With contributions from: Collin Burns, Avital Balwit, Carl Shulman, Jan Leike, Ilya Sutskever, Holden Karnofsky, Sholto Douglas, James Bradbury, Dwarkesh Patel.

“Dedicated to Ilya Sutskever”
situational-awareness.ai

June 22, 2024,
Franc Smidt,
Editor, FUTURUM magazine, Germany

Leopold Aschenbrenner’s publication contains key aspects and forecasts regarding the future of artificial intelligence (AI) and its impact on the global community. We read the extensive 165-page document and summarized the main points for our readers in this overview.

Main Points

The Emergence of AGI by 2027 is Quite Realistic Aschenbrenner asserts that the progress from GPT-2 to GPT-4 over four years demonstrates that the emergence of artificial general intelligence (AGI) by 2027 is quite achievable. This progress is based on current trends in increasing computational power (~0.5 orders of magnitude per year) and algorithmic improvements (~0.5 OOM per year). If these trends continue, we can expect a new leap in AI model development, similar to the transition from GPT-2 to GPT-4. “The progress from GPT-2 to GPT-4 has shown that AI models can quickly transition from simple tasks to complex ones, such as coding and solving mathematical problems.” This progress has been achieved through increased computational power and algorithmic improvements like RLHF and Chain of Thought.

AGI as the Most Important Geopolitical Resource AGI is becoming the most important geopolitical resource, replacing nuclear weapons as the main instrument of global influence. The desire to possess AGI is driven by its potential to radically change the economy, military capabilities, and the global balance of power. Aschenbrenner emphasizes that “every country is willing to do whatever it takes to be the first to acquire AGI, just like the atomic bomb.” “Every country strives to be the first to develop AGI to gain economic and military superiority.” Possessing AGI will provide countries that develop it first with enormous advantages, allowing them to dominate international politics and economics.

Creating AGI Will Require a Unified Computing Cluster Worth a Trillion Dollars Aschenbrenner provides figures to underscore the scale of the project: “Models like GPT-4 used 8e24 to 4e25 flops (floating-point operations) for their training. Future generations of models are expected to require exponentially larger volumes of computational resources.” Aschenbrenner describes a project already being implemented by Microsoft for OpenAI. “The construction of a trillion-dollar cluster has already begun. This cluster will include hundreds of millions of GPUs and consume an enormous amount of electricity.” These computational resources are necessary for training and operating future AGI models.

Energy Consumption of the Computing Cluster It is expected that this cluster will consume more electricity than the entire United States combined. Supporting it will require a significant increase in electricity production. Aschenbrenner writes: “By the end of the decade, electricity production in the US will increase by tens of percent. From oil fields in Pennsylvania to solar farms in Nevada, hundreds of millions of GPUs will be working, requiring a colossal increase in energy consumption.” The document also highlights the need for investments in energy infrastructure and the development of new technologies to ensure sustainable growth in energy consumption.

Funding from Big Tech Aschenbrenner points out the role of the largest tech companies in funding AGI development. “AGI funding will come from the largest tech companies, such as Microsoft and Google, which already allocate $100 billion to AI quarterly.” These investments will continue to grow, and by 2030, annual AI investments will reach $8 trillion. This underscores the importance and potential benefits of AGI development for tech giants.

AGI is Just the Beginning The creation of AGI will lead to an almost immediate transition to ASI, which will have even more significant consequences for humanity. Aschenbrenner writes: “Superintelligence (ASI) will significantly surpass AGI in capabilities and power, opening new horizons and creating new challenges for humanity.” Superintelligence will be able to solve tasks that seem impossible today and offer solutions to global problems such as climate change and diseases. However, these capabilities will also bring new risks related to controlling and ensuring the safety of superintelligence.

Conclusion Leopold Aschenbrenner’s report “Situational Awareness: The Decade Ahead” provides a comprehensive analysis of the current state and future prospects of AI development. Aschenbrenner emphasizes the importance of strategic planning and coordinated efforts for the successful and safe development of AGI and ASI. Industrial mobilization, large-scale investments, and attention to safety are key elements necessary to achieve these goals. This document is an important contribution to the discussion of AI’s future and related challenges and opportunities, offering deep analysis and practical recommendations for ensuring the safe and effective development of technologies.

Read the report: situational-awareness.ai