KnowraAI alignmentLinked fromLinked fromThe 16 pages that link to AI alignment, each with the reason it gives.All 16Related 14Compared with 2Artificial intelligenceRelated: More capable systems make the consequences of poorly specified objectives more consequential.Artificial general intelligenceRelated: A capable general system may pursue goals in ways its designers did not intend.Geoffrey HintonRelated: It addresses the control problem raised by Hinton’s concerns about increasingly capable systems.AI safetyRelated: It addresses the risk that capable systems pursue goals different from those intended.Existential riskRelated: It addresses one proposed route from advanced AI capability to loss of human control.Technological singularityRelated: Loss of control is a central concern when singularity scenarios involve superhuman AI.Nick BostromRelated: Bostrom’s control problem helped motivate this broader research program.Effective accelerationismRelated: Alignment concerns challenge any assumption that faster AI progress is automatically beneficial.ChatGPTRelated: ChatGPT's instruction-following and safety behavior raise practical alignment questions.Existential risk from artificial intelligenceRelated: Failures to align advanced systems with human intentions are a central proposed pathway to loss of control.HAL 9000Related: HAL’s conflicting instructions dramatize what happens when objectives cannot be reconciled.Philosophy of artificial intelligenceRelated: It turns disputes about value and intention into a problem of system design.Ray KurzweilCompared with: Kurzweil emphasizes the potential benefits of advanced AI, while alignment work foregrounds control and safety problems.AI boomRelated: Scaling capability makes the question of reliably directing systems more urgent.Andrew NgCompared with: Ng’s focus on near-term deployment risks differs from approaches centered on controlling hypothetical superintelligence.Generative AIRelated: A generator can satisfy a prompt while producing harmful or unwanted content.
KnowraAI alignmentLinked fromLinked fromThe 16 pages that link to AI alignment, each with the reason it gives.All 16Related 14Compared with 2Artificial intelligenceRelated: More capable systems make the consequences of poorly specified objectives more consequential.Artificial general intelligenceRelated: A capable general system may pursue goals in ways its designers did not intend.Geoffrey HintonRelated: It addresses the control problem raised by Hinton’s concerns about increasingly capable systems.AI safetyRelated: It addresses the risk that capable systems pursue goals different from those intended.Existential riskRelated: It addresses one proposed route from advanced AI capability to loss of human control.Technological singularityRelated: Loss of control is a central concern when singularity scenarios involve superhuman AI.Nick BostromRelated: Bostrom’s control problem helped motivate this broader research program.Effective accelerationismRelated: Alignment concerns challenge any assumption that faster AI progress is automatically beneficial.ChatGPTRelated: ChatGPT's instruction-following and safety behavior raise practical alignment questions.Existential risk from artificial intelligenceRelated: Failures to align advanced systems with human intentions are a central proposed pathway to loss of control.HAL 9000Related: HAL’s conflicting instructions dramatize what happens when objectives cannot be reconciled.Philosophy of artificial intelligenceRelated: It turns disputes about value and intention into a problem of system design.Ray KurzweilCompared with: Kurzweil emphasizes the potential benefits of advanced AI, while alignment work foregrounds control and safety problems.AI boomRelated: Scaling capability makes the question of reliably directing systems more urgent.Andrew NgCompared with: Ng’s focus on near-term deployment risks differs from approaches centered on controlling hypothetical superintelligence.Generative AIRelated: A generator can satisfy a prompt while producing harmful or unwanted content.