Explore LLM word representations using similarity analysis
SMRTR summary
Representational Similarity Analysis (RSA) is a neuroscience-born technique now being used to compare how different AI language models, like BERT and GPT-2, internally represent words. Though their embeddings vectors are completely different, RSA reveals they share surprisingly similar internal semantic structures across word categories like fruits, furniture, and space.
SMRTR provides this summary for quick context. The original article belongs to Daily.dev.
Read the original article