Sam Altman Unveils ChatGPT’s Energy Costs and Superintelligence Path

The CEO of OpenAI, Sam Altman, recently shared insights in a blog post entitled “The Gentle Singularity,” where he discussed the energy consumption associated with each query made to ChatGPT. Altman indicated that, on average, a single query consumes approximately 0.34 watt-hours. To put that into perspective, this amount of energy is comparable to what an oven uses in just over a second or what a high-efficiency light bulb uses in a few minutes.

Additionally, Altman highlighted that each query also utilizes about 0.000085 gallons of water, which is roughly equivalent to one-fifteenth of a teaspoon. He suggested that the overall cost of intelligence is likely to align closely with electrical expenses in the future.

What particularly stood out was Altman’s dramatic opening statement about our rapid progression toward superintelligence.

“We have crossed the event horizon; the takeoff has begun. Humanity is on the verge of creating digital superintelligence, and so far, it is less bizarre than one might expect.”

Altman appears to be informing us that we are on the brink of advancements in transformative AI technology. He further elaborated, stating, “The hardest part of the journey is behind us; the scientific breakthroughs that led us to systems like GPT-4 and others were challenging, but they will propel us forward significantly.”

This statement implies that existing AI technologies could pave the way toward Artificial General Intelligence (AGI) and, eventually, Artificial Superintelligence (ASI). This view contrasts with that of several AI skeptics, including Yann LeCun, the chief scientist at Meta AI, who argue that large language models (LLMs) have reached their limits and are not capable of leading us to AGI or ASI.

Altman outlined a timeline for upcoming advancements:

• 2025: Arrival of agents capable of performing cognitive tasks like coding
• 2026: Introduction of AI systems that can generate novel insights
• 2027: Development of robots able to undertake real-world tasks
• 2030: Significant productivity boosts for individuals
• 2035: Emergence of brain-computer interfaces, potentially aiding in space colonization

The Future of AI

Altman also mentioned the concept of “recursive self-improvement,” where AI systems begin to enhance their own capabilities autonomously. He explained that current AI technologies are not fully independent yet, but they represent an early form of this self-improvement. Existing AI tools are already helping in creating better systems, implying that future technologies could lead to even more sophisticated AI.

Experts at OpenAI are reporting significant productivity boosts, with some scientists noting they have become two to three times more efficient thanks to current AI systems. However, Altman acknowledges that this technological transformation will likely lead to significant job displacement and societal upheaval.

He stated, “There will be tough challenges ahead, such as entire job sectors disappearing. Yet, on the flip side, the world will become so much wealthier so quickly that we might be able to explore policy ideas we couldn’t previously consider.”

In a recent talk at the University of Toronto, OpenAI’s former chief scientist, Ilya Sutskever, echoed these sentiments:

From AI researchers to industry leaders, many view AI as a game-changing technology that could lead us into an era of abundance. However, the journey to that future is likely to involve various societal challenges. Whether this ambitious vision will ultimately become a reality remains to be seen.