Two headlines from the same week tell seemingly opposite stories about artificial intelligence in finance.
At one end, Ken Griffin declared that generative AI “just falls short” when it comes to uncovering alpha at Citadel. At the other, Viking Global reported that its in-house VikingGPT 2.0 system now fields three queries per minute from traders, with usage up 400% year over year.
So which is it? Is AI transforming hedge funds or falling flat? Both stories are true, and together they capture the crossroads where financial AI now stands.
The answer reveals a crucial divide that every institutional investor must grasp. Griffin and Viking aren't contradicting each other; they're solving different problems. Griffin is right to be skeptical of AI systems that claim to autonomously generate alpha. Viking is right to invest heavily in AI that helps their best analysts generate alpha faster.
There is a fundamental difference between AI that creates alpha and AI that clarifies the intelligence behind it. The best firms are using AI to amplify the judgment of their people, not to automate it. Others are still waiting for autonomous "oracle" systems that may never materialize or lack the internal architecture to build for their own domain-specific workflows.
The Griffin Critique: Why AI-as-Oracle Fails
Griffin's skepticism is more insightful than cynical. Speaking at the JPMorgan Robin Hood Investors Conference, he noted that generative AI “clearly enhances productivity,” but hasn't replaced the “meaningful research” that drives his $69 billion firm. His CTO, Umesh Subramanian, was even blunter: “In a world where the AI was making the trading decisions, everyone would know what it would do.”
That line exposes the fallacy of AI-as-oracle in competitive markets. Alpha exists at the edge of asymmetry, in knowing something others don’t or interpreting the same facts differently. If every fund queries the same foundation model for investment ideas, the collective advantage collapses. You’re not discovering alpha; you’re commoditizing analysis.
The mistake is treating AI as a replacement for human discretion in a domain where discretion itself is the moat. Griffin’s caution is less about technological limits and more about competitive structure. Alpha is a relative game.
The Viking Model: AI as an Intelligence Multiplier
Viking Global took the opposite approach. They didn't ask AI to make investment calls. They built VikingGPT to help their best people make better ones, faster, more rigorously, and with less friction.
Deputy CIO Previn Mankodi described using the system as a “skeptical avatar” to challenge investment theses, not a machine oracle but a disciplined second voice. The tool accelerates research, pressures assumptions, and reduces the time between hypothesis and validation.
This is intelligence illumination, not intelligence substitution. The model doesn't decide what to buy or sell. It enables analysts to reason at scale by synthesizing, testing, and refining their insights in real time.
The results are tangible. Viking's main fund underperformed in 2025, partly due to their caution on AI-linked equities, but their internal productivity and research throughput surged. Constant system engagement, three queries per minute, signals authentic utility, not novelty.
The Design Principle: Augmentation Beats Automation
The contrast between Citadel's caution and Viking's experimentation reveals a broader truth:
The most valuable AI doesn't replace expertise. It amplifies it.
In high-stakes, information-dense environments like global macro investing, the winning AI architectures share common traits:
As Viking's co-founder Andreas Halvorsen put it, “human decision-making remains central.” The technology is valuable precisely because it strengthens, not substitutes, the human edge.
Applying the Principle: Menos AI and the Amplification Model
This is the same philosophy guiding our work at Menos AI, an agentic AI firm based in Silicon Valley. We partner with global macro and multi-strategy investors to build intelligence amplification systems that strengthen analysts' reasoning and accelerate idea generation without eroding human discretion.
This approach is particularly well suited to macro and multi-strategy investing, where success depends on synthesizing complex, multi-variable dynamics across rates, FX, commodities, and credit. These strategies demand the ability to track dozens of interrelated narratives simultaneously, assess how macro regime shifts ripple through asset classes, and quickly evaluate whether a thesis is gaining or losing evidential support.
That’s exactly the kind of cognitive load AI can relieve while preserving the judgment calls that separate skilled investors from the rest.
Our core platform, Sonαr, operationalizes this approach. It converts vast volumes of unstructured financial commentary, research, and market narrative into an interactive reasoning environment. Instead of asking, “What should I trade?”, users interrogate the market's collective thinking: Why do participants believe what they believe, and how is that reasoning evolving?
Voice Scoring™ and the Narrative Frontier: Measuring Reasoning, Not Just Results
Two proprietary technologies underpin this framework.
Voice Scoring™ evaluates analytical skill not just by P&L outcomes but by causal accuracy: whether the analyst's reasoning matched how the world unfolded. This distinction matters. In macro investing, being right for the wrong reasons is luck, not edge. Conversely, being wrong for the right reasons often signals genuine insight masked by noise.
For instance, an analyst shorting JGBs in early 2025 might have anticipated insurer selling and rising issuance, both fundamentally sound calls, yet been blindsided by a one-off geopolitical shock that temporarily moved prices the other way. Traditional attribution would mark that as failure; Voice Scoring™ recognizes the underlying analytical validity.
Building on that, Narrative Frontier maps the entire ecosystem of market theses, continuously tracking which narratives are gaining evidential strength, which are weakening, and how price action responds to shifts in reasoning. The result is a living knowledge graph of market cognition.
Portfolio managers can use this to:
In short, it allows firms to see how markets are thinking and which ideas rest on firm logical ground versus shaky inference.
The Broader Lesson: Build With, Not Against, Human Judgment
The next wave of AI adoption in investing will not be led by firms chasing autonomous trade generators. It will be led by those designing systems that help their experts think more clearly, move more quickly, and see more deeply.
Ken Griffin is right: AI will not replace the deep research that drives alpha. But Viking Global's experience, and the work we're doing with leading macro and multi-strategy funds, shows that it can redefine what deep research feels like.
The future isn't man versus machine. It’s man with machine, building judgmental advantage through synthetic intelligence.
That’s where the real alpha lives: in human insight illuminated by AI.
Menos AI partners with global macro and multi-strategy investors to build intelligence amplification systems for research and portfolio construction.
Learn more about Sonαr here.