Zero-loss convergence for EM43 multiplication

Determine whether the EM43 Emergent Model can eventually reach zero loss on the two-integer multiplication task when given sufficient training computation.

Background

The two-input multiplication experiment does not converge within the available training budget and performs poorly on both training and test sets. The authors explicitly leave unresolved whether the failure is computationally temporary or reflects limitations in the model, interface, or search procedure.

References

More complex tasks such as a \cdot b showed very slow convergence, and within the available training budget it was not possible to determine whether zero loss would eventually be reached.

— Emergent Models: Intelligence from Tiny Substrates  (2608.14019 - Bocchese et al., 14 Aug 2026) in Section 3.1, “EM43,” Results discussion