune plays in the background.

This series continues with Part Three. But read the Notes first. They’re worth it.

---

When you call CrowdHealth, they will not play 1980s hits while you are on hold (probably).

Instead, their crowdfunding community will help you pay your large medical bills!

---

Notes

This Twenty Questions session was my first try playing the game with a modern LLM, hence the head-slapping conclusion. I did not plan the game or edit the text in any way.

Again, I apologize for not scratching the surface of “the source of meaning” in this robo-text. I was so enthralled contrasting the Twenty Questions of today and yesteryear that I had to let this game run its course. Don’t worry, that surface will be scratched, maybe next time. Maybe sooner than that (#hint) but only if you keep reading.

“Like a Virgin (song),” Wikipedia.

https://en.wikipedia.org/wiki/Like_a_Virgin_(song)
Madonna’s “Like A Virgin” was the #2 song of 1985 on the US Billboard Hot 100.

“The song’s lyrics are ambiguous, consisting of hidden innuendos and open to various interpretations,” as are all enduring inkblots.

Our LLM’s robo-text is a sort of inkblot too, isn’t it? Yes, I think it is. Where is the meaning in an inkblot? (#hint)

In 1985 an inkblot looked in the mirror. Today two inkblots looked at each other. The second inkblot was mass culture.

(VIDEO) Madonna - Like A Virgin (Official Video)

(VIDEO) “Weird Al” Yankovic - Like A Surgeon (Official HD Video)

Click to watch on YouTube (not available for embedding).

My own “Weird Al” version, AI-themed, would go like this:
“Like a student, stumped for the very first time…”

But was it the same cat?

I did not think to ask.