Reachability analysis determines the set of all states a system can reach under uncertainties in initial conditions, inputs, and parameters. It plays a central role in the formal verification of cyber-physical systems (CPS), ensuring that unsafe states can be systematically avoided. This talk highlights recent advances in reachability analysis across continuous and discrete domains, with an emphasis on applications to large language models (LLMs). In the continuous domain, the challenge of constructing accurate models for CPS is addressed through data-driven reachability analysis algorithms that rely solely on input/output data while providing formal guarantees. The effectiveness of these methods is demonstrated through applications such as ensuring safety for an LLM-controlled robot. In the discrete domain, the focus shifts to logical zonotopes and polynomial logical zonotopes, novel binary set representations capable of compactly capturing exponentially large Boolean state spaces. These tools enable efficient verification and model checking of discrete systems and open new directions for applications such as cryptographic key search.