Bloomberg: Google insiders question Bard chatbot’s usefulness

IT House News on October 12, according to Bloomberg last night, Google and Discord jointly invited loyal users of their own AI chatbot Bard to set up a chat room. Among them, Bard’s product managers, designers and engineers discussed the effectiveness and practicality of AI in the chat room, but some people began to question it.

Two participants in the Bard community on Discord shared to Bloomberg details of discussions in the chat room between July and October. Bard's senior product manager Dominik Rabiej said in the discussion that he "doesn't trust" the responses generated by large language models, and recommended that people only use Bard for "creative" and "brainstorming" purposes. Rabiej also said that using Bard for programming "is also a good choice" - because people inevitably want to verify that the code works.

Additionally, Rabiej this month emphasized the importance of the new “Double-Check Replies” button in Bard. The feature is said to highlight "potentially incorrect" content in orange. He also reiterated that Bard doesn't really understand the text it's getting and simply responds with more text based on the user's prompts. "Please remember that Bard, like any large model, is generative - it is not looking up information or summarizing for you, but generating text.

"

This product manager once said bluntly in a chat in July, Don’t trust the output results of large language models unless (users themselves) can independently verify it. "We also want it to meet your expectations, but it's not there yet."

Another product manager Warkentin also wrote in the discussion, "Humanized improvements are crucial, only in this way can Bard become a product suitable for everyone. Otherwise, users will not have the ability to judge the function of the product, I think this would be a huge mistake.” “We don’t need products in ivory towers, but products that everyone can use!”

IT House previously reported that when Google expanded the deployment of Bard AI tools in July this year, it added the "share conversation link" function. The user can export the conversation between him and Bard as a web link and share it with others, so that others can continue the conversation with Bard.

An analytics consultant Gagan Ghotra discovered a vulnerability on the

Gagan Ghotra accordingly warned users not to share any "private content" when talking to Bard, otherwise the relevant information may be leaked.

The above is the detailed content of Bloomberg: Google insiders question Bard chatbot’s usefulness. For more information, please follow other related articles on the PHP Chinese website!