This is an automated rejection. No LLM generated, assisted/co-written, or edited work.
Read full explanation
Breakthroughs do not always come from within a field; sometimes they come from those unbound by its assumptions. Freed from its conceptual constraints, outsiders might be able to reframe problems in ways insiders rarely do. For example, in 1828, George Green, a self-taught miller with no place in the scientific establishment, introduced what are now known as Green’s functions—tools that became foundational across physics and engineering.
Certainly, an expert of your level does not require a reminder about the urgency of the fundamental problem of AI risk and the growing need to find ways of efficient 𝗔𝗜 𝗮𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁. As the gap between what we can build and what we can control is widening, we can not afford to ignore even the most unconventional approaches.
Such a breakthrough from an outsider might come from my friend, Prof. Michael Zibulevsky (Technion – Israel Institute of Technology), whose main professional work is in the domain of optimization. Over the past few years, he has taken a deep personal interest in AI alignment, approaching it not as a formal research program but as an independent line of thought developed alongside his primary work. During this time, he has been steadily reflecting on a particular direction, revisiting and refining it over an extended period. While this effort has been informal in nature, it has matured enough to take the form of a coherent perspective shaped by sustained, long-term consideration.
The Professor's 𝘂𝗻𝗰𝗼𝗻𝘃𝗲𝗻𝘁𝗶𝗼𝗻𝗮𝗹 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵 𝗵𝗮𝘀 𝗿𝗲𝗰𝗲𝗻𝘁𝗹𝘆 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝗱 𝗮𝗰𝘁𝗶𝘃𝗲 𝗱𝗶𝘀𝗰𝘂𝘀𝘀𝗶𝗼𝗻 in informal circles, where it continues to provoke sustained engagement. While this is not evidence in itself, such reactions can occasionally indicate that this approach is probing a genuinely ignored direction—at a time when alignment may require exactly that.
At this stage, the only thing missing is a clear signal from a domain expert. A concise, high-level judgment would be sufficient to determine whether this direction should be taken further, or set aside without additional investment. Even a brief assessment at this level would provide the clarity needed to decide the next step. In case that you are a domain expert and would be ready to provide you feedback, feel free to take a read. The article looks rather long, but no need to be scared, since most of it is the appendix. Here it is:
Breakthroughs do not always come from within a field; sometimes they come from those unbound by its assumptions. Freed from its conceptual constraints, outsiders might be able to reframe problems in ways insiders rarely do. For example, in 1828, George Green, a self-taught miller with no place in the scientific establishment, introduced what are now known as Green’s functions—tools that became foundational across physics and engineering.
Certainly, an expert of your level does not require a reminder about the urgency of the fundamental problem of AI risk and the growing need to find ways of efficient 𝗔𝗜 𝗮𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁. As the gap between what we can build and what we can control is widening, we can not afford to ignore even the most unconventional approaches.
Such a breakthrough from an outsider might come from my friend, Prof. Michael Zibulevsky (Technion – Israel Institute of Technology), whose main professional work is in the domain of optimization. Over the past few years, he has taken a deep personal interest in AI alignment, approaching it not as a formal research program but as an independent line of thought developed alongside his primary work. During this time, he has been steadily reflecting on a particular direction, revisiting and refining it over an extended period. While this effort has been informal in nature, it has matured enough to take the form of a coherent perspective shaped by sustained, long-term consideration.
The Professor's 𝘂𝗻𝗰𝗼𝗻𝘃𝗲𝗻𝘁𝗶𝗼𝗻𝗮𝗹 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵 𝗵𝗮𝘀 𝗿𝗲𝗰𝗲𝗻𝘁𝗹𝘆 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝗱 𝗮𝗰𝘁𝗶𝘃𝗲 𝗱𝗶𝘀𝗰𝘂𝘀𝘀𝗶𝗼𝗻 in informal circles, where it continues to provoke sustained engagement. While this is not evidence in itself, such reactions can occasionally indicate that this approach is probing a genuinely ignored direction—at a time when alignment may require exactly that.
At this stage, the only thing missing is a clear signal from a domain expert. A concise, high-level judgment would be sufficient to determine whether this direction should be taken further, or set aside without additional investment. Even a brief assessment at this level would provide the clarity needed to decide the next step. In case that you are a domain expert and would be ready to provide you feedback, feel free to take a read. The article looks rather long, but no need to be scared, since most of it is the appendix. Here it is:
https://medium.com/@michaelzibulevsky/motis-journey-growing-aligned-superintelligence-from-infancy-d6d7894c134b