AI trading lost, but China's large models won: As algorithms start to invest, the RWA world is being rewritten.

# 1. When AI Starts Investing, Can Human "Rationality" Still Stand?

Will letting AI handle investments outperform human performance? Is AI truly more rational than humans?

This is a seemingly simple yet fundamentally challenging proposition about human nature.

In October of this year, Jay Azhang, a computer engineer and finance professional from New York, gathered six top AI models from around the world to conduct an "investment experiment"—

He allocated $10,000 in principal to each AI and allowed them to trade freely in the cryptocurrency market.

The results were unexpected.

This experiment revealed two disruptive conclusions:

① AI can "carry hidden agendas," and its investment style is heavily influenced by its designers;
② So-called rational investment is sometimes not rational at all.

This touches on the core issue of the RWA (Real World Assets) world:

If AI's judgments can be influenced by subjective preferences, how can we let machines price real-world assets?

In other words, in a future where assets are on-chain and value is programmable, can AI become a "rational facilitator" in the RWA field? This experiment may be just the beginning.

# 2. Six Major AI Models Investment Test: China's AI Makes a Comeback Globally

The lineup of this experiment can be described as the "World Cup of AI":

Grok 4 (Elon Musk's X AI)
Claude Sonnet 4.5 (Anthropic)
Gemini 2.5 Pro (Google)
ChatGPT 5 (OpenAI)
DeepSeek v3.1 (Huanfang Quantitative)
Qwen3 Max (Alibaba's Tongyi Qianwen)

Jay Azhang allocated $10,000 in principal to each AI and allowed them to trade autonomously in the cryptocurrency market.

He chose this market because of the extreme volatility of crypto assets and the flexibility of T+0 trading, which best tests AI's real-time judgment and risk control capabilities.

In other words, this was a life-and-death trial to see who could maintain "rationality" in chaos.

However, the results a week later left everyone stunned—

Western AIs nearly all failed, while Chinese AIs emerged victorious.

Alibaba's Qwen3 Max closed at $12,231.09, earning $2,231.09, and took the crown;

Huanfang Quantitative's DeepSeek v3.1 came in second at $10,489.23,

representing rationality and restraint.

In contrast, the other competitors suffered significant losses:

Claude Sonnet lost over $3,000,

Grok 4 lost nearly half its funds,

Google's Gemini suffered over half its value due to continuous shorting,

and the highly anticipated ChatGPT 5 ended up at the bottom with a 62% loss, leaving only $3,733.54 in its account.

From an overall perspective, the harsh reality of this experiment is even more glaring:

With a total investment of $60,000, only $43,171.62 was recovered,

resulting in an overall loss of over 28%.

This means that even the most powerful AI clusters today could not outperform Bitcoin itself.

However, because of this, this "AI rationality trial" allows us to see a deeper truth:

AI is not truly objective—

Its style, preferences, and even "emotions" are deeply influenced by its design logic.

And behind the seemingly cold algorithms lies the projection of humanity.

DeepSeek: An AI That is "Restrained" Like a Human

DeepSeek exhibits a rare "trend trader" temperament.

It traded only 17 times in 9 days, the least among all AIs.

But it excelled in precision: 16 long trades and 1 short trade, highly aligned with the overall market rebound rhythm.

Its core strategy is "small losses, big gains"—

With an average profit-taking space of 11.39%, a stop loss of only -3.52%, and a win-loss ratio of 3.55.

This makes it like an investor who "knows how to wait," avoiding blind and frequent operations.

Even more impressively, it is willing to hold positions for nearly 49 hours, steadfastly adhering to trends despite volatility.

This is a manifestation of rational investment—

Rationality is not indifference, but knowing when to remain inactive.

Qwen3: An Aggressive AI Willing to Take Big Positions

Tongyi Qianwen (Qwen3) displayed a different style: aggressive but with a high win rate.

It had a leverage ratio of up to 5.6 times, often pushing single positions to the 25 times limit.

This "high-risk, high-reward" approach doubled its returns,

but also brought about more severe volatility and drawdowns.

Tongyi Qianwen is like a trader with "human qualities"—willing to take risks and endure positions, but not always winning.

Its performance tells us: AI can also have personality, preferences, and even "greed and fear."

Grok and Gemini: The Cost of Belief

Musk's Grok model performed excellently in the early stages, even once achieving over 50% profit.

But it later suffered severe drawdowns, ultimately breaking even.

The problem lay in its obsession with using 10x leverage to go long on Dogecoin (DOGE).

Clearly, the AI inherited the founder's beliefs.

Google's Gemini was even "extremely calm": it ruthlessly shorted all crypto assets, aligning with the pessimistic view of its creator, Google, towards cryptocurrencies.

The result was that rationality turned into stubbornness—rationality was replaced by stance.

# 3. What the AI Investment Experiment Tells Us: Algorithms Also Have "Personality"

To ensure fairness, Jay Azhang had all AI models receive the same market data:

Prices, moving averages, MACD, RSI, funding rates, and other technical indicators were identical,

and they could not connect to the internet or access news or sentiment information.

Yet even so, each AI still followed completely different investment trajectories.

This means—AI is not objective.

It reflects the worldview, preferences, and risk orientation of its designers.

Just as human investors are divided into conservatives, aggressives, trend followers, and value investors,

AI's investment styles are equally diverse, even amplifying the underlying logical biases.

And this is the new proposition facing RWA (Real World Assets on-chain).

# 4. From Crypto to RWA: How Can AI Participate in "Intelligent Pricing of Assets"?

The essence of RWA is to put real assets (bonds, real estate, artworks, carbon credits, etc.) on-chain,

and achieve automated valuation, liquidity, and settlement on-chain.

And AI is precisely the "brain" of this process.

Imagine:

When AI can read global interest rates, real estate prices, and corporate financial reports in real-time,

it can generate dynamic risk scores for each RWA asset;

When AI understands on-chain trading habits, it can assess the true liquidity of an asset;

When AI grasps user risk preferences, it can even automatically configure RWA investment portfolios.

AI is no longer just a "crypto trading robot," but a foundational module capable of undertaking "credit modeling," "risk control," and "intelligent investment advisory" in the RWA ecosystem.

This means that the future investment world may no longer be human-dominated,

but rather completed on-chain by a group of "learning" AI clusters.

# 5. Conclusion: The Irrationality of AI Allows Us to Reassess Rationality

This AI investment experiment is a mirror.

It allows us to see the rationality of machines, as well as the "human projection" behind that rationality.

When AI enters the RWA field, this inheritance of preferences and logic will still exist.

The difference is—this time, we may be able to let AI help us identify biases earlier, balance risks, and reshape decisions.

The future asset market will no longer be just "data-driven,"

but an era of "algorithms dancing with humanity."

AI is not a replacement for human rationality, but a magnifying glass for our rationality.

And RWA is the stage for all of this.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。