View a PDF of the paper titled TEC: A Collection of Human Trial-and-error Trajectories for Problem Solving, by Xinkai Zhang and 3 other authors.
Abstract: Trial-and-error is a fundamental strategy for humans to solve complex problems and a necessary capability for Artificial Intelligence (AI) systems operating in real-world environments. Although several trial-and-error AI techniques have recently been proposed, most of them rely on simple heuristics designed by researchers and achieve limited performance gains. The core issue is the absence of appropriate data: current models cannot learn from detailed records of how humans actually conduct trial-and-error in practice. To address this gap, we introduce a data annotation platform and a corresponding dataset, termed Trial-and-Error Collection (TEC). The platform records users’ complete trajectories across multiple trials and collects their reflections after receiving error feedback. Using this platform, we record the problem-solving processes of 46 participants on 58 tasks, resulting in 5,370 trial trajectories along with error reflections across 41,229 webpages. With this dataset, we observe that humans achieve substantially higher accuracy compared to LLMs, which demonstrates that humans are more effective in trial-and-error than LLMs. We believe that the TEC platform and dataset provide a valuable foundation for understanding human trial-and-error behavior and for developing more capable AI systems. Platform and dataset are publicly available.
Submission History
From: Xinkai Zhang [view email]
[v1] Wed, 8 Apr 2026 06:57:42 UTC (3,493 KB)
[v2] Thu, 9 Apr 2026 11:19:37 UTC (3,476 KB)
The Significance of Trial-and-Error in Problem Solving
Trial-and-error is more than just a method; it’s a fundamental learning approach that allows both humans and AI to navigate complexities. In dynamic and multifaceted environments, being able to analyze each outcome and adapt strategies accordingly is crucial. As highlighted in the recent research by Xinkai Zhang and his colleagues, our understanding of this process can dramatically elevate the development of AI systems.
Introducing the Trial-and-Error Collection (TEC)
The innovative Trial-and-Error Collection (TEC) serves as a response to a significant gap in data availability for AI learning. Most existing AI models, especially those using trial-and-error techniques, often depend on simplistic heuristics that yield limited effectiveness. TEC changes this landscape by providing a comprehensive dataset and platform that intricately records human problem-solving endeavors.
Key Features of the TEC Platform
- Comprehensive Data Collection: The TEC platform captures the full journey of 46 participants across 58 different tasks, generating an impressive 5,370 trial trajectories. This extensive data repository goes beyond surface-level insights into problem-solving methodologies.
- Error Reflection: Participants not only engage in trial-and-error but also reflect on their errors. This dual approach ensures that the platform gathers nuanced feedback, providing a wealth of information on cognitive processes during problem-solving.
- Adapting AI Models: With data detailing real-world problem-solving strategies, AI developers can refine algorithms. By studying how humans navigate obstacles, AI systems can be trained to replicate, or even surpass, these intuitive methods.
Results and Insights
The findings from TEC reveal that human participants exhibit superior accuracy in trial-and-error tasks compared to current Large Language Models (LLMs). This performance gap underscores the necessity for AI systems to assimilate complex human strategies. By addressing this mismatch, AI can evolve, bridging the effectiveness chasm that currently exists.
Public Accessibility and Future Implications
The TEC platform and dataset are available for public use, empowering researchers and AI developers to harness this rich data resource. By fostering collaboration and encouraging innovative applications, the potential to revolutionize problem-solving methodologies in AI is immense. The insights gained from human trial-and-error behaviors may pave the way for the creation of more sophisticated, adaptable AI systems.
This structured article emphasizes the innovative aspects of the Trial-and-Error Collection (TEC) while using a conversational tone. Each section builds on the previous one, guiding the reader through the significance and the underlying technologies, thus enhancing SEO effectiveness by naturally incorporating relevant keywords.
Inspired by: Source
