Using Computer Vision for Analysis of Collapsed Bose-Einstein Condensate in Dual Sagnac Interferometer; The Rise of Generative AI: Emergent Intelligence, Extreme Risks, and the Race Toward AGI

This thesis examines two notable machine learning developments: the rapid creation of large language models and generative AI, and the application of computer vision to analyze Bose-Einstein condensates in atom interferometer gyroscopes for navigation.

"The Rise of Generative AI: Emergent Intelligence, Extreme Risks, and the Race Toward AGI", provides an overview of the sudden progress in AI capabilities driven by transformer-based architectures. It highlights the emergence of sophisticated abilities that match or exceed human performance across diverse domains. However, the creators of these systems have limited understanding and control over what these models can do. The paper outlines extreme risks posed by current and near-future AI, including the creation of bioweapons, autonomous weapons, and amplified misinformation. Competitive pressures are causing AI labs to deprioritize safety and rush to deploy ever more powerful systems. Many experts believe artificial general intelligence (AGI) that meets or surpasses human-level intelligence is plausible in the near-term. Lastly, the paper discusses how the losing control of advanced AI could pose an existential threat for humanity.

"Using Computer Vision for Analysis of Collapsed Bose-Einstein Condensate in Dual Sagnac Interferometer ", demonstrates a novel approach for predicting the rotation angle of an atom interferometer gyroscope by applying machine learning to images of a collapsed Bose-Einstein condensate. Efficiently measuring orientation changes is crucial for navigation, especially where conventional GPS is unreliable. The proposed solution uses principal component analysis and multivariate linear regression to construct an optimal basis image that captures interference patterns correlated with the rotation angle. The results show a 0.91 correlation between image-derived scores and true angles, suggesting that gyroscope orientation can be largely inferred from the collapsed BEC images using simple machine learning methods. Further work is proposed to improve sensitivity and robustness.