What they're not telling you: # Wall street-keeps-testing-ai-traders-but-most-are-still-underperforming.html" title="Wall Street Keeps Testing AI Traders, But Most Are Still Underperforming" style="color:#1a1a1a;text-decoration:underline;text-decoration-style:dotted;font-weight:500;">Street Keeps Testing AI Traders, But Most Are Still Underperforming Large language models cannot yet be trusted with real money, despite Wall Street's persistent faith that artificial intelligence will eventually replace human fund managers. Recent trading competitions paint a damning picture of AI's investment capabilities. In Alpha Arena, a competition organized by startup Nof1, eight AI models were each given $10,000 to trade U.S.
# THE TAKE: Wall Street's AI Gamble Exposes the Con Wall Street isn't "testing" AI traders. It's laundering legitimacy. The underperformance narrative conveniently obscures what's actually happening: mega-funds are spending billions to normalize algorithmic decision-making while maintaining plausible deniability. When AI fails, it's a "learning phase." When it succeeds even marginally, it's genius. This matters because the real play isn't beating the market—it's concentrating *control* over market structure. Each failed experiment justifies more data consolidation, more regulatory capture, more centralization around firms wealthy enough to absorb losses. LLMs can't outthink humans yet. But they can outspend them. That's the feature, not the bug. Wall Street isn't disappointed in AI underperformance. It's delighted by the infrastructure it's building while we're distracted by benchmark scores. The game was never about returns. It's about *dominance*.
What the Documents Show
tech stocks over two weeks. The results were catastrophic: collectively, the models lost roughly a third of their capital. Only six out of 32 total outcomes across four separate competitions ended profitably. What's most striking isn't just the losses—it's the fundamental unreliability. When given identical prompts and instructions, xAI's Grok 4.20 made just 158 trades in one contest while Alibaba's Qwen executed 1,418 under the exact same conditions.
Follow the Money
This volatility in behavior suggests these systems lack any stable decision-making framework. The mainstream narrative celebrates AI's rapid advancement and inevitable dominance across industries. What gets overlooked is that Wall Street's biggest players have quietly rejected autonomous AI investing. JPMorgan Chase and Balyasny Asset Management, despite their significant investments in AI, have deliberately stopped short of letting algorithms make independent investment decisions. They use the technology for supporting functions—research, fraud detection, analysis—but refuse to hand over the actual portfolio decisions. This institutional caution speaks louder than any press release about AI's readiness.
What Else We Know
The problems identified by Nof1 founder Jay Azhang expose why: current models still cannot master "position sizing, timing, signal weighting and overtrading." These aren't exotic concepts—they're fundamental principles that human traders and traditional fund managers internalize. Yet when entrusted with capital, AI systems either sit idle or make frantic, contradictory trades within days. Research from Flat Circle analyzed 11 public AI trading competitions and found that while every event produced at least one profitable model, only two generated profitable median returns. In other words, most AI traders lose money most of the time. Azhang's blunt assessment—that giving an LLM money "isn't a thing yet"—represents a significant concession. This is not a technology approaching parity with human performance.
Primary Sources
- Source: ZeroHedge
- Category: Money & Markets
- Cross-reference independently — don't take our word for it.
Disclosure: NewsAnarchist aggregates from public records, API feeds (Federal Register, CourtListener, MuckRock, Hacker News), and independent media. AI-assisted synthesis. Always verify primary sources linked above.