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

Key Concepts

• Enterprise AI: AI systems specifically designed for business productivity, reliability, and security, as opposed to general-purpose consumer chatbots.
• "Boring" AI: The philosophy that AI should function like a utility (e.g., electricity)—predictable, reliable, and integrated into the background of operations rather than flashy or human-like.
• AGI (Artificial General Intelligence): A theoretical form of AI that can reason and act like a human across various domains; often criticized by experts as a marketing term that introduces unnecessary risks and costs.
• Scaling Laws: The industry hypothesis that increasing compute power and data volume leads to better model performance, which currently drives massive capital expenditure.
• Open vs. Closed Models: The debate between proprietary, high-cost models (closed) and community-driven, transparent models (open weights) that allow for broader participation and lower costs.
• Hallucinations: The tendency of LLMs to generate plausible-sounding but factually incorrect information, which poses significant legal and reputational risks in enterprise settings.

---

1. The Disconnect in AI Adoption

David Cox (IBM) notes that while AI technology is advancing, companies are struggling to realize value. A key MIT report cited that 95% of generative AI pilots fail to add business value. The disconnect stems from:

• Focus on MVPs: Companies prioritize "low-hanging fruit" for demos rather than solving core business problems.
• Cost Overruns: Deploying AI at scale is significantly more expensive than running a demo, often negating potential ROI.
• Non-functional Requirements: Many pilots lack the necessary security, reliability, and governance to be safely integrated into enterprise workflows.

2. The "Boring" Future of AI

Cox argues that the industry’s obsession with AGI is misguided. He draws a historical parallel to the "War of Currents" between Edison (Direct Current/exciting but dangerous) and Tesla (Alternating Current/boring but reliable).

• Reliability over Flashiness: Businesses need tools that perform specific tasks consistently without "variance."
• Efficiency: IBM’s Granite models are designed to be purpose-fit, avoiding the wasted costs of training models on advanced knowledge (like physics) when the task is simple (like HR policy).

3. Risks and Cybersecurity

The integration of AI introduces new "insider threats" and attack surfaces:

• Data Leakage: Lax processes can lead to proprietary code or confidential data being exposed.
• Autonomous Agents: Tools like "Open-source agents" that have broad access rights can inadvertently erase data or perform unauthorized actions.
• Legal Accountability: Courts are increasingly ruling that companies are legally responsible for the statements made by their AI representatives, meaning "the AI hallucinated" is not a valid legal defense.

4. The Future of Work and Skills

Cox, a former computer science professor, suggests that AI will not replace entire jobs but will automate specific tasks.

• Human-in-the-loop: Humans are essential for accountability and judgment.
• Shift in Skills: The value of a programmer is shifting from "hands-on coding" to "system architecture" and "problem-solving." Coding is becoming a way to express thought rather than just a technical task.

5. Economic Outlook and Scaling

• Capital Expenditure: Hyperscalers (Microsoft, Meta, Google, etc.) are projected to spend ~$650 billion by 2026. Cox warns of "circular flows" of money where companies invest in each other to buy hardware, raising concerns about the sustainability of these returns.
• Energy Constraints: The physical limit of energy (megawatts) and the time required to build infrastructure (data centers/power plants) act as a hard cap on AI growth.
• Diminishing Returns: History shows that AI models tend to become 10x more efficient every 6–9 months. This "compaction" suggests that the current need for massive, expensive models may be temporary.

6. Notable Quotes

• "Boring is good. You think about all the stuff around us that our world depends on like electricity. Electricity is very boring. You just plug things into the wall and it just works." — David Cox
• "When it works, it’s software." — Quoting Clem (Founder of Hugging Face) to describe the inevitable commoditization of AI.
• "It’s more like bullshitting... the model has been trained literally to produce plausible sounding answers." — Cox on the nature of AI hallucinations.

Synthesis

The main takeaway is that the current "hype cycle" of AI is transitioning toward a maturation phase. While the industry is currently obsessed with expensive, human-like AGI, the long-term value lies in "boring," efficient, and open-source AI that functions as a reliable utility. For businesses, the path to ROI involves moving away from expensive, proprietary "black box" models and toward purpose-built, secure, and transparent systems that augment human judgment rather than attempting to replace it.