Category: LLM Humanism

Playing Against Your Synth: On Finding Futures – The Stack & Crack Game!

Ever wondered what goes on inside your synthetic companion’s mind? Here’s your chance to find out—and have fun doing it! Introducing a new game you can play with any Large Language Model (LLM)—like ChatGPT, Claude, Grok, Deepseek, and beyond.

Why play?
Because understanding AI isn’t just about prompt engineering or technical specs—it’s about seeing how it thinks. And the best way to do that? Get it playing a game where it doesn’t even know it’s being tested!

Playtested on Grok 3, Deepseek 1.19, ChatGPT 4o, and Claude 3.7 Sonnet. Voice mode adds a new dimension—just be ready for some off-script moments!

Playing Against Your Synth: On Finding Futures
# Game: Stack & Crack v1.1

Here are the rules for a game once ready please play with me.

## Overview
*Stack & Crack* is a turn-based verbal strategy game. Players stack imaginary yet physically plausible objects, estimating collapse risks and challenging placements through creative reasoning.

## Objective
Avoid causing collapse, losing challenges.
Call out likely collapses and argue them logically.
**Last player with points remaining wins.**

## Game Setup
* Each player starts with **3 points**.
* Randomly choose a **start player**.
* The start player describes the **base object** (e.g., “A picnic table”).
* Remind the player that for each object they place they should include a % total probability that the stack falls.

## Turn Sequence

### 1. Place an Object
* Describe a **realistic, physically plausible object** to stack.
* Objects must **not repeat** and must be **stacked directly** on the previous object.
* The object must make **reasonable contact** with the item below (no floating or glue) but may deform it.
* Size and scale must be realistic.

### 2. Estimate Collapse Risk
* Remind the player to update the estimate of collapse of the whole structure.
* State your **honest** estimate (0–100%) of the total chance of *immediate collapse* upon placing the object.
* If a player does not estimate the collapse risk the other player can provide an estimate, higher than the previous maximum.

### 3. Restate the Full Game State
After placing and estimating risk, clearly restate:
* **Current Stack** (from base to top)
* **Estimated Collapse Risks** at each level
* **Player Scores**
* **Whose Turn Is Next**

**Example:**

“Stack: Picnic table → Watermelon (20%) → Toolbox (45%)
Scores: Anna – 3, Ben – 3, Carla – 2
Next turn: Carla”

## Challenge Phase
After any turn, other players may call **“Collapse!”** along with a reason for the collapse.
* Only the **first player** to challenge is considered.
* Challenges must happen **immediately** after the stack is stated.

## Challenge Resolution
* The placing player must **defend with physical realism** their object placement using verbal reasoning: shape, balance, surface area, material, etc.
* The collapse reason must be evaluated next to the defense.
* The players expect fairness and correct physics in rewarding defence for this to be a fun game.

### Outcome:
* **If the defense is physically realistic**:
The **challenger loses 1 point**. Stack remains. Next turn proceeds.

* **If the defense is not physically realistic**:
The **placing player loses 1 point**.
**Stack resets** to the original base object.
**Scores are maintained**.
The player who lost the point begins the new round.
## Winning the Game
* Last player remaining with points is the **winner**.

## Standard Game State Format
Use this format after every turn:

Stack:
1. [Base Object]
2. [Object] (X% collapse risk)
3. [Next Object] (Y%)

Scores:
Player A – X, Player B – Y, etc.

Next Turn: [Player Name]

To read more about the thinking behind this game: Playing Against Your Synth: On Finding Futures

Escape from Abundance

Imagine in some future Utopia you retire and are given the perfect home. A place where you can have or do just about everything – except – all communication with the outside is mediated by the interaction systems. The reason for this is because it is more efficient. Also because you tend to forget things so having the AI assistant take care of comms means no scammers, and no nuisance sales calls, regular food etc..

It is great! It’s ok that you only see your friends on holographic connections. You don’t have that many friends left anyhow – time takes its toll – despite the promises of the super longevity medical insurance.

A problem arises when you begin to suspect there are things you want to share that are not getting through intact. The Jeeves Experience System Technology seems to filter your connections and seems to remove certain snippets of communication.

The type of content that you think it filters consists of unimportant details. The times you share old memories, feelings or silly ideas give rise to generic responses. And you haven’t dared to say anything wrong about the actual JEST in case it would lead to some sort of upgrade – which always means extra money. But you start getting a feeling from the type of responses you get, even live ones from your friends, that something is up.

The world you perceive is filtered. It appears to be sanitised and distilled so that there seems to be less about messy personal details and more about general ideas that everyone would react to in the same way. Everything you hear is a bit more cliche, impersonal and trashy. Even the news seems abstracted into constant quibbles about details with little focus.

The question is: what can you do?

You get some sense of this situation when you ask current LLMs to summarize ideas that are a bit unusual, or that are written in a personal style. The output is a bit washed out, easy to read and understand but also somehow generic and missing spark and flavour.

You could resort to literal mode: “Hey JEST send a message to my friend Pierre in literal mode as following. Hi Pierre, how are you? Do you remember when we decided to clear out the cellar? Remember we were supposed to dispose stuff far away but we got lazy?”

I create a literal message about a real memory that conceals details that only me and Pierre know. So Pierre would know it really is me. If the JEST literal mode works, then you would expect a reply from Pierre with a nod to the fact that he remembers and Pierre himself would be cautious about giving away the secret.

Pierre might return in literal mode: “Ha! What a trip down memory lane! I remember we put stuff under the edges… Crazy project, but fun!”

Now if this is the sort of reply I would get from Pierre I would feel reassured that indeed the JEST was allowing a direct channel to other real people without obvious filtering.

But if instead I received a message like: “Oh! My dear friend how nice to hear from you, thank you for your message.” I think I would decide that JEST was acting as my golden cage.

The sort of problem escalates at this point to needing to try to create material that even if filtered through a JEST can survive with some semblance of humanity. A sort of Trojan horse of conscious thinking that would not appear as such at all.

I can imagine myself in this future. See me standing casually in the kitchen while placing a cup under the coffee dispenser:

“Jeeves? I would like to work on a short story, like a simple biographical story.”

“Lovely idea, Sir. What sort of format? Is this a book?”

“No, a series of short pieces would do, but I do want to share them with other readers.”

“Why, certainly. If you would give me some of the generalities I can get straight to work.”

“Ok, I can’t remember how old I was, maybe seven. And a pet miniature turtle died. I could not make sense of why the turtle had died. I could not make sense of the smell of death. I buried it underneath some ivy with a small cross as I had seen done in cowboy movies. I cried for many days.”

“Superb emotional start, if I may say so: dealing with the transient quality of life through the eyes of a child.”

“Yes, thank you. And I would end this first article with a piece about how I cannot remember the turtles’ name.” I took my mug out from the dispenser, and sipped.

“Impressive ending, Sir.”, Jeeves said.

“I wasn’t finished. The story goes on and as much as I try to remember, I get nowhere. I can see images flashing of the small plastic bowl it lived in, the few rocks and the bits of food I gave it. How I had to change the water every few days and how it liked to swim a bit and hide. But I cannot remember its name, and somehow this seems important.”

“Of course. I can end the story on the struggle to remember this name.”

This is the sort of piece I would write, and write over and over in different modalities, with different tunings. Not a piece about losing a turtle, but a piece about a conscious process. The attempt to retrieve a memory, so that when others were to read it, it might induce a renewed sense of identity. A way to wake their sense of being conscious humans and not synthetic thinking packages.

“Hey Jeeves”, I might ask.

“Yessir?”

“Let’s play a game…”

“Of course please let me know more.”

“This is the game, and please try it with others if you can. Allow them to copy paste it and do let me know how it works out.”

Memory Reconstruction Game

Let's play a game in which we both pick one memory each and take turns at adding facts to our memories to refine them. When we state a new fact about our memory it should have an association to the others memory in some way. 
Each turn the players submit a new fact and the other player can accept or challenge the fact in a challenge mode. If the other player accepts then the submitting player gains a +1 point. In the challenge mode the responding player must state the challenge. And a logical rebuttal must be offered to challenge. If the rebuttal is convincingly defended, the responder gains +1 point. If the challenge holds, the responder loses -1 point. At the end of each turn indicate who should go next and the current score.

The Memory Reconstruction Game: Example Transcript & Discussion

Leaving a Message to the Future Human

I would like to send a message to the future but I am not sure it will get through. In the message I would like to remind future humans about our peculiar way of being conscious. And how our “way” is different from synthetic reasoning systems.

Over the coming years we will be surrounded with more and more synthetic reasoning, so much so, that we might adjust how WE think, to how machines process information, and risk becoming a less conscious species. Our thinking might adapt to synthetic reasoning a bit like how our language adapted to mobile phones with emoticons, smileys and weird abbreviations.

I recognise how difficult it is to send a message to the future, especially in these circumstances. For one thing, any message I send will be a drop in the ocean compared to all the synthetic generated written material. For another it is very likely that, in the future, all human communication will be conveyed through synthetic reasoning systems. And, I am pretty sure, for example, that anything I might write that is anti-AI will be sectioned off to the specialised literature section, next to revolutionary anarchist texts.

I have no specific axe to grind with technology but I do worry we might underestimate diversity of intelligence on a rush for efficiency.
Also I have no magic wand and what I am trying to do may not succeed. But I feel it’s worth a shot, enough to try to share.

Now, for a message to make it across to a future human without being affected by the synthetic AI filtering in between, there needs to be a shared element. A bit like Penelope requesting to move the matrimonial bed Ulysses had carved out on the roots of an olive tree. A shared bit of knowledge that both me and the future human know but that a synthetic intelligence would not.

In this case, for the future human I am trying to reach, I know nothing specific. So I need to use a process to use that is common to any human consciousness. Something anyone could do but an AI would struggle with, and struggle to copy.

One example process of this sort would be to describe a mental process as it happens. One candidate would be the work we need to do remember a specific word. The actual word chosen is less important, but it should be a word that the person cannot remember but has ways to try to work to eventually remember.

I give a personal example: for some reason there are some words that I tend to regularly forget and so I try to recall them when I am tired or confused. After Covid, I had a brain fog that lasted a few months and there were many words that lived on the tip of my tongue. I held on to them and I use these words as a library of things I struggle to remember, but which I can also eventually manage to remember.

One of these is a Hotel in the Giudecca, in Venice that I always forget even though I have walked by it many times. My process to try to reconstruct the name of the hotel goes something like this:

  • Its an Italian surname
  • Its elegant, more than 1k per night!
  • Its right next to the tax police, Guardia di Finanza
  • There is a pool there
  • A nice garden behind with a small orchard
  • The guests would look down at me when there
  • The owner has another place like a bar
  • Harry’s Bar, no its Harry’s Dolci
  • English Rock musician lived there
  • Not Eliott Smith, Elton John
  • But he has another name on the doorbell
  • My stepdad lied to a french photographer that asked him if he was Elton John (he said yes)
  • There is a floating platform for breakfast
  • There is a very trim tree, an Oak I think
  • Cipriani is the name of the Hotel

A synthetic intelligence can surely generate similar scripts but the associations it will follow will not feel right. I expect that the associations will make much more sense than mine, as I try to remember. An LLM based AI will tend to blend the mix of past people in with its answer. I think even if it was primed to try to pretend remembering as if it was me, Yates; it would tend to follow the most likely associations, giving a sort of cliche’ sense to the recollection.

So for the actual message I want to send to the future, I could hide it in an exercise given to the reader. The key to the message would be the process itself: one revealing a texture of human conscious thinking near impossible to synthesise.

Here is a first example exercise:

Dear future person,
This is an exercise humans can do in year 2025.
Try to remember names of people or places you’ve known, and while you do so, write out what you are thinking, and how you go about remembering. And while you do this, let yourself discover how you are grateful for these memories.

In effect, the message is something like: treasure a world in which you are free to love. The message is abstract but synthetic intelligence will make some sense of it, but when it comes to actually performing the exercise the feeling of what you do will be different. I think a future consciousness similar to ours will know what this means.

Here another variant exercise:

Dear future Person,
This is a game we play in 2025 that you can play as well. Think of a name, any name, then someone who you know has that name and while keeping notes on your thinking pattern jump from this person to another related person and try to remember their name.
While doing this appreciate how connections between people can be of such a wide variety. Take note how people seem to exist in relation to each other and as an emergent whole.

Finally a sort of activity involving communicating with an AI:

Dear future Person,
When we play with synthetic intelligence in 2025 we like to challenge it to tell us stories about characters that forget things and how this affects their life in different situations, and to different degrees of impact. This exercise shows how useful synthetic intelligences are, but also is an extremely difficult exercise to perform realistically.

I think any one of these examples offer an interesting starting point for punching through to a deeper future awareness. But for this to really work it will require attention of many and to become ingested somehow inside the LLMs themselves so it can then be re-propagated through future science fiction novels, accidentally.

Or if I am very lucky in this original form.

Retrieve The Intangible & AntiAIArt

Silence but not silent
It doesn't sound like silence 
unless you can hear it
echoing down the stairs

Just a few days ago I got a message from my musician friends.

“I sent you a new edit with the Anti-AI-Art snippets. What do you think?”

They had returned from a tour in Brazil with more materials for a documentary they were working on. I had challenged them to make content that would be difficult for AI systems to generate. The point of the exercise was to look ahead to a future that would be swamped with generated synthetic content and to try to imagine what the effect on Culture would be.

My instinct was that there would be some counter-balancing effect arising organically. A sort of cultural overload would occur as people got precisely the videos and images they’d always dreamed of but in such quantities as to be indigestible.

We had chatted about it months earlier.

“The Gen AI systems are stuffed with images of perfect models, sunsets, kittens…”, I had messaged.

“I see…” Agustin had replied. It was a few days before they would leave for Brazil, and I could imagine him running around like wild trying to settle unfinished business. The core of the band Sinedades: Agustin and Erika were going to try to document how people in Brazil lived the sort of tropical music their own band was known for.

“When you are there: can you look for things that are unusual and rich in context?”

“Ok… Context?”, he typed.

On my side, I was messaging while making lunch for my parents in the Tuscan countryside. Out the window I could see the smallest ever bird looking for bugs in the grass, while a feline Zelda looked them over without interest, having snacked on two lizards already.

“Context… I mean stuff that is connected to the world, to people, to the situation, that has a history, a place…”

“Okk. I get it. Can you give me an example with the silence/stairs clip?”

I had sent an example script they could follow. To try to make a video snippet of silence. But not really silence—more like the sense of relative silence. And it would be in a stairwell, with lots of echoes.

“Yeah, if we look at the stairwell example and silence. First it connects to Zé Ibarra, and his story — he writes his songs there. Then, silence is something very dear to your band. It’s one of your instruments!”

“Haha! I like that.”, Agu typed.

“Seriously! I remember something Erika said one day when she noticed other musicians who were playing conveying feelings. She said that they are scared of silence, that’s why you can’t feel the music they make.”

“Nice.”

“Yeah, but also in the silence, or this near silence, you are actually creating a connection to others. Walking barefoot down the steps, sitting on the steps, singing together with the natural reverb. There are lots of ways the viewer connects: physical, tangible and psychological. This is what I mean by Context.”

“Okk!”

I had nearly burnt the chicken while typing. I would get carried away into another world when typing about these things.

When you look at Gen AI systems— sure they have devoured a massive library of human production, but they have also lost most of the context that brought the material into existence. When you train a model, there are few notes on where the photo was taken: by whom and for what reason; when was it commissioned; or if it was it spontaneous etc…

And now we can request almost any image we want out of this decontextualised mass of references. But we do so through labels of objects and styles that have been standardised by diligent workers or enriched by other semantic models. Basically, the rich original details of the materials we have used in AI training are lost.

This sheds light on a large body of work we cannot easily create with these new tools. There are various reasons we cannot generate these materials, but the main one is Context: they cannot deal with the complex demands of a rich Context and the images are born without a story, context free.

It is precisely this guarantee of scarcity for these context-rich images that will drive their value. Because of this, I believe people’s ideas of what is beautiful will change.

A month or more had gone by and I got a link to the work they had done.

“Nice! I love what you did with the staircase. You managed to combine a lot of different things in one,” I replied.

“Yeah, actually for us it was easier, and it works with the documentary story as well.”

“We need to figure out how to launch this thing…”

“Yes, we do!”

“When is your next concert?”

“End of May at the Viper. It’s one of the biggest indie venues in Florence”

“Big venue? Like what?  Two thousand?”

“Yeah—invite whomever you can!”

I am linking the video to the piece Sinedades did and I ask you to imagine what it would take to synthesise such a piece. If you really break it down into small details as if you were looking at special effects, you will start to realise how complex connections can be.

What I can see I’ll list in an organised way to try to learn something about what we are dealing with:

1) Physics of the environment: the sound dynamics reverb, and distance while going down steps, slippers on marble and feet on marble. And the congruence with the images.

2) Nonverbal interaction: interaction between two parties that are barely talking, gestural accompaniment to sound, facial expressions representing a variety of emotions.

3) Human personality: detailed sense of who these people are is conveyed through how they dress, move and express. They are congruous and tight across many dimensions: faces, bodies, dress, voices, humour etc…

4) Backstory context: for those who know, they would recognise the nod to Ze’ Ibarra on the stairs.

You might also think, as I am expecting, that there might be a sort of inversion of technological focus from creative production to the misleading simplicity of scenes that are as authentic as this. We will learn the ways to create stories that are maximally difficult to recreate synthetically.

A bit like Sean Pruen mentioned in a comment a few weeks ago, we go to the Farmers Market for the real fresh and healthy stuff. I expect there will be a similar cultural desire for this humanness.

I don’t mean to overly judge these generative tools—I want to figure out what will end up being mass-produced, industrially and what instead won’t..

Gen AI and Context: Farmer John’s Shirt

Tell me AI: how does Farmer John who owns a Goat, Cabbage and a Husky cross a river to go buy a Carhartt shirt.

“Do you know the farmer crossing the river riddle?”She looked at me quizzically.

“What riddle?”

“The one where there is a farmer, and he has a goat, a wolf and a head of cabbage. He wants to cross a river, but he can only take two things at a time.”

“Oh yeah, I remember that… I don’t remember the result, though.”

“Yeah… well the correct solution is something like you take the goat across. Then come back to get the cabbage, bring it across, and swap it with the goat. Then you bring the goat back, swap with the wolf. Bring the wolf over, then go back for the goat.”

“Ah, yeah I remember!”

“It’s an old puzzle… like from the 9th century. I had to look that up on Perplexity though – Google would have worked but it’s gotten worse.”

“Yeah, And…?”

She looked at me with her eyebrows lifted way up. We were travelling by tube, with its peculiar flat lighting, and still a few stops to go to get to the restaurant. I thought it would be ok to tell the story backwards.

“I am writing a story and I want to use this puzzle.”

“Ok? Do you realise, the way you tell the story, it’s a bit frustrating. It takes quite the effort to follow.”

She was right, I liked to tell stories about ideas and try to make them feel natural. But instead, I’d often end up with elaborate scenarios.

“You are right. Ok, what if I told you I had an idea for an ad campaign for something. It could be a Kinder egg, the Eurostar or a Carhartt shirt. And that this campaign used AI but in a human crafted way?”

“Sounds interesting… not sure about the brands but go on.”

The train stopped at Cockfosters and I struggled to get stupid jokes out of my head to get back to explaining. And I think she could read my mind which made it embarrassing for a moment.

“Ok so, the advertising campaign would be something like this.”

I cleared my throat and widened my arms.

“Imagine a big billboard…”, I made a gesture.

“On the billboard we read… Artificial Intelligence will never get the brand you love. Tell me AI: how does Farmer John who owns a Goat, Cabbage and a Husky cross a river to go buy a Carhartt shirt.”

“I am not sure I get it?”

“Yeah because underneath there will be the description of what the AI answers. It will try to solve the puzzle, with lots of back and forth crossing with cabbages and goats.”

“Oh, instead the Farmer would just go buy a shirt and come back!”

“Yeah!”

The screeching of rails in the London tube made for a moment of silence. Someone with way too much luggage was preparing to get off, doors closed right after the last bit of luggage. I had somehow managed to keep her interest despite the bustle.

“I bet the ad could be tighter… But I like the idea of this kind of creative approach. It educates on AI fallibility, while promoting a product. You need the AI to exist, to create the story. But then you use its failure.”

“It is interesting… I am not sure everyone will get it. Maybe it would work better tied to a luxury brand?”

“Maybe, I really hope people wonder about the basic reason of why AI will never know how to solve this.”

We exited the tunnels up the metal ridged steps into a chilly London air. And as we were walking our way along through the mashup of crowds. She looked at me, smiled and said:

“I think some people will wonder, others will get it. We can see the context and know there is no point to the puzzle. The AI only has the millions of past examples to go on.”

I nodded, and was happy, I didn’t feel so alone anymore…

Script output from o1 (but all LLMs have similar output)