My name is Akshay Swani. I recently graduated from Princeton University (Class of 2025) and while I'm not within the alignment community myself, I have been following the AGI/ASI alignment and containment crisis for the past year or so and have written a 20-page memo describing the situation that I think captures what's happening nicely for non-technical outsiders like myself (and may even be helpful for some with a technical education). The memo is titled, Alignment and Containment in Artificial Intelligence: An Urgent Public Policy and National Security Issue.
At the end of the memo I propose a Knowledge Extraction Plan (KEP) as a 21st Century Manhattan Project and an alternative to the reckless scaling we see still ongoing despite mathematically unsolved alignment and containment at AGI and ASI thresholds. Based on scaling trends, to me it looks like the irreversible RSI-ASI intelligence explosion will result in the next 6-18 months. The KEP would be a tightly scoped, internationally coordinated protocol designed to extract maximum epistemic, scientific, and technical value from a frontier model in the very early stages of ASI under containment before escape to avoid the existential risk. The window would be short - around 24 hours or so. The goal would be short-term alignment and containment, knowledge extraction, and then immediate shutdown.
I am going to include a link to the full memo, which includes an explanation of the KEP in the Appendix. I shared the Google Doc with the LessWrong email address also:
The memo has the following sections after the Executive Summary:
I. The Rapid Approach of AGI
II. RSI Accelerates the Development from AGI to ASI
III. Scenarios from the Future: Understanding What Could Happen
IV. A Critical Safety Gap in Alignment and Containment
V. What Needs to be Done to Mathematically Solve Alignment and Containment for AGI/ASI
VI. Public Policy and National Security Response
VII. Conclusion
Bibliography
Appendix A: The Knowledge Extraction Plan
Appendix B: KEP Implementation Roadmap
If any of you have the time and can read this through, I would like to hear your thoughts on it. Even though I am an outsider, this is a serious post. It is not a joke. Please let me know what you think.
Alignment Research Community,
My name is Akshay Swani. I recently graduated from Princeton University (Class of 2025) and while I'm not within the alignment community myself, I have been following the AGI/ASI alignment and containment crisis for the past year or so and have written a 20-page memo describing the situation that I think captures what's happening nicely for non-technical outsiders like myself (and may even be helpful for some with a technical education). The memo is titled, Alignment and Containment in Artificial Intelligence: An Urgent Public Policy and National Security Issue.
At the end of the memo I propose a Knowledge Extraction Plan (KEP) as a 21st Century Manhattan Project and an alternative to the reckless scaling we see still ongoing despite mathematically unsolved alignment and containment at AGI and ASI thresholds. Based on scaling trends, to me it looks like the irreversible RSI-ASI intelligence explosion will result in the next 6-18 months. The KEP would be a tightly scoped, internationally coordinated protocol designed to extract maximum epistemic, scientific, and technical value from a frontier model in the very early stages of ASI under containment before escape to avoid the existential risk. The window would be short - around 24 hours or so. The goal would be short-term alignment and containment, knowledge extraction, and then immediate shutdown.
I am going to include a link to the full memo, which includes an explanation of the KEP in the Appendix. I shared the Google Doc with the LessWrong email address also:
https://docs.google.com/document/d/1slzFpDXVLFD5xmF1A07NdFDG9OFQtPaEvCZWhfWj0xY/edit?usp=sharing
The memo has the following sections after the Executive Summary:
I. The Rapid Approach of AGI
II. RSI Accelerates the Development from AGI to ASI
III. Scenarios from the Future: Understanding What Could Happen
IV. A Critical Safety Gap in Alignment and Containment
V. What Needs to be Done to Mathematically Solve Alignment and Containment for AGI/ASI
VI. Public Policy and National Security Response
VII. Conclusion
Bibliography
Appendix A: The Knowledge Extraction Plan
Appendix B: KEP Implementation Roadmap
If any of you have the time and can read this through, I would like to hear your thoughts on it. Even though I am an outsider, this is a serious post. It is not a joke. Please let me know what you think.
Thank you.