Network Security Internet Technology Development Database Servers Mobile Phone Android Software Apple Software Computer Software News IT Information

In addition to Weibo, there is also WeChat

Please pay attention

WeChat public account

Shulou

Google Outsourcing Review complains: I don't understand many topics, how to judge whether Bard is right in a short time

2025-04-06 Update From: SLTechnology News&Howtos shulou NAV: SLTechnology News&Howtos > IT Information >

Share

Shulou(Shulou.com)11/24 Report--

On April 5, Google recently launched Bard, a chat robot, and called on its employees to conduct internal tests, while entrusting a number of external contractors to conduct evaluations. However, some contractors revealed that they simply did not have enough time to verify whether Bard's answer was correct, and in the end they had to guess.

After the explosion of OpenAI chat robot ChatGPT, Google quickly followed suit, launching a limited beta version of Bard chat robot in March. Similar to ChatGPT, users can ask questions or give task orders to Bard, and Bard will give a human-like response.

Currently, contractors at Appen, a multinational AI training data service, are helping to improve Google's chat robots. Although these people were not explicitly told that the tasks they were assigned were related to Bard, the internal discussion of the new tasks dates back to February 7, when Google first released Bard. Appen internal documents show that the contractor needs to review the quality of responses provided by the AI chatbot.

These contractors often help evaluate the relevance of Google's search algorithm and ads in search results, and mark harmful sites so that they do not appear in search results.

Four contractors interviewed said that since January, most of their work has shifted to censoring AI chatbot prompts. In the evaluation process, the contractors were disappointed with the performance of the chatbot and said they did not have enough time to accurately assess whether the chatbot responded correctly to the prompt, sometimes only by guessing. But they can still get paid.

Bard was criticized for giving the wrong answer in a demonstration. Google said that the chatbot will get better and better over time, and it should not be seen as a substitute for search.

Before the official release, Google asked its employees in February to spend two to four hours a day to help test the chat robot, including asking questions and marking answers that did not meet the company's accuracy standards and other metrics. Employees can rewrite the answer to any question for Bard to learn from. Google and Appen didn't respond to requests for comment.

If there is not enough time according to the contractor's guidance documents, they will receive prompts (such as questions, instructions or statements) sent by the user to the AI chatbot, as well as responses generated by the two machines. The contractor needs to help determine which response is better. They can also specify the reasons for the choice in the text box to help chatbots learn to find specific properties in acceptable responses. The answers given by chatbots should be coherent and accurate and introduce up-to-date information.

The contractor said they were set a fixed time to complete each task. The task time prompted by the review ranges from 60 seconds to a few minutes, which varies widely. These people admit that it is difficult to rate AI's responses if they are not familiar with the topics chatbots are talking about, such as blockchain technology.

Because there is a fixed reward for each task, some contractors say they will try their best to complete the task even if they realize that they cannot accurately assess the response of the chatbot.

An evaluator said: "in just 60 seconds, I don't have enough time to understand areas I don't understand, so I can only give the best guess so that I can continue to work and get paid."

Another contractor expressed a similar view, saying that they also wanted to get the right answer and provide the best possible chatbot experience, but they did not have enough time to study certain topics before conducting the evaluation. He added: "to be honest, many of us are about to collapse!"

The third contractor said: "it takes three hours of research to complete a task in as little as 60 seconds, which clearly highlights the problems we are facing now."

Demand for better working conditions at present, contractors working for Google through outsourcing companies are increasingly asking for better working conditions.

In February, a number of contractors visited Googleplex, Google's headquarters, and submitted a petition to Prabhakar Raghavan, head of search, hoping to increase pay. They work for Appen and earn between $14 and $14. 50 an hour. The businesses they support (search and advertising) are Google's main source of revenue.

The Alphabet workers' union has expressed support for the contractors and helped them take action, but the group cannot formally negotiate with Google on behalf of the contractors.

In Austin, Texas, YouTube contractors announced plans to unionize with AWU at the end of last year. The group estimates that Google employs more than 200,000 contractors, but they are not included in the company's official workforce.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.

Views: 0

*The comments in the above article only represent the author's personal views and do not represent the views and positions of this website. If you have more insights, please feel free to contribute and share.

Share To

IT Information

Wechat

© 2024 shulou.com SLNews company. All rights reserved.

12
Report