This concept is central to how AI performs pattern recognition. Large models, trained on billions of parameters, learn to generate vector embeddings—numerical representations of data. These embeddings are stored and indexed in a vector database. In this high-dimensional space, the proximity between vectors corresponds to semantic similarity, allowing the AI to 'see' relationships as clusters of related data points.
I wrote this in the Kurgan´s blog. But it suits here as well.
I used chatGPT for the first time about 2 weeks ago. I am in the chemistry field and asked to formulate a product that is as good or better than the market leader. I am a one man show, vs companies with 100s of employees and let's say 10s of lab coats. This particular chemistry I was not too familiar with so I thought I would finally see what this AI thing is about.
It is incredible. In a day I had literally all the info I needed to start formulation, even very good guesses of what the competition product is made from and saved myself weeks of fluffing around... But yes it makes A LOT of mistakes. This is the crux. Since I am an "expert" in formulating I can sniff the BS. The beauty was once you tell it it is BSing it reconfigures and boom..
People like me (hopefully of IQ) are going to be able to do the work a whole room full of 110 IQs only faster. If you do not understand the topic well you can definitely get arse raped into going down the wrong rabbit hole.
This will be the same for the idiots who work for Mckinsey etc. Why the hell would you hire 5 x 120IQ consultants on 2000 USD a day plus expenses to offer shitty advice? hire a 140-150IQ guy who's got a keen nose for sniffing out BS and he will do a better job far faster at much lower cost with literally a 20 dollar a month subscription.
The old adage. Tech is only as good as the person using it.
Agreed. Using minor scripting/coding knowledge and platform experience I was able to develop a script to do a task. It took 3 days. 1 to get the search and the default stuff down. The next 2 were finishing the task from the search results.
At one point, the script kept failing due to data errors even though it's supposed to transfer like to like. It persisted until I said, "Why don't you try putting the data into a list?" It recalibrated and the script worked.
You're absolutely correct. Higher IQ folks with AI are absolutely going to outperform others. Annoying that a bunch of gammas will be included.
I think simulator games tend to be examples of designed for simulation.
So something like Flight Simulator has highly detailed controls and simulated flight physics, where a more arcade-style flight game would give you simplified controls so you can focus on feeling like a pilot.
There's also this hilarious walking simulator QWOP.
Normal games let you get the effect of walking by pushing a single arrow key.
QWOP gives you control over individual leg muscles (4 keys to control L/R thighs and calves), and you have to control the sequencing/timing in order to walk. This is hard. It's a free browser game that you should try out and see.
The utter frustration of getting basic things to work is why fun games use a lot more effect than simulation.
Don't recall the exact name of the game, but it was a Civil War game covering the 7 days battles. In that game you had to write orders out and your messengers had to deliver them and then you had to roll for them to be accepted. This created a somewhat random but bounded lag between when you wanted to start something and when it would actually happen and made coordination of movement pretty challenging. I think that fits in the design for simulation space.
"AI as DM/GM" is an excellent rhetorical model. I think we'll make real headway with this one.
This concept is central to how AI performs pattern recognition. Large models, trained on billions of parameters, learn to generate vector embeddings—numerical representations of data. These embeddings are stored and indexed in a vector database. In this high-dimensional space, the proximity between vectors corresponds to semantic similarity, allowing the AI to 'see' relationships as clusters of related data points.
The formulas can be read in books from 25 years ago, people have no idea about it.
The math and concept is so far above them.
Being a gamer helps understand it, another good reflection.
People want a robot friend. Having several good tools cant occur to MPAI.
I wrote this in the Kurgan´s blog. But it suits here as well.
I used chatGPT for the first time about 2 weeks ago. I am in the chemistry field and asked to formulate a product that is as good or better than the market leader. I am a one man show, vs companies with 100s of employees and let's say 10s of lab coats. This particular chemistry I was not too familiar with so I thought I would finally see what this AI thing is about.
It is incredible. In a day I had literally all the info I needed to start formulation, even very good guesses of what the competition product is made from and saved myself weeks of fluffing around... But yes it makes A LOT of mistakes. This is the crux. Since I am an "expert" in formulating I can sniff the BS. The beauty was once you tell it it is BSing it reconfigures and boom..
People like me (hopefully of IQ) are going to be able to do the work a whole room full of 110 IQs only faster. If you do not understand the topic well you can definitely get arse raped into going down the wrong rabbit hole.
This will be the same for the idiots who work for Mckinsey etc. Why the hell would you hire 5 x 120IQ consultants on 2000 USD a day plus expenses to offer shitty advice? hire a 140-150IQ guy who's got a keen nose for sniffing out BS and he will do a better job far faster at much lower cost with literally a 20 dollar a month subscription.
The old adage. Tech is only as good as the person using it.
Agreed. Using minor scripting/coding knowledge and platform experience I was able to develop a script to do a task. It took 3 days. 1 to get the search and the default stuff down. The next 2 were finishing the task from the search results.
At one point, the script kept failing due to data errors even though it's supposed to transfer like to like. It persisted until I said, "Why don't you try putting the data into a list?" It recalibrated and the script worked.
You're absolutely correct. Higher IQ folks with AI are absolutely going to outperform others. Annoying that a bunch of gammas will be included.
Chat 4o opined:
Personalization as Persistent Influence
Instead of treating every prompt as a blank slate, personalization means the model:
• Remembers how you think
• Adjusts tone and complexity
• Prioritizes familiar topics or phrasing
• Uses your past interactions to weigh probable next tokens differently
So “most probable token” now becomes:
“Most probable next token, given the context, style, tone, topic, and prior choices of this user.”
Sorry for the side track, but Squad Leader is indeed one the the greatest games ever.
Very noticeable when homonyms throw off Suno. Wind != Wind, Wound != Wound
Context doesn't particularly help either.
Wuhowend
Sometimes the phonetics need to be manually injected for Suno, especially for unusual nouns.
Thank you. I'll try that.
Great explanation! For the sake of my full understanding, could someone give an example of a game or system that is Design for Simulation?
I think simulator games tend to be examples of designed for simulation.
So something like Flight Simulator has highly detailed controls and simulated flight physics, where a more arcade-style flight game would give you simplified controls so you can focus on feeling like a pilot.
There's also this hilarious walking simulator QWOP.
Normal games let you get the effect of walking by pushing a single arrow key.
QWOP gives you control over individual leg muscles (4 keys to control L/R thighs and calves), and you have to control the sequencing/timing in order to walk. This is hard. It's a free browser game that you should try out and see.
The utter frustration of getting basic things to work is why fun games use a lot more effect than simulation.
Don't recall the exact name of the game, but it was a Civil War game covering the 7 days battles. In that game you had to write orders out and your messengers had to deliver them and then you had to roll for them to be accepted. This created a somewhat random but bounded lag between when you wanted to start something and when it would actually happen and made coordination of movement pretty challenging. I think that fits in the design for simulation space.