Associate Professor at the University of Copenhagen. I work on Vision-Language and Tokenization-free NLP. EMNLP 2026 PC.
- Tweets3.4K
- Following467
- Followers3.3K
- Likes1.1K
![Overview of the paper.
Title: The Emergent Symbolic Structure of Artificial Neural Networks
Authors: Tom McCoy, Paul Soulos, Tal Linzen, Paul Smolensky
Left: Neural networks encode information in vectors (there is then an image of a vector), yet they excel at tasks long thought to require symbolic structure (there is then an image of a symbolic representation, specifically a syntax tree). How do LLMs do it?
Right: We find that LLM representations can be closely approximated with symbolic structures. This approximation lets us edit the structure of an LLM’s output by editing the structure of its internal representations, as shown. There is then an image of two edits to LLMs. In the first one, the original input is 3 + 6 * 8, with an answer of 51. But if we swap the positions of the 3 and the 6, the output becomes 30. In the second one, the original input is a Python command repeating the list [Z, U] three times, producing [Z, U, Z, U, Z, U]. But if we edit the input in a way that adds](https://pbs.twimg.com/media/HRJ5Jr5W0AAQ-sN.jpg?name=orig)
ALT Overview of the paper. Title: The Emergent Symbolic Structure of Artificial Neural Networks Authors: Tom McCoy, Paul Soulos, Tal Linzen, Paul Smolensky Left: Neural networks encode information in vectors (there is then an image of a vector), yet they excel at tasks long thought to require symbolic structure (there is then an image of a symbolic representation, specifically a syntax tree). How do LLMs do it? Right: We find that LLM representations can be closely approximated with symbolic structures. This approximation lets us edit the structure of an LLM’s output by editing the structure of its internal representations, as shown. There is then an image of two edits to LLMs. In the first one, the original input is 3 + 6 * 8, with an answer of 51. But if we swap the positions of the 3 and the 6, the output becomes 30. In the second one, the original input is a Python command repeating the list [Z, U] three times, producing [Z, U, Z, U, Z, U]. But if we edit the input in a way that adds
ALT UT to lead new $30M NSF robotics center.\n\nNew five-year award will establish the Center for Human and Robot Co-Adaptation.
ALT The center will bring together 39 researchers from six universities to study how people and robots adapt to one another over time in everyday environments.
ALT \u201CThe next leap in robotics is not just getting robots to perform more tasks,\u201D said Joydeep Biswas, an associate professor of computer science who will serve as the center\u2019s director. \n\n\u201CIt is enabling robots to understand the people around them: their needs, preferences, values and constraints, while recognizing that people also adapt their behavior when they integrate robots into their environment.\u201D
ALT The research explores how robots and people learn from each other \u2014 robots adapting to individual needs and abilities, and people discovering new ways to live more independently.
