Google executives say the Bard AI meeting isn’t all for research

Google executives continue to deal with the fallout from last month’s flopped announcement of the company’s artificial intelligence engine called Bard, but their efforts to clean up the mess are causing more confusion among the workforce.

In a plenary meeting Thursday, the executives answered questions from Dory, the company’s internal forum, with most of the top-ranked issues related to priorities around the Bard, according to audio obtained by CNBC. It’s the first company-wide meeting since Google employees have criticized leadership, most notably CEO Sundar Pichai, for the way it handled the announcement of Bard, Google’s ChatGPT competitor.

Wall Street has punished Alphabet, Google’s parent, for its offering of Bard, sending the stock lower on concern that the company’s primary search engine is at risk of being replaced as consumers eventually turn to AI-powered responses that allow for more conversational and creative answers. Employees called Google’s initial public presentation “haste,” “a failure,” and “no Google.”

Jack Krawzyk, head of product at The Bard, made his debut Thursday, answering Dory’s next question, which was seen by CNBC.

“Bard and ChatGPT are big language models, not cognitive models. They’re great at creating text that looks like a human, and they’re not very good at ensuring their text is based on facts. Why do we think the first big application should be search, which at its heart is finding the real information ?

Krawczyk responded by immediately saying, “I just want to be very clear: Bard is not research.”

“It’s a collaborative AI service experience that we’ve been talking about,” Krawczyk said. “The magic we find in using the product is that it is this creative companion to help you be a spark of imagination, explore your curiosity, etc.”

But Krawczyk was quick to follow up by saying, “We can’t stop users from trying to use it like search.”

He said that Google is still providing its services to people who want to use it for search, noting that the company has built a new feature for internal use called “Search It”.

“We will try to improve the performance of linked queries there, as well as convey our trust to users,” Krawczyk said. He added that users would see a tab that said “View other drafts,” which should direct people away from search-like results.

“But while you want to get into more search-oriented travel, we already have a product for that — it’s called search,” he said.

An attempt to separate the Bard from the search seemed to indicate a pivot in the initial strategy, based on what employees told CNBC and on internal memes that have circulated in recent weeks. In the lead-up to Bard’s announcement, Google executives had said repeatedly that the technology it was developing internally would integrate with search.

Several Google employees, who asked not to be named because they are not authorized to speak on the matter, told CNBC that inconsistent answers from executives have added to the confusion.

Elizabeth Reed, vice president of engineering for research, echoed Krawczyk’s comments Thursday, highlighting the company’s extensive use of Language Large Models (LLMs).

“As Jack said, Bard is really separate from research. We have a very long history of bringing LLMs into research,” Reid said, citing models named Bert and Mum.

Reed said that while the company is experimenting with LLMs, it wants to “preserve the essence of what research is.”

In last month’s Google announcement, I mentioned search several times.

“We’re working to bring the latest advances in artificial intelligence into our products, starting with research,” the company said. blog post.

In the same week, at an event in Paris, Google Search Head Prabhakar Raghavan revealed some new examples of using Bard in search. After the announcement, company leaders urged all employees to help out by spending a few hours taking the Bard test and rewriting the wrong answers, citing “a huge responsibility to get it right.”

CNBC also previously reported that the company was testing different designs for Bard’s integrated homepage.

Another top-rated question Thursday from Pichai asked about different use cases for Bard, in which Googlers were asked to help with research and “rewrite queries with factual information.”

“It is important to acknowledge that it is experimental,” Pichai said in his response. “It’s very important to acknowledge the limitations of these products as well.”

With Bard, Pichai said, “you’re exposing users’ ability to talk to LLMs,” which will improve over time. “Obviously we’re engineering products on top of that,” he said.

“Products like this get better the more people use them,” Pichai said. “It’s a virtuous cycle.”

After Google launched Bard in February, Alphabet’s share price fell nearly 9%, suggesting that investors were hoping for more in light of growing competition from Microsoft, a big investor in ChatGPT creator OpenAI.

Employees are well aware of how an introduction is received.

“The first public demo was underwhelming, demoralized, and invited massive media coverage,” read an employee comment from my role read aloud. Then came the question, “What really happened?” He asked, “Please share your candid thoughts on what went wrong with Bard’s launch.”

Pichai pointed to the answer as Krawzyk, who danced around the topic without giving a direct answer.

Krawczyk said, “Questions like this can be fair and we want to reiterate the fact that Bard didn’t launch it. We’ve acknowledged to the world that this is something we’re testing — we’re testing. But there’s a lot of excitement in the industry right now.”

Krawczyk also pointed to an event held at Microsoft headquarters that week, where the company showed off how OpenAI technology could power Bing search results and other products.

“You see the ChatGPT stories coincide with an event we were having that was actually focused on research,” said Krawczyk. “There can be challenges around external perception but, as you heard today, we remain focused on cold testing.”

Krawczyk added that Google is excited about putting the technology in “users’ hands to capture their creativity.”

Pichai shouted to say, “It’s a tough time.”

“The purpose of the blog post was once we decided we were going to trusted third-party testers, and things could leak, it was important to put it in place,” Pichai said. “We haven’t launched the product yet. And obviously when we do launch it, we’ll make it clear that it’s a beta product.”

Pichai said the company hopes to provide more details after Google IO, its annual developer conference. Google has not yet announced the dates for the event.

Another comment from the top-rated staff from my role said, “Releasing an AI feels like an out-and-out reaction without a strategy.”

Pichai began his response by noting that Google spends more money on AI research and development than any other company.

He said with a laugh: “I do not agree with the introduction of this question.” “We’ve been working deeply on AI for a long time. You’re right in the sense that we have to stay focused on the users and make sure we’re building things with impact.” He said, “User input is going to be an important part of the process, so it’s important to get it right.”

Pichai called Jeff Dean, Google’s head of artificial intelligence, into a plenary meeting to answer a question about the company’s loss of top talent. Specifically, he asked the question why Google lost so many key people who were included in a paper about the prominent architecture used in artificial intelligence.

Dean said, “I think it’s important to realize that this is a very competitive field. People who have these kinds of skills are in great demand.”

Dean said Google has “two of the best AI research teams in the world” and “people working side by side to drive forward the state of the art in AI.”

“We have the ability to get things in the papers here but also work on products that touch millions of users every day,” Dean said.

Pichai added that, “For the last couple of weeks, we’ve been talking to some people who would like to join Google who are literally some of the best ML researchers and engineers on the planet.”

A Google spokesperson did not immediately respond to a request for comment.

He watches: Google could have a second-mover advantage with its chatbot technology