AI attention span so good it shouldn’t be legal

This episode of the Stack Overflow Podcast features interviews with Pathway and Mary Technology, both companies pushing the boundaries of AI in distinct fields. Pathway is developing Baby Dragon Hatchling, a post-transformer model designed to overcome the limitations of current LLMs regarding memory and continual learning. CEO Zuzanna Stamirowska explains that their model mimics the structure of the human brain, utilizing neurons and synapses to achieve efficient computation and long-term reasoning. This architecture allows for a vastly improved attention span, potentially exceeding current models by a significant margin and reducing the likelihood of hallucinations.

In the second segment, Rowan McNamee, co-founder of Mary Technology, discusses their AI-powered fact management system for attorneys. Mary Technology helps lawyers manage the overwhelming volume of legal evidence by extracting and organizing facts, identifying relevance, and providing a clear audit trail. The system utilizes LLMs and machine learning to streamline the review process, while emphasizing the importance of lawyer oversight and source verification to maintain legal integrity.

Both companies highlight the need for responsible AI development and deployment, particularly in sensitive areas like legal proceedings. Pathway focuses on building models that are more interpretable and efficient, while Mary Technology prioritizes transparency and provides tools for lawyers to validate AI-generated insights.

(Source：Plato Data Intelligence)