Why Do LLM Assistants Stop Learning After Deployment?
Rich Sutton says the popular language-model assistants are not learning from experience. Their weights never change. A memory of the user, or extra context, can change the state the model conditions on, and the next word can use that state. The model itself is not learning. The structuring and the new concepts that built the model were weight learning, and that should not happen only once.
He says the field does too much pre-training and post-training before launch, then does not let the model learn while someone is using it. Pre-training and post-training are fine. The disagreement is the stop. Khurram Javed says a few products, including Cursor's tab autocomplete, do update weights from a large batch of many users, which is a poor way to teach one person's own model something specific.
Rich Sutton and Khurram Javed: Why AI Models Stop Learning, and How to Start It Again ·
From the transcript
Like an intelligent assistant, you know, has to become better over time, has to know what you want. Would your contention be that the current paradigm of the popular LLM-based assistants, would your contention be that these are not experiential learners or continual learners? And if so, what is the fundamental gap? Are you serious? I mean, obviously- They learn memories about me.
They're doing some in-context learning. Their weights never change. And by the way, is a small number of the weights changing sufficient or do you need all the weights to be changing? Well, so think of all the structuring and generation of new concepts that went into creating the large language models. All that is the weight learning.