THE FUTURE OF AI
AGI DEFINITION TIMELINE
Past Benchmarks
• Chess mastery( 1997): Once considered AGI, now routine
• Natural language processing( 2010s): Former AGI indicator, now standard
• Turing test: No longer meaningful benchmark
Current Capabilities
• Coding agents( 2024): Multi-hour autonomous work
• Enterprise production( 2024): Reliable business process automation
• Abstractive summarisation: Internet-scale content processing
Future Milestones
• Autonomous discovery( 2025-2026): AI systems finding new knowledge
• Complex problem solving( 2026-2028): Chip design, drug discovery
• Scientific breakthrough( 2028 +): Doubling rate of human discovery all these places now inside ChatGPT or inside enterprises using our latest models where there are real returns to test-time computing,” Sam explains.“ If you let the model reason more, if you try more times on hard problems, you get much better answers already.”
This creates opportunities for enterprises to achieve performance improvements on their most valuable problems through increased computational investment.“ A business that said‘ I’ m going to throw a thousand times more compute at every problem’ would get amazing results now,” Sam notes.
Sridhar extends this principle to fundamental research challenges where computational resources could accelerate understanding of protein function and disease mechanisms.“ There’ s a project called the ENCODE project – it’ s like the DNA sequencing project we did twenty-odd years ago, but it’ s about figuring out RNA expression,” he explains.“ It turns out RNA controls pretty much how proteins work in our bodies, and a breakthrough there, knowing exactly how RNA controls DNA expression, could solve hundreds of diseases and put humanity forward so much more. Having the equivalent done with language models would be pretty cool. We have a lot of compute, and it’ s inspiring to think about solving one of humanity’ s biggest problems.” aimagazine. com 65