-->

OpenAI CEO Sam Altman published a post titled “The Gentle Singularity” on his blog. In this piece, he provides deep insights into the current state of AI development and future prospects.

The Beginning of the Singularity: We’ve Already Crossed the Event Horizon

Altman declares that “we have crossed the event horizon, and the leap has begun.” Humanity has approached building digital superintelligence, and the situation so far has been much less bizarre than expected.

The Paradox of the Current Situation

  • Robots aren’t walking the streets, and most people aren’t talking to AI all day
  • People still die from diseases, we can’t easily go to space, and there’s much we don’t know about the universe
  • But simultaneously, we’ve built systems smarter than humans in many ways and are significantly amplifying the productivity of people who use them

ChatGPT’s Real Impact

Altman describes ChatGPT as already being “more powerful than any human who has ever lived.” Hundreds of millions of people depend on it daily and are increasingly using it for important tasks.

2025-2027: Heralding Rapid Change

2025: Agents Performing Real Cognitive Tasks

  • The way computer code is written will change forever

2026: Systems Discovering New Insights

  • Systems that find new insights that didn’t exist before will emerge

2027: Real-World Task Robots

  • Robots capable of performing tasks in the actual physical world will arrive

2030s: Abundance of Intelligence and Energy

Altman predicts that intelligence and energy will become abundant in the 2030s. These two have long been fundamental constraints on human progress.

Current Productivity Increases for Scientists

  • Scientists are already reporting being 2-3 times more productive with AI assistance
  • Using AI for AI research is particularly important
  • “If we can do 10 years of research in 1 year or 1 month, the pace of progress will be completely different”

The Beginning of Self-Reinforcing Loops

Technical Self-Reinforcement

  • The tools we’ve already built help create better AI systems
  • While not fully autonomous code updates, it’s an early form of recursive self-improvement

Economic Self-Reinforcement

  • Economic value creation has started the flywheel of building compound infrastructure to run more powerful AI systems
  • The era of robots making other robots and data centers building other data centers is not far away

Cost Reduction and Improved Accessibility

Convergence to Electricity Costs

  • As data center production becomes automated, the cost of intelligence will eventually approach electricity costs
  • A ChatGPT query uses an average of 0.34 watt-hours (the amount of power an oven uses for just over a second)

Energy Efficiency

  • Similar to the power a high-efficiency light bulb uses for a few minutes
  • Water usage is about 0.000085 gallons (about 1/15 of a teaspoon)

Social Adaptation and New Jobs

Altman is optimistic about humanity’s ability to adapt to new technologies, citing job changes since the Industrial Revolution.

A Farmer’s Perspective from a Thousand Years Ago

  • A subsistence farmer from a thousand years ago would think our current jobs are “fake jobs”
  • Similarly, jobs a thousand years from now might seem “fake” to us, but they’ll be very important and satisfying to people of that era

Safety and Accessibility: Important Challenges

Two Key Challenges

  1. Solving the alignment problem: Ensuring AI systems learn and act toward what we collectively want
  2. Broad accessibility of superintelligence: Making it cheap and widely available, not concentrated in specific individuals, companies, or countries

Lessons from Social Media Feeds

  • Current social media algorithms are examples of unaligned AI
  • They understand short-term preferences but exploit brain vulnerabilities by ignoring long-term preferences

Building the “World’s Brain”

Altman describes the entire industry, including OpenAI, as building “the world’s brain.”

Characteristics

  • Highly personalized and easily accessible to everyone
  • Will only be limited by good ideas
  • The era of “idea guys” will come

OpenAI’s Identity and Future

Superintelligence Research Company

  • OpenAI is many things, but above all, it’s a superintelligence research company
  • Most of the path ahead has now been illuminated, and the dark areas are rapidly receding

Reality of 2030 Predictions

  • If we had predicted the current situation in 2020, it would have sounded even crazier, emphasizing the high feasibility of 2030 predictions

Conclusion: Smooth and Exponential Scaling

Altman concludes with “scaling smoothly, exponentially, and serenely through superintelligence.”

“Intelligence too cheap to meter” is within reach, and this will be the starting point of the greatest transformation in human history.

Original Link: The Gentle Singularity - Sam Altman

This piece contains candid prospects from one of the most influential figures in AI development, making it an important reference for anyone interested in the AI industry and technological advancement.

참고