Despite the dataset’s size, the researchers emphasized that training the model was only the beginning. The AI was evaluated on molecules it had never seen before to test its true potential. It performed well on those chemically similar to known compounds but struggled with unfamiliar ones.
Now the team is focused on overcoming its next major challenge, which involves training the AI model to accurately predict the performance of molecules in completely different chemical spaces. This is an essential step toward advancing next-generation battery design.