I think this can transform market research and sociology. We shouldn't view these simulations as a replacement for going out into the field and talking to real people. But they can enhance the design of that kind of field work.
I think the idea that we may live in a simulation — that reality itself is a simulation — taps into some deep change in the way we're thinking about our ultimate place in the universe. The more compelling these LLM-fuelled simulations become, the more this idea will take hold.
And yes, this has got to be the future of video games!
Cue the sceptics saying they're constrained by their training data. But no, this is about complex relationships and the layering of outcomes. There will be probabilities at play.
Market research, political science, VC & seed capital, wargaming....
Just wow.
David, any idea if this has been commercialised by Stanford yet?
Enjoyed your thought-provoking email about AI agents and the work of Joon Sung Park. I’d love to hear more about how you anticipate government, scientists and corporations will leverage this social simulations. Perhaps you could go into that in a deeper dive on a pod ep, please? I always enjoy those deep conversations you have with Raoul Pal… And I can’t help but feel this simulation is just a simulation within a simulation (and so on and so on). Great work as always, David - keep going!
You know I am a longtime skeptic of these methods. Individual modeling in reductionist isolation is one thing in a deterministic system. But that’s nothing like the world we live in now.
When interactions and complexity take over, the individual matters far less than their relationships to each other. That is where the weight of the models need to preside. It is why Romania almost elected a complete outsider this month.
Outliers are contorting and redefining the world, and yet these models are built upon linear predictive outcomes. Not the non-linear effects of interactive social complexity.
LLMs can only regress towards the mean with predictive algorithms. This is why AI was a complete failure during Covid: there was no prior data to build upon that would tell us WHO declaring a pandemic would cause a global consumer run on toilet roll. You cannot model where there isn’t data for it.
Good stuff
Harbinger of the death of Market Research as we know it
I think this can transform market research and sociology. We shouldn't view these simulations as a replacement for going out into the field and talking to real people. But they can enhance the design of that kind of field work.
Agreed. I fear however that the cost savings will be irresistible to many clients
Once again, science fiction coming to fact. A future of individual Matrixes could be coming.
Also, if I worked in video games, I’m pouring my R&D investments into this work!
I think the idea that we may live in a simulation — that reality itself is a simulation — taps into some deep change in the way we're thinking about our ultimate place in the universe. The more compelling these LLM-fuelled simulations become, the more this idea will take hold.
And yes, this has got to be the future of video games!
100% yes!!
This is fascinating? And I’m curious? How can this improve our lives?
Beyond exciting.
Cue the sceptics saying they're constrained by their training data. But no, this is about complex relationships and the layering of outcomes. There will be probabilities at play.
Market research, political science, VC & seed capital, wargaming....
Just wow.
David, any idea if this has been commercialised by Stanford yet?
Enjoyed your thought-provoking email about AI agents and the work of Joon Sung Park. I’d love to hear more about how you anticipate government, scientists and corporations will leverage this social simulations. Perhaps you could go into that in a deeper dive on a pod ep, please? I always enjoy those deep conversations you have with Raoul Pal… And I can’t help but feel this simulation is just a simulation within a simulation (and so on and so on). Great work as always, David - keep going!
You know I am a longtime skeptic of these methods. Individual modeling in reductionist isolation is one thing in a deterministic system. But that’s nothing like the world we live in now.
When interactions and complexity take over, the individual matters far less than their relationships to each other. That is where the weight of the models need to preside. It is why Romania almost elected a complete outsider this month.
Outliers are contorting and redefining the world, and yet these models are built upon linear predictive outcomes. Not the non-linear effects of interactive social complexity.
LLMs can only regress towards the mean with predictive algorithms. This is why AI was a complete failure during Covid: there was no prior data to build upon that would tell us WHO declaring a pandemic would cause a global consumer run on toilet roll. You cannot model where there isn’t data for it.