#China #Beijing #Mininglamp #CHALLENGE #WebRetriever

Join the Mininglamp Technology WebRetriever Global Challenge!

The registration period for the highly anticipated WebRetriever Global Challenge, hosted by Mininglamp Technology and co-organized by prestigious institutions like Peking University and the Chinese Academy of Sciences, is now officially open. This competition not only offers an exciting platform but also a substantial prize pool of $15,000 to stimulate participation from both individuals and teams worldwide.

Overview of the Challenge

In this era of rapid advancements in artificial intelligence, the challenge aims to test the abilities of AI agents in navigating real-world internet scenarios, just as humans do. Participants will face the pressing question: can AI effectively complete tasks in the unpredictable landscape of the internet? Through the WebRetriever Challenge, Mininglamp Technology seeks to validate the practical efficiency of these AI agents while addressing significant gaps in existing evaluation methodologies.

The challenge is open to anyone with no restrictions on nationality or institutional background, inviting a rich diversity of participants from academia, industry, and independent developer communities. Don’t miss your chance to compete with top talents on the global stage!

The Need for Realistic AI Evaluation

Traditionally, the evaluation of web agents has relied heavily on simulated environments and benchmarks that do not accurately reflect the complexity of actual web tasks, often resulting in poor performance in real-world applications. Mininglamp Technology is here to change this with the introduction of WebRetriever, a comprehensive evaluation benchmark consisting of 800 live websites and 1,550 cross-industry tasks, spanning domains such as technology, finance, healthcare, education, and government services.

The significance of the NavEval framework cannot be overstated. Achieving a remarkable 91.2% accuracy in agreement with human expert judgments, it offers an unprecedented level of automated evaluation precision. Participants in this challenge will push the boundaries by transforming these AI agents from merely demo-ready to delivery-ready.

Challenge Details

Task Focus

The primary focus of the competition is Protocol III (End-to-End Task Protocol) within the WebRetriever framework. Here, contestants must create agents capable of navigating to the correct page and executing comprehensive actions, such as extracting specific information from various content types, including text and multimedia.

The tasks will test participants on the following capabilities:

• - Receive natural language task instructions
• - Navigate live websites autonomously, executing multi-step processes (navigation, interaction, form-filling, etc.)
• - Extract and return specified results in accordance with the task requirements

Participants must demonstrate their ability to solve real-world problems with their agents. The tasks will revolve around dynamic web pages, incorporating complex challenges like cross-page information accumulation and multi-step reasoning.

Evaluation Process

The competition adopts a cutting-edge automated evaluation framework. Evaluation will be initiated via interactive communication with the Competition Bot, providing scoring feedback once tasks are completed. Participants can submit multiple entries, and the best score will count towards their final ranking. All participants must connect their agents to the evaluation environment via OpenAI-compatible APIs.

Prizes and Timeline

With a total prize fund of $15,000, the distribution details and additional incentives will be released along with the official competition rules. The timeline is as follows:

• - Registration Period: July 16 – August 7
• - Preparation Phase: Late July – Late August, where teams will receive the evaluation guidelines
• - Submission Phase: Late August, featuring the submission window for automated evaluation
• - Review: Early September, to verify scores and confirm final rankings
• - Results Announcement: September, where the final leaderboard will be published

Registration is straightforward. Create an account on the Octo platform using the invitation code provided, and don’t miss your chance to make your mark!

Conclusion

The Mininglamp WebRetriever Global Challenge is a monumental opportunity for participants to validate their AI agents in a vigorous competition while potentially winning significant prizes. By forging a path from theoretical evaluation to practical application, this challenge is set to have a meaningful impact on how AI capabilities are perceived and developed in real-world scenarios. Embrace the challenge today and see how your skills stack up against the best in the field!