From Simulation to Enaction: Post-trained language models recognize and react to their own generations

1. 1Train a language model using a pre-training dataset
2. 2Post-train the model by generating its own responses and adjusting the output distributions
3. 3Evaluate the model's ability to recognize its on-policy generations using metrics such as output distribution entropy
4. 4Fine-tune the model to improve its performance on specific tasks
5. 5Test the model's ability to react to its own generations in real-world scenarios

AI engineers and researchers can benefit from understanding how post-trained models recognize their own generations, allowing them to develop more effective language models

💡 Post-trained models can implicitly encode recognition of their on-policy generations in their output distributions

🤖 Post-trained language models can recognize & react to their own generations! 📊