Artificial Intelligence: Superhuman Breakthrough or Smarter Tool? | Don't Short Yourself

Key Concepts

• Enterprise AI: AI systems specifically designed for business reliability, security, and efficiency rather than general-purpose consumer interaction.
• "Boring" AI: The concept that the most valuable AI is predictable, reliable, and integrated into the background of infrastructure, similar to electricity.
• Scaling Laws: The industry hypothesis that increasing compute power and data volume leads to better model performance.
• Open Weights/Open Source: Models (like Llama or those from IBM/Mistral) that allow community contribution and prevent vendor lock-in.
• Hallucinations/BS: The tendency of LLMs to generate plausible-sounding but factually incorrect or fabricated information.
• Agentic AI: AI systems capable of autonomous action, which introduce new cybersecurity risks like "insider threats."

1. The Disconnect in AI Adoption

David Cox (IBM) notes that while AI technology is advancing rapidly, companies are struggling to move from "cocky adolescence" to reliable, secure enterprise deployment.

• The 95% Failure Rate: Citing an MIT report, Cox explains that many AI pilots fail because companies prioritize "low-hanging fruit" (easy MVPs) that lack genuine business value.
• Cost Overruns: Deploying AI is significantly more expensive than building a demo. Many projects fail because the operational costs exceed the value generated.
• Non-Functional Requirements: A major barrier is the lack of security and reliability. Unlike consumer chatbots, enterprise AI must be legally and operationally sound to avoid lawsuits and reputational damage.

2. The "Boring" Future of AI

Cox argues that the industry’s obsession with Artificial General Intelligence (AGI)—systems that think and reason like humans—is often a marketing distraction.

• Utility vs. Flash: He compares AI to electricity. Electricity is "boring" because it is predictable and reliable. He advocates for "purpose-fit" models, such as IBM’s Granite models, which are scoped to specific tasks (e.g., HR policy queries) rather than wasting compute power on unnecessary general knowledge (e.g., advanced physics).
• Efficiency: He notes that every 6–9 months, the industry finds ways to perform the same tasks with models 10 times smaller and more energy-efficient.

3. Cybersecurity and Risk Management

The integration of AI into business workflows creates a new "attack surface."

• Insider Threats: AI agents with broad access rights can inadvertently erase data or execute harmful commands. Cox compares this to the early days of the internet, where poor coding practices (like SQL injection) created vulnerabilities that the industry eventually learned to patch.
• Legal Accountability: Courts are increasingly holding companies responsible for the statements made by their AI representatives. If an AI promises a refund that violates company policy, the company is legally bound to that promise.

4. The Future of Work and Skills

• Augmentation, Not Displacement: Cox believes AI will automate specific tasks within jobs rather than replacing entire roles.
• Human-in-the-Loop: He emphasizes that humans are essential for accountability and high-level architectural thinking. "Coding is a way of expressing thought," and the ability to solve problems and manage systems will remain a human-centric skill.
• Adaptability: He advises students to focus on "thinking problems" (architecture, strategy, and soft skills) rather than just syntax-based coding.

5. Market Dynamics and Open Source

• Circular Investment: Cox expresses concern over the "circular" flow of capital in the AI industry, where companies invest in each other to buy hardware, potentially inflating the perceived value of the sector.
• The Open Source Counterpoint: He argues that open-source models are essential to prevent a few proprietary players from controlling the foundational technology of the global economy. He draws a parallel to Linux, which became the backbone of the internet because it was open and collaborative.
• Diminishing Returns: As open-source models close the performance gap with proprietary models, the "moat" of the big tech companies may shrink, making it difficult for them to recoup the trillions of dollars currently being invested.

6. Synthesis and Conclusion

The main takeaway is that the current "hype cycle" of AI is unsustainable in its current form. The industry is moving toward a maturation phase where efficiency, reliability, and security will outweigh raw, flashy capability. For businesses, the path to ROI lies in boring, purpose-built applications that solve specific problems without the risks of hallucination or excessive cost. Ultimately, the long-term success of AI depends on whether it remains an open, accessible utility or becomes a proprietary tool controlled by a few dominant entities.