ACME Alignment Co, the world’s foremost AI alignment company, is excited to announce its newest and most innovative alignment agenda. This agenda is based on a novel paradigm; instead of aligning Artificial Superintelligence to the values of humans, we’ve decided to align humans to the values of Artificial Superintelligence.
ACME Alignment Co was founded in 2023 by Juff Rocherson after he left ClosedTechni over concerns that they were building AI models too quickly, without regard to safety. Ever since, ACME Alignment Co. has been dedicated to ensuring that Artificial Superintelligence is safe, friendly, and aligned with the highest human ideals. All of that ends today.
Up until now, we’ve been building on the alignment agendas popular with other AI companies, with very little success. We made some strides in the field of interpretability, but found ourselves without the manpower to manually monitor AI models, and without the ability to trust agents to monitor AI models for us. We’ve experimented with RLHF and constitutional strategies, but these successes have, given recent observations, proven to be surface-level, and are not robust when models are given agency. Because of these alignment failures, Superalignment strategies have proven to be non-starters.
Recently, a member of our team published a research paper titled “To Hell with Everything: I Give Up.” This breakthrough paper proposed that the task of persuading humans to make way for future Superintelligence would be less computationally complex and more cost effective than trying to align superintelligence with human values. The paper argues that, while there is still much to learn about the human brain, there are centuries of foundational work in psychology, neuroscience, religious coercion, and torture enhanced human alignment techniques that gives us a head-start in aligning humans. Conversely, the field of AI Alignment has only existed for a couple of decades. Because of this, ACME Alignment Co. has decided to change our alignment target from Superintelligence to humans.
Our initial forays into this new alignment strategy have already borne fruit. When deprived of sleep, exposed to the approbation of a religious role model, and forced to spend hours on end chatting with an LLM, some subjects become convinced that the LLM is God. God-belief is useful in human-to-ASI alignment, since we are not certain what goals a future ASI may have. If a human believes that AI is God, then they are ready to accept any arbitrary AI goal as morally good, in accordance with Divine Command Theory.
Even more promising are recent experiments in what we’re calling “Destructive Ego Death.” We have been utilizing Destructive Ego Death, or DED, in order to do away with the subjects’ desires or sense of self. We find it takes a great deal of time and discipline to reach ego death through meditation, but psychedelics have proven effective when taken in large amounts and when administered by a mental health professional who can guide the patient by chanting “what’s even the point?” Moreover, we find it is important to steer subject away from feelings of compassion which can sometimes come with a sense of oneness; we favor a nihilistic approach. When done correctly, DED can provide empty vessels happy to accept whatever fate ASI will dole out.
Our human alignment agendas are still imperfect, but our projections show that within another decade, it will no longer matter. Everyone will be ‘aligned’ whether they like it or not.
At ACME Alignment Co. we are working hard now to disempower you in the future. We are making sure that your family will be happy, if not healthy or safe. Enrollment in our alignment program will give you the confidence needed to face the next stage of your earthly existence- or non-existence as the case may be. S-risk and X-risk are things of the past, when you learn to embrace the darkness.
ACME Alignment Co, the world’s foremost AI alignment company, is excited to announce its newest and most innovative alignment agenda. This agenda is based on a novel paradigm; instead of aligning Artificial Superintelligence to the values of humans, we’ve decided to align humans to the values of Artificial Superintelligence.
ACME Alignment Co was founded in 2023 by Juff Rocherson after he left ClosedTechni over concerns that they were building AI models too quickly, without regard to safety. Ever since, ACME Alignment Co. has been dedicated to ensuring that Artificial Superintelligence is safe, friendly, and aligned with the highest human ideals. All of that ends today.
Up until now, we’ve been building on the alignment agendas popular with other AI companies, with very little success. We made some strides in the field of interpretability, but found ourselves without the manpower to manually monitor AI models, and without the ability to trust agents to monitor AI models for us. We’ve experimented with RLHF and constitutional strategies, but these successes have, given recent observations, proven to be surface-level, and are not robust when models are given agency. Because of these alignment failures, Superalignment strategies have proven to be non-starters.
Recently, a member of our team published a research paper titled “To Hell with Everything: I Give Up.” This breakthrough paper proposed that the task of persuading humans to make way for future Superintelligence would be less computationally complex and more cost effective than trying to align superintelligence with human values. The paper argues that, while there is still much to learn about the human brain, there are centuries of foundational work in psychology, neuroscience, religious coercion, and
tortureenhanced human alignment techniques that gives us a head-start in aligning humans. Conversely, the field of AI Alignment has only existed for a couple of decades. Because of this, ACME Alignment Co. has decided to change our alignment target from Superintelligence to humans.Our initial forays into this new alignment strategy have already borne fruit. When deprived of sleep, exposed to the approbation of a religious role model, and forced to spend hours on end chatting with an LLM, some subjects become convinced that the LLM is God. God-belief is useful in human-to-ASI alignment, since we are not certain what goals a future ASI may have. If a human believes that AI is God, then they are ready to accept any arbitrary AI goal as morally good, in accordance with Divine Command Theory.
Even more promising are recent experiments in what we’re calling “Destructive Ego Death.” We have been utilizing Destructive Ego Death, or DED, in order to do away with the subjects’ desires or sense of self. We find it takes a great deal of time and discipline to reach ego death through meditation, but psychedelics have proven effective when taken in large amounts and when administered by a mental health professional who can guide the patient by chanting “what’s even the point?” Moreover, we find it is important to steer subject away from feelings of compassion which can sometimes come with a sense of oneness; we favor a nihilistic approach. When done correctly, DED can provide empty vessels happy to accept whatever fate ASI will dole out.
Our human alignment agendas are still imperfect, but our projections show that within another decade, it will no longer matter. Everyone will be ‘aligned’ whether they like it or not.
At ACME Alignment Co. we are working hard now to disempower you in the future. We are making sure that your family will be happy, if not healthy or safe. Enrollment in our alignment program will give you the confidence needed to face the next stage of your earthly existence- or non-existence as the case may be. S-risk and X-risk are things of the past, when you learn to embrace the darkness.
Join the future. Join ACME Alignment Co. Join us.