Dictionaries and Knowledge Sources for Language Models
One of the critical problems with very large language models is output validation. Many generative models produce a small percentage of unexpected results. Those results might still be unacceptable given the social norms, explainability, or bias. One way to deal with the problem is a top-down approach in which we infuse a filtering system with domain knowledge. Unfortunately, for large domains, this is a complex task. This paper discusses a way to build domain dictionaries. This approach can be used to generate #responsibleai knowledge sources for generative language models.
#nlp #ai #deeplearning #distributeddictionaries #languagemodels #generativeai #ethicalai #responsibleai
https://link.springer.com/article/10.3758/s13428-017-0875-9
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.