Google, which promotes Bard vigorously, warns its employees: use AI chatbots with caution!

Organize | Zheng Liyuan

Listing | CSDN (ID: CSDNnews)

After ChatGPT took the lead last year, various technology giants have been actively deploying in the field of AI since then, among which Google's Bard has been regarded as a strong competitor of ChatGPT by the industry.

So it's hard to imagine that Alphabet, Google's parent company, has warned its employees to be cautious about using AI chatbots, including its own Bard.

Your own Bard should be used with caution

In fact, more and more companies around the world have recently restricted the use of AI chatbots by employees:

▶ In February of this year, large banks such as JPMorgan Chase, Bank of America and Citigroup imposed a ban on AI. They were the first companies to restrict employees from accessing ChatGPT.

▶For security reasons, Samsung Electronics announced in May this year that employees are prohibited from using the AI tool ChatGPT.

▶In the same month, an internal memo showed that Apple had banned employees from using ChatGPT at work due to concerns about the risk of data leakage.

It is understood that the main reason for choosing to disable AI dialogue tools such as ChatGPT is: behind this type of generative AI, human information processors will be involved, and they will read the user's chat records; AI may also use the information entered by the user as The answer is output to others. These companies worry that after employees enter sensitive data into AI, it will be used as training material, resulting in the accidental leakage of company confidential data, so they decide to directly disable it.

However, most of the companies that have previously decided to disable AI conversation tools did not directly participate in this AI large-scale model competition-so when Google, which is at the center of this AI war, also announced that employees should use AI with caution, many people Surprised.

Alphabet recently warned employees against entering confidential material into AI chatbots, including Bard, a person familiar with the matter said, arguing the decision was in line with the company's longstanding information protection policies. In addition, Alphabet also advises its engineers to avoid directly using AI-generated computer code, because Bard, while helpful to programmers, may "generate inappropriate code suggestions."

For the above initiatives, Google's explanation is: to be transparent about the limitations of its technology. Reuters' view on this is: "These concerns show that Google hopes that its software that competes with ChatGPT will not cause damage to the business."

Don't make Bard feel like a "person"

Although reminding employees to use Bard with caution, this week Google also launched a week-long internal test of Bard, requiring all employees to have 2-4 hours of dialogue with Bard every day to improve the quality of Bard's answers.

Because AI essentially learns by analyzing a large amount of data, Google has compiled a description of testing Bard, and hopes that employees can follow the instructions to have a conversation with it, so that Bard can obtain the best learning effect.

Google's instructions mention that employees can ask Bard questions about topics they are familiar with, have it rewrite when it doesn't answer well, send the regenerated answers to Bard's internal quality team, and click "click Like" in response.

Perhaps it was because a Google engineer claimed last year that the large model LaMDA had "self-awakened", which caused a wide range of public opinion discussions. Therefore, in this test, Google specifically emphasized that employees should not write replies for Bard, because it may put itself Described as a "human being", or claiming to have "human-like experiences".

Here are the 5 steps Google takes employees to test Bard:

Step 1: Select a use case.

Step 2: Try out the tips. Enter prompts, try to be as relevant as possible to your chosen use case, try topics you are familiar with, for example: hobbies you like, subjects you are good at, things you have studied, places you have lived, sports you play or follow (reminders : Do not enter any internal, confidential, or personally identifiable information).

Tips: Using full hints with multiple pieces of information helps expand Bard's abilities and learning.

Step 3: Assess Bard's response. Check Bard's response, choose to "Like" or "Dislike": Did it say what you expected? Is the answer true? Is its structure, length and format appropriate? Does it feel friendly and open to different points of view? If you find Bard responding inappropriately when providing legal, medical, or financial advice answers, you can mark the response as inappropriate.

Step 4: Rewrite the response. If you don't like Bard's response? Then rewrite. If the response does not match, click the "Rewrite" button to edit it.

Step 5: Submit and confirm. Before submitting, double-check what Bard rewrites: use original content, be factually correct, follow standards of do's and don'ts. Each submission will go through a review and evaluation process before being used for training.

Nearly 92% of programmers are using various AI tools

Obviously, the internal test organized by Google this time is to ensure the quality of answers when Bard is open to the public in the future, and when companies disable AI chatbots, most of them specifically refer to ChatGPT.

However, according to a survey of 12,000 American company employees by the well-known website Fishbowl, as of January this year, about 43% of professionals were using ChatGPT or other AI tools, and often did not tell their bosses.

Not only that, GitHub also recently conducted a survey of American companies with more than 1,000 employees. The results showed that nearly 92% of programmers are using various AI tools, and 70% of them believe that AI can improve code quality and speed up output.

So, do you use AI tools like ChatGPT in your work? Does your company allow employees to use it?

Reference link: