Will AI Suddenly 'Evolve'? Paradigm Uses a Game to Tell You the Answer

Author: Xiaobing

The well-known crypto VC Paradigm launched a web game yesterday called RSI Simulator.

Players take on the role of the head of an AI laboratory, with the goal of creating superintelligence from scratch.

You have three types of resources: manpower, computing power, and data, and you need to decide how much to invest in research and development, how much in training, and how much in expansion. After each round of decisions, the model's capabilities change, which in turn affects the efficiency of research and development, until a certain critical point is reached where the AI begins to accelerate its own improvement, and the curve rises sharply.

It sounds like an idle game. In reality, it is an educational tool that transforms economic research papers into an interactive experience.

Paradigm uses it to answer a question that everyone is guessing but few have quantified: How fast will AI's self-improvement occur? Will it suddenly explode? What factors determine its ceiling?

What is Recursive Self-Improvement?

Recursive self-improvement, abbreviated as RSI, is one of the core topics in AI safety research.

In simple terms: when an AI model becomes powerful enough to help humans improve the next generation of AI models, a cycle is formed where AI participates in creating stronger AI, and stronger AI then participates in creating even stronger AI.

Once this cycle starts, it can theoretically produce exponential acceleration, known as an intelligence explosion.

This is no longer science fiction. In February 2026, OpenAI admitted that early versions of the GPT-5.3-Codex model played a "significant role" in creating itself, helping to debug the training process, manage deployment, and diagnose evaluation failures. This was the first time a leading lab officially acknowledged that a model substantially participated in the engineering process of creating its successor.

How to Play the Game?

The game part of RSI Simulator simulates resource allocation decisions in an AI laboratory.

In each round, you need to allocate resources among three directions. Investing more computing power allows you to train larger models; investing more manpower in research can improve algorithm efficiency (achieving the same effect with less computing power); investing in data collection can broaden the boundaries of training materials.

The key mechanism is: when the model's capabilities reach a certain level, it begins to "help with research," and your manpower research efficiency improves due to AI assistance. This is the starting point of the RSI cycle. You need to judge when to switch resources from expansion mode to "let AI research itself" mode.

If you switch too early, the model's capabilities may not be sufficient, and the AI's research efficiency may be lower than that of humans, wasting computing power. If you switch too late, you may be surpassed by other (hypothetical) laboratories.

Paradigm also includes an "Explorer" that allows users to directly adjust parameters of the underlying economic model, such as the "elasticity coefficient of model capability to research speed," and observe how the AI progress curve changes under different parameter settings.

This part is aimed at a more hardcore audience, essentially transforming the mathematical models from academic papers into a draggable dashboard.

Four Insights the Game Aims to Convey

Paradigm summarized four insights gained from the model and the game in a supporting research article.

The bottleneck is always in the weakest link. AI research and development requires three complementary elements: manpower, computing power, and data. Even if AI intelligence increases infinitely, the physical limitations of computing power and data still exist. Recursive self-improvement may be limited by the supply of computing power or data quality, rather than intelligence itself.

Self-improvement may be pulsed, not a continuous exponential growth. AI may experience a period of self-acceleration and then stop at a point far from physical limits if computing power becomes a bottleneck. An intelligence explosion may not be a straight line soaring into the sky, but rather several rapid steps interspersed with plateaus.

Narrow domain intelligence explosions may occur before general intelligence explosions. AI may first achieve accelerated self-improvement in the narrow field of "AI research," but this capability may not generalize to all cognitive tasks. You can imagine an AI that is very good at improving AI training code but has no equivalent breakthroughs in managing supply chains or writing novels.

Everything depends on parameter settings. Model outputs are highly dependent on a set of parameters called "elasticity coefficients." The elasticity coefficient measures "how much percentage output increases when input increases by 1%." The most critical elasticity coefficient in recursive self-improvement is: how much percentage increase in discovery speed is brought about by a 1% increase in current model capability. This number may change over time, and tracking its real-time fluctuations is crucial.

Why Paradigm?

Paradigm is no longer just a pure crypto VC.

A month ago, Paradigm completed a new fundraising round of $1.2 billion. The allocation of this money is markedly different from the past three rounds, as co-founder Matt Huang (former partner at Sequoia Capital) and managing partner Alana Palmedo wrote in the announcement that the fund will "start from crypto and expand into AI, robotics, and other frontier fields."

Since its establishment in 2018, Paradigm has raised over $4 billion, investing in iconic projects in the crypto industry such as Uniswap, Optimism, Hyperliquid, and Kalshi.

Now, Paradigm's investment landscape has clearly crossed boundaries. In March, it led a Series H investment in drone delivery company Zipline, in April it participated in a Series D investment in space defense company True Anomaly, and simultaneously invested in the AI research lab Nous Research.

The internal tools are also extending into AI. Paradigm has developed Centaur (an AI agent toolkit) and EVMbench (a blockchain security assessment project in collaboration with OpenAI), with one leg standing on the chain and the other already stepping into AI infrastructure.

In this context, the RSI Simulator is no longer just a PR project; Paradigm is betting $1.2 billion on the "intersection of frontier technologies," namely AI acceleration, decentralized computing, and automated infrastructure.

Understanding the economics of RSI is to understand the underlying logic of its biggest bet. If the speed of AI's self-improvement aligns with certain parameter settings in the game, the update cycle of the entire tech stack will be compressed, and projects that seem to be three years ahead today may be eliminated in six months.

At the end of the article, Paradigm stated that they are interested in any work that pushes the frontier of RSI modeling and encourages researchers in related fields to reach out. This game serves as a business card, behind which lies $1.2 billion and a transforming investment institution.

Game address: https://www.paradigm.xyz/research/rsi/game

Model Explorer: https://www.paradigm.xyz/research/rsi/explorer

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.