‍Introduction

The AI revolution is here, but it's not what you think. While chatbots dazzle us with poetry and code, true machine intelligence remains elusive. This article dives into the heart of modern AI's paradox, exploring how the GPU gold rush might unlock superintelligence. It also looks at AI's resilience through periods of reduced funding and interest, and offers guidance on how organizations can begin implementing AI solutions amidst technological uncertainties.

Current AI: Impressive Yet very Limited

ChatGPT can write poetry, debug code, and even pass law exams, yet according to Yann LeCun, Meta's AI chief scientist, it's not even as intelligent as a dog (Novet, 2023). This paradox lies at the heart of our current AI conundrum: systems that appear remarkably clever in domains yet lack the fundamental attributes of true intelligence.

LeCun argues that true intelligence extends far beyond language processing. He posits that most human knowledge has little to do with language, and that current AI systems, including Large Language Models (LLMs) like ChatGPT, lack crucial aspects of intelligence such as emotions, creativity, sentience, and consciousness. Moreover, these systems fall short in their ability to sense, plan, exhibit common sense, or reason based on real-world experiences.

Despite these limitations, the capabilities of current AI systems are undeniably impressive. They can process and generate human-like text, translate between languages, and even create art. However, their intelligence remains narrow and specialized, far from the general intelligence exhibited by humans or even animals. Is it enough to unlock economic productivity? Yes. Does it make us obsolete? Far from it.

LLMs: Catalysts for AI Adoption and Development

While artificial intelligence has been a field of study since the 1950s, it was the advent of Large Language Models (LLMs) that made AI accessible for the masses. These models, exemplified by tools like ChatGPT, have not only captured the public imagination but have also ignited a firestorm of investment and development in AI technologies, leading to a rush for hardware that enables training and scaling of AI models (Schor, 2023).

Scale and GPUs: Enablers of Superintelligence?

In the realm of AI, bigger often means better. This principle, championed by researchers like Richard Sutton, posits that major AI breakthroughs are more likely to come from scalable methods that can harness ever-increasing computational power, rather than from encoding human knowledge (Sutton, 2019). In other words, he argues that game changing AI requires more computation power and new approaches to training AI. An approach that goes beyond dumping labeled human knowledge into data models.

Speaking of computational power, LLMs have sparked a new gold rush into the developments of GPUs. The GPU is the workhorse of modern AI training, whose performance-to-price ratio doubles roughly every two and a half years (Epoch, 2023). With more capital flowing into the R&D of companies like NVIDIA, there is no doubt that the research of new computational hardware will accelerate quicker than ever before. Will it one day be enough for Superintelligence? We can’t tell today, however what we know is that this acceleration allows larger and more complex models to be trained at a lower cost.

Navigating the uncertainty

The unpredictable nature of AI advancement presents both challenges and opportunities for organizations. While the potential of AI is immense, its development path is far from linear. The field has experienced periods of rapid progress followed by "AI winters" - times of reduced funding and interest. Today's AI boom could either lead to transformative breakthroughs or face a slowdown if expectations aren't met.

Given this uncertainty, organizations must adopt a balanced approach to AI integration. Rather than betting everything on speculative future capabilities, they should focus on implementing practical, measurable AI initiatives that deliver immediate value. This could involve using current AI technologies to automate routine tasks, enhance decision-making processes, or improve customer interactions.

At the same time, organizations should remain flexible and open to emerging AI technologies. By building a culture of continuous learning and adaptation, they can position themselves to quickly leverage new AI breakthroughs as they occur. This might involve setting up cross-functional AI teams, investing in employee training, or partnering with AI research institutions.

Summary

• Current AI systems, while impressive, lack true general intelligence and fundamental aspects of human-like cognition.
• The advent of Large Language Models has catalyzed widespread AI adoption and investment.
• Scaling computational power through advanced GPUs may be key to achieving breakthroughs in AI capabilities.
• Organizations must prepare for AI adoption with measurable, simple initiatives while remaining adaptable to future developments.

Sources

1. Epoch. (2023). Trends in GPU Price-Performance. Retrieved from https://epochai.org/blog/trends-in-gpu-price-performance‍
2. Novet, J. (2023). A.I. is not even at dog-level intelligence yet: Meta A.I. chief. CNBC. Retrieved from https://www.cnbc.com/‍
3. Schor, D. (2023). AI Capacity Constraints - CoWoS and HBM Supply Chain. SemiAnalysis. Retrieved from https://www.semianalysis.com/p/ai-capacity-constraints-cowos-and‍
4. Sutton, R. (2019). The Bitter Lesson. Retrieved from http://www.incompleteideas.net/IncIdeas/BitterLesson.html

‍