Mohave County
Inspecting and Editing Knowledge Representations in Language Models
Hernandez, Evan, Li, Belinda Z., Andreas, Jacob
Neural language models (LMs) represent facts about the world described by text. Sometimes these facts derive from training data (in most LMs, a representation of the word "banana" encodes the fact that bananas are fruits). Sometimes facts derive from input text itself (a representation of the sentence "I poured out the bottle" encodes the fact that the bottle became empty). We describe REMEDI, a method for learning to map statements in natural language to fact encodings in an LM's internal representation system. REMEDI encodings can be used as knowledge editors: when added to LM hidden representations, they modify downstream generation to be consistent with new facts. REMEDI encodings may also be used as probes: when compared to LM representations, they reveal which properties LMs already attribute to mentioned entities, in some cases making it possible to predict when LMs will generate outputs that conflict with background knowledge or input text. REMEDI thus links work on probing, prompting, and LM editing, and offers steps toward general tools for fine-grained inspection and control of knowledge in LMs.
Exercise Reveals Advantages Artificial Intelligence Gives in All-Domain Ops
U.S. Northern Command hosted a global information dominance exercise and the results point to the tremendous advantages the Defense Department would receive by applying machine learning and artificial intelligence to all-domain information, Air Force Gen. Glen D. VanHerck said yesterday. All 11 combatant commands participated in the exercise, which was based on a global scenario involving two peer competitors. "What we were looking to do is show the incredible value of information and how information can be used today," he said. This is especially true if commanders can take information from all domains -- undersea, space, cyberspace, air, land, sea and share it through machine-learning artificial intelligence, he said. This would serve to make data and information available in a timely manner to produce space for decision makers.