We propose the Indirect Neural Corrector (INC) for hybrid PDE solvers, integrating corrections into PDEs to reduce errors, enable long-term rollouts, and speed-up chaotic/turbulent systems.