The Gentle Singularity

a year ago

Humanity is close to building digital superintelligence, which is less weird than expected.
Current AI systems like GPT-4 and o3 are already smarter than humans in many ways and amplify human output.
AI will significantly improve quality of life through faster scientific progress and increased productivity.
ChatGPT is already more powerful than any human, impacting hundreds of millions daily.
By 2025, AI agents will handle real cognitive work; by 2026, they may generate novel insights; by 2027, robots could perform real-world tasks.
The 2030s will see abundant intelligence and energy, fundamentally changing human progress.
AI is accelerating scientific research, potentially compressing a decade’s work into a year or month.
Self-reinforcing loops, like economic value creation and automated infrastructure, will further accelerate progress.
The cost of intelligence may converge to the cost of electricity as datacenter production becomes automated.
Technological progress will keep accelerating, with society adapting to rapid changes and new policy ideas.
Challenges include solving AI safety issues and ensuring widespread access to superintelligence.
The best path forward involves solving alignment problems and making superintelligence cheap and widely available.
OpenAI is focused on superintelligence research, with most of the path now lit and dark areas receding.
Intelligence becoming too cheap to meter is within reach, with predictions for 2030 seeming less crazy than past predictions.

Hasty Briefsbeta