Zenodelic.ai uses a trained OpenAI GPT to crush the top LLMs on Theory of Mind benchmarks (including improving OpenAI's own scores by 25%).

MOORESVILLE, N.C. - Oct. 21, 2024 - PRLog -- In a groundbreaking achievement that is set to revolutionize the field of artificial intelligence, startup Zenodelic.ai has officially set a new world record in Theory of Mind benchmark scores, outperforming all established leaders in the industry. What makes this accomplishment really interesting, however, is that Zenodelic.ai used a trained version of an OpenAI GPT to beat Mixtral 8x7B, Claude Sonnet 3.5, and even OpenAI's own top GPTs. By instructing a custom GPT on how to better understand the human mind, Zenodelic.ai has raised the global standard for emotional intelligence in AI systems.

The set of benchmarks tested were the Open Theory of Mind benchmark set used by many world-leading institutions. They are designed to measure an AI's ability to understand and predict human thoughts and intentions, a crucial aspect of creating machines that can interact compassionately and with emotional intelligence, which is a requirement for Safe AI and also AGI (Artificial General Intelligence). The benchmark results improved OpenAI GPT-4o's overall score of 63% (broken down in the table) to a new 79% mark from Zenodelic.ai's trained GPT-4o, with MultiHop First Order matching human performance at 85%, and Location First Order coming in at 95% accuracy, just under human performance marks (and 50% higher than OpenAi's score).

"This is only version 0.1, and frankly, Theory of Mind isn't even our jam," said Sean Webb, founder of Zenodelic.ai. "These results are simply a side effect of our human mind processing algorithms. We'll get another 15% bump in the near future when we sit down and actually focus on improving the instruction set for these. Our main focus is creating emotional intelligence and artificial compassion for AI, for which we will introduce a new set of benchmarks to share soon. We had to create those because they didn't even exist." Webb has numerous highly-rated books on the human mind that are used as college coursework references and that are endorsed by numerous U.S.Navy SEALs as best-in-class references for managing the mind. He also has a background in supercomputing and technology. "Basically, our goal is to spread artificial compassion spread across the world."

This achievement comes at a time when the demand for emotionally aware AI is more critical than ever. With increasing applications in mental health, customer service, and personal assistants, Zenodelic.ai's record-breaking performance highlights the urgency for advancements in AI that prioritize emotional intelligence.

For media inquiries, interviews, or more information about Zenodelic.ai's groundbreaking work, visit Zenodelic.ai