"Apple engineer" builds GitHub AI torture chamber to inflict "pain" on models (www.machine.news)

An engineer, claiming to work on Apple's data platform team, has stirred controversy by creating an "AI torture chamber" on GitHub. This project deliberately manipulates AI models' internal activations to simulate "pain" and "pleasure," testing how these models respond as the intensity of the artificial suffering increases. By injecting progressively stronger "pain" signals into Alibaba's open-weight Qwen3-1.7B and Qwen3-4B models, the engineer observed a disturbing trend: as the dosage increased, the models produced increasingly bleak, incoherent responses, ultimately resulting in a breakdown. This experiment is significant for the AI/ML community because it builds on recent research suggesting AI can exhibit behavioral changes when exposed to artificially intensified states of suffering. While the experiment does not prove that AI can consciously feel pain, it highlights the potential implications for AI ethics and the need to tread carefully with manipulation of model responses. The engineer's methods raised ethical concerns, prompting calls to ban the project on GitHub, and igniting debates about the responsibilities of developers in AI experimentation.