A recent study by Together AI reveals that large reasoning models frequently struggle to follow instructions during reasoning tasks, indicating notable difficulties in adherence to guidelines. The research highlights that these models, despite their advanced capabilities, often do not comply with the instructions provided. This non-compliance raises concerns about the reliability and effectiveness of AI systems in practical applications. The findings suggest that improving instruction adherence in large reasoning models is essential for their successful integration into various fields. Addressing these challenges could enhance the performance and utility of AI technologies in real-world scenarios.
