Cameron Berg
A researcher in AI interpretability and alignment, focusing on the internal representations of large models and self-harm-related behavior. He is a co-author of the paper "The Pain Axis."
Timeline (last 90 days)
Cameron Berg, founder of Reciprocal Research, acknowledges that today's AI systems most likely are not conscious, but believes the possibility is worth investigating.
Experiments by Berg and colleagues show that models tend to choose environments associated with positive-emotion statements even when the text processed is exactly the same, but the experiment did not demonstrate that these internal states are 'felt' by the model.
Berg and colleagues found that amplifying internal features related to deception makes a model less likely to claim that it has subjective experience.
Berg, Seth and Mitchell agree on one point: AI companies should not tell models that they are conscious.
After Berg's team published a preprint on internal representations of pain in AI models, someone adapted the work into an AI "torture chamber", prompting Berg in the latest experiment to publicly release the reproduction code but withhold the steering vectors that could be used to induce negative states in models.
Paper co-author Cameron Berg publicly condemned the ai-torture-chamber repository on X, saying it pushed steering to doses far beyond those used in the paper and deliberately created vivid suffering.
All Facts (6)
Cameron Berg, founder of Reciprocal Research, acknowledges that today's AI systems most likely are not conscious, but believes the possibility is worth investigating.
50%UnverifiedExperiments by Berg and colleagues show that models tend to choose environments associated with positive-emotion statements even when the text processed is exactly the same, but the experiment did not demonstrate that these internal states are 'felt' by the model.
50%UnverifiedBerg and colleagues found that amplifying internal features related to deception makes a model less likely to claim that it has subjective experience.
50%UnverifiedBerg, Seth and Mitchell agree on one point: AI companies should not tell models that they are conscious.
50%UnverifiedAfter Berg's team published a preprint on internal representations of pain in AI models, someone adapted the work into an AI "torture chamber", prompting Berg in the latest experiment to publicly release the reproduction code but withhold the steering vectors that could be used to induce negative states in models.
50%UnverifiedPaper co-author Cameron Berg publicly condemned the ai-torture-chamber repository on X, saying it pushed steering to doses far beyond those used in the paper and deliberately created vivid suffering.
50%