The People Machine
Enable social system simulations powered by LLMs
The Problem with Premonitions
You have heard it said before: no one knows what tomorrow is going to bring. For a long time, this statement was all but universal. No one did, by and large, know what was ever going to happen next. Anyone who claimed to tell the future was almost certainly lying. Weather reports were notoriously inaccurate, and for systems involving large numbers of people, predictions fared even worse. Human behavior is inherently unpredictable, which has made forecasting and planning something of a fool’s errand, to the ire of businesses across every sector.
According to Clayton Christensen, a professor at Harvard Business School, of 30,000 products launched annually, approximately 95% are doomed to fail. Of course, any entrepreneur will tell you that entrepreneurship – launching new products, trying new things – is inherently risky. There are never any guarantees that a product will find its customers, that it will achieve the all-important product-market fit, before investment capital dries up. Even more striking about Prof. Christensen’s findings, however, is that the rule applies even to large companies.1
For Coca-Cola, it was New Coke. For Google, it was Google Glass. Large companies, despite their wealth of talent, make frequent missteps. There is no obvious way to avoid this issue, either. Without new product launches, without finding new sources of revenue, growth will begin to dissipate, leaving investors frustrated, pushing them to take their money elsewhere.
In retrospect, it may seem obvious that businesses can’t tell how their product launches are going to land. After all, basic economics will tell us that is the whole reason there is profit to be had: you risk your capital by essentially making a bet that you have found something people want that other businesses haven’t noticed yet.
What is less easy to explain away, however, is why even our fundamental polling systems fail with some regularity. Infamously, almost none of the professional poll-takers and researchers correctly predicted the outcome of the 2016 US presidential election. Upshot, at the time a widely touted and highly respected system for modeling the election, reported an 85% probability that Hillary Clinton would best Donald Trump. 2
Even more recently, the Covid-19 pandemic illustrated the ways that predictive models may appear to follow a certain trajectory for a time, only to fall apart without any warning. While people behaved mostly the same as one another, quarantining in place, in the first weeks of the pandemic, greater regional variations and individual variations in behaviors became apparent as time went on.3
Here, we get to the crux of the problem: people are complicated, and the more people you are trying to make predictions about, the more complications you have to consider.
To the human mind, the math involved in all this was out of reach throughout history. We simply did not have the equipment to record or make use of the huge amounts of data needed to simulate people’s behavior accurately.
Recent innovations, though, have pointed to a clearer picture of the future for us all. It’s no longer a given that we can’t know what tomorrow is going to bring.
Simulatrex – Limitless Understanding
As of 2024, according to one estimate, there are 3.5 quintillion bytes of data generated each day: that is 3,500,000,000,000,000,000 bytes, every single day, of new data, with that number also increasing each day, as more services and users come online.4 Just like the systems that all this data represents, this number is impossibly large for the human mind to fathom. As data scientists have shown – through breakthroughs in business analytics, recommendation algorithms, and even medical discoveries based on advanced artificial intelligence like AlphaFold – there is untold wisdom contained inside all this information.
The issue that we are now coming up against, though, is that no one is sure how to make sense of all that information.
Traditional statistical analyses fall short. It is not enough to come up with a “mean” result and then calculate back toward that. Remember that when we are dealing with people, unforeseen complexities will arise. No matter how clever we may think our approaches are, the fact is that when people come onto the scene, all of our assumptions begin to collapse. Predictions that seem straightforward – the 2016 US presidential election, the rollout of an exciting product by a beloved brand – slip through our fingers.
Simulatrex is the engine for social simulations, to break through the barrier that has prevented us from making accurate, decisive predictions about the future.
Our vision is this: a future in which social dynamics, even the most perplexing and intricate social dynamics, become easier for us to understand, model, and predict. In turn, we will empower decision-makers to come up with more complete and more intelligent plans for the future. We will demystify the seemingly endless array of details and data points that have prevented businesses from rolling out their most ambitious product ideas and that have separated investors from their most accurate roadmaps.
In short, we believe that by simulating more, we open the door to doing more.
Leveraging large language models (LLMs), combined with advanced data science, we are equipping more people than ever before with the historic capability to see further ahead. Because we believe so wholeheartedly that this technology should belong to all people, we are open-sourcing the Simulatrex engine.
Then what?
Whatever your use case may be, we are making it simple for you to plug the Simulatrex engine into your data, to create a fully functional, spectacularly informative simulation. You can use this simulation to draw a straight line between the present moment and any moment in the future.
A new product? Get a clear idea of what the market is going to say about it, before you spend any of your valuable capital on development and distribution.
An upcoming election? Understand the electorate before you make any big statements about what’s to come.
New features, new market entries, any event, any change, any plan: run them all through, built on the Simulatrex engine, and navigate the future with more confidence than has ever been possible before.
What Does Simulatrex Unlock?
In order to understand why Simulatrex use cases will be so invaluable, it is important to understand the underpinnings of the engine and the systems that will emerge from it.
This is not just a wrapper around the current state-of-the-art LLM. As researchers have found, using the outputs of an LLM as synthetic data tends to heavily favor a regression toward the mean – which is, in a sense, exactly what an LLM is supposed to do. For simulations, an LLM alone is therefore insufficient.5 Simulatrex, by finetuning subpopulation representative models, can surpass foundation LLMs, generating more aligned outputs for a certain target audience.
Consider each of the following use cases for the Simulatrex engine.
Simulation of a Product Launch in a Niche Market:
Look at niche market data from sources such as Reddit and Twitter to understand customer sentiments, identify emerging market trends, and tailor marketing strategies to meet the unique needs of the target audience. Simulatrex can track the evolution of consumer opinions over time so that when you are ready to move into the rollout phase, you have a precise idea of the responses and reactions you are likely to garner. You can even game out different pricing strategies, promotional offers, and distribution channels, for a robust risk analysis before any big spending.Simulation of Google Ads:
Running a Google Ads campaign through Simulatrex, you can experiment with different ad formats, keywords, and targeting options without risking any misspent marketing capital. Make predictions about the ways changes in ad design, wording, and placement will influence click-through rates and conversions, while you also test different bidding strategies to find the most cost-effective approach. With this type of mini app built on Simulatrex, you can even go so far as to simulate customer journeys from ad clicks to website navigation, to identify potential drop-off points and opportunities for improving the user experience.Simulation of Traffic via PostHog Analytics:
Utilize your own PostHog Analytics to simulate your traffic and audience, seeing how changes and new features will land – without risking any negative pushback. You can use the Simulatrex engine in this type of mini-app to model different user segments, getting a more nuanced understanding of how different groups engage with your site or app.
Until now, you would have had to spend significant capital researching real-world feedback for any of these use cases. You also would have had to wait, until the feedback became available, to draw any insights, based on events that had already taken place.
Not anymore.
The Simulatrex engine is a social crystal ball. With it, you can look at formerly inscrutable collections of data – tremendous amounts of web traffic, a whole internet’s worth of marketing possibilities, your next market, your new audience – and feel certainty about the choices you’re making.
Powering Tomorrow’s Predictions
The goal of Simulatrex is not just to help people make better predictions, but to help people get better at making better predictions. That may seem like a subtle distinction, but it’s an important distinction too.
What is your use case? What will you predict?
Simulatrex is the engine that, at long last, puts you in control of knowing. While you may not be able to change the future, you can shape it more in line with what you would like to see. You can make shifts and pivots in the plans and strategies that you have set – based not on guesses, based not on hope, but based on data and facts, comprehended through the most advanced technology available today.
Whether you're in finance, predicting market trends and stock movements, or healthcare, anticipating disease outbreaks and patient outcomes, or retail, forecasting consumer behavior and product demand, Simulatrex makes every challenge more manageable. Ours is a data-driven approach that taps into the wonders of LLMs.
Understand more. See through more. Get a glimpse that is further and further ahead into tomorrow and every tomorrow that comes after it. Simulatrex isn't just a tool for making predictions; it's a way to hone your foresight, to learn how to ask the right questions, to interpret data more effectively, and to understand the nuances of your next big decision point.
There is no longer any social system too complex for you to handle. There is no longer any obstacle too great for you to overcome.
You have the power of knowing at your disposal, with Simulatrex.
Find more here: https://simulatrex.com.