MIT researchers found that massive neural network models that are similar to large language models are capable of containing smaller linear models inside their hidden layers, which the large models could train to complete a new task using simple learning algorithms.
Image: Jose-Luis Olivares, MIT
Large language models like OpenAI’s GPT-3 are massive neural networks that can generate human-like text, from poetry to programming code. Trained using troves of internet data, these machine-learning models take a small bit of input text and then predict the text that is likely to come next.
But that’s not all these models can do. Researchers are exploring a curious phenomenon known as in-context learning, in which a large language model learns to accomplish a task after seeing only a few examples — despite the fact that it wasn’t trained for that task. For instance, someone could feed the model several example sentences and their sentiments (positive or negative), then prompt it with a new sentence, and the model can give the correct sentiment.
Jacob Andreas and other scientists from MIT, Google Research, and Stanford University are striving to unravel this mystery. They studied models that are very similar to large language models to see how they can learn without updating parameters.
Read the full story by the MIT News Office