However, the large language models did not provide information to users already working to change their behaviors about using a reward system to maintain motivation or about reducing environmental stimuli that might increase the risk of a relapse, the researchers found.