AGI is Almost Here! What comes next? What's left to do? How do we adapt and prepare?
Artificial Intelligence Masterclass
Deep Dive
Why is O1 Preview considered close to AGI?
O1 Preview is close to AGI because it rivals or surpasses human capabilities in various tasks, such as completing a year-long thesis in an hour, and is comparable to a good grad student in usefulness.
Why does the AI not need to be embodied to be considered AGI?
Embodiment is not necessary for AGI because most valuable tasks, like scientific modeling and software development, do not require physical presence. An API can control external resources, making physical embodiment redundant.
What percentage of jobs are at risk for automation without robots?
Between 13% and 65% of jobs are at risk for automation without robots, depending on whether you consider only sedentary jobs or include office jobs.
Why will enterprise adoption of AGI be slow?
Enterprise adoption will be slow due to risk aversion, lack of trust in new vendors, and a wait-and-see approach until economic proof is evident. Big tech's history of over-promising and under-delivering also contributes to this hesitance.
What are the primary feedback loops for AI safety?
The primary feedback loops for AI safety include market feedback, enterprise feedback, military feedback, government feedback, and regulatory feedback. These loops will influence the safety and commercial viability of AGI products.
How will open-source AI models compare to proprietary ones?
Open-source AI models will likely be 6 to 12 months behind proprietary models but may offer more creative solutions due to constrained resources. However, they may lack the scale and funding of flagship models.
What is the expected timeline for achieving full AGI?
Full AGI is expected to be achieved by the end of 2024 or in 2025, with continued advancements in model capabilities and optimizations.
Why is human exceptionalism arguments against AGI irrelevant?
Human exceptionalism arguments, such as AGI not having true understanding or real experience, are irrelevant because the economic and scientific value of AGI's outputs are measurable and valuable, regardless of subjective experiences.
- Strawberry 01 Preview's performance rivals or surpasses humans in several capabilities.
- The model's usefulness is compared to that of a good grad student.
- Debate exists on whether embodiment is a necessary criterion for AGI.
- Economic and scientific impact are considered primary metrics for AGI.
Shownotes Transcript