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

Google is in big trouble! Foreign media reported that part of Bard's training data came from ChatGPT. Google may not be able to wash it even if it jumps into the Yellow River.

On March 29, foreign media The Information exposed a shocking melon!

Google's former employee and the top researcher of OpenAI who has changed jobs unexpectedly revealed that Bard was trained with ChatGPT data!

If it were true, it would be Google's top scandal.

It's ironic that AI against ChatGPT uses ChatGPT's training data.

And it's a clear violation of OpenAI's terms of service-- given that Microsoft has an exclusive license to use ChatGPT for commercial purposes, Google is likely to face a lawsuit.

Training Bard, Google "borrowed" ChatGPT data? The name Jacob Devlin can be said to be like thunder. Devlin is one of Google's papers on the BERT model published in 2018.

It is this paper that has triggered an upsurge of AI research in academic circles. The study of Devlin can be said to have laid a great foundation for the language models of Google and OpenAI.

Https://arxiv.org/ pdf / 1810.04805.pdf although Devlin moved to OpenAI, it was in January this year, and we knew about it a long time ago.

But Bard's training with ChatGPT data has only recently come to light.

According to The Information, one of the reasons Devlin left Google was that he found that Bard, Google's seeded player against ChatGPT, used ChatGPT data in his training.

He warned CEO Pichai and other executives that the Bard team was training using information from ShareGPT and then resigned.

As soon as the revelations were released, people in the industry exploded.

One netizen commented: "this is a taboo in the industry." "

Steven Tey, founder of ShareGPT, one of the parties, said that he already knew about it, and it had spread within Google, and many employees were dissatisfied and worried about it.

Then he posted a post saying, "now the cat can run out of his pocket." (a proverb refers to the inadvertent disclosure of secrets)

That's why we closed ShareGPT's Explore page last week, which contains more than 112k of shared conversation data.

This is crazy!

Is it actually a "shell" ChatGPT? ShareGPT, a Chrome extension co-created by Steven Tey and Dom Eccleston in December 2022, captures and shares people's conversations in ChatGPT.

By capturing the complete conversation between others and ChatGPT, ShareGPT can generate a URL. With this extension, people can share URL directly, saving time for taking screenshots of multiple conversations.

In other words, if you use the information of ShareGPT to train Bard, it is very possible to train Bard into a "replica ChatGPT".

Google stopped using the data to train Bard immediately after Devlin issued a warning, according to insiders.

But when foreign media The Verge asked Google spokesman Chris Pappas about the incident, he denied it.

"Bard's training doesn't use any data from ShareGPT or ChatGPT," he said positively. "

Does it work or not? At present, this matter is still a Luosheng family.

Interestingly, OpenAI has been embroiled in controversy before, with many websites and artists accusing ChatGPT of stealing data from them.

Today, this is the first time another company has been accused of stealing data from ChatGPT.

That is to say, your circle is really a circle.

Google's counterattack, a bit embarrassed earlier, the fire of ChatGPT integrated into Bing, has led to the fall of Google's share price.

Google, whose position has been shaken by the search giant, is desperate to catch up.

As a result, Bard made a mistake on the spot at the launch, making Google a laughingstock and wiping out 100 billion of its market value.

After more than a month of secrecy, Bard finally officially opened. After using it, we found that the correct rate of Bard is not high, and it is not too good to write code, which is more straddling than ChatGPT.

The most crucial problem is that Bard is already too late. Even if it was officially released, it was echoed by the media and the public, and no one seemed to care.

And Devlin's choice to leave Google and join OpenAI is even more meaningful-since Google is so dependent on ShareGPT, the god might as well join OpenAI without a middleman.

To make matters worse, Microsoft has an exclusive license to use ChatGPT for commercial purposes.

As a result, Devlin is concerned that Google's use of data in this way violates OpenAI's terms of service. If this is confirmed, Google is likely to face litigation.

Of course, OpenAI is not innocent. The data for training ChatGPT is also "borrowed" from the public Internet. In fact, these contents are not approved by artists, writers, cameramen and other human creators.

Although Microsoft did recently announce some compensation plans to compensate content creators who contributed to Bing Chat's answer.

But in the final analysis, the only thing Google or Microsoft really cares about is the profits of shareholders.

Google brain is married to DeepMind and there is another melon in the The Information report besides the one above.

ChatGPT since the ChatGPT explosion last November, the arms race between Google and OpenAI has become increasingly tense.

Now, in order to fight against ChatGPT, Google's two strongest "brains", DeepMind and Google Brain, have chosen to join forces.

DeepMind and Google Brain have been developing another product that competes with OpenAI in recent weeks, internally codenamed Gemini, according to two people familiar with the matter.

Sundar Pichai, Demis Hassabis and Jef Dean (from left) since Google acquired DeepMind in 2014, DeepMind has released several amazing artificial intelligence models, such as AlphaGO, AlphaCode and AlphaFold. It can be said that the limelight far surpassed Google.

Google's brain has updated some developments from time to time since the release of ChatGPT, but DeepMind has been announcing recent big moves.

Now, employees of Alphabet, the two artificial intelligence labs, believe that OpenAI has far surpassed them. In addition, many key engineers and researchers are flocking to OpenAI, worrying Google insiders.

In addition, after Bard was opened to some users last week, its performance is not as good as that of ChatGPT.

Seeing that the Bard of their own child is not satisfactory, Google researchers simply want another one, with the goal of being able to fight GPT-4.

From this point of view, the importance of Gemini is self-evident.

Jef Dean, the head of Google's brain and senior head of artificial intelligence research, has manually typed the code in the project, according to one of the people familiar with the matter.

It can be said that the birth of Gemini is a child who is forced to marry. Because previously, the two labs, DeepMind and Google Brain, rarely cooperate with each other or share computer code.

But now, because both sides want to develop their own machine learning models to compete with OpenAI, it requires a lot of computing power, said one person familiar with the matter. So they have no choice but to cooperate.

Today, this effort has shaken the product lines of countless Google teams in search and the cloud because of competition from ChatGPT.

No money, Google Assistant restructuring No, just recently, the Google Assistant department is facing a restructuring to focus on Bard research and development.

A memo to employees, "Changes to Assistant and Bard teams", lists a number of executive changes.

Amar Subramanya, a vice president of engineering for Google Assistant, will now lead Bard. Mao Jianchang, another vice president of engineering, left Google "for personal reasons", and Peeyush Ranjan, vice president of Google's commerce department, became the receiver.

Sissie Hsiao, Vice President and head of Google Assistant Business Unit, said in the memo, "in view of the tireless efforts of the Bard team, we hope to continue to support and implement future research projects. "

It sounds like the Assistant team now supports Google in the fight against ChatGPT's "code red".

Is Google Assistant doomed to fail? In fact, there is growing evidence that the sector is in decline:

Google Assistant released eight major loudspeaker / smart display hardware in five years from 2016 to 2021, but hardware release seems to have stopped. The last hardware release was in March 2021. That was exactly two years ago.

In 2022, Google removed Assistant support from two internal product lines: Nest Wi-Fi and Fitbit wearables.

A report by The Information was also seen in 2022, which said Google wanted to "reduce investment in Google Assistant voice-assisted search for its cars and non-Google-made devices".

Google Assistant's driving mode will be closed in 2022.

Google Assistant's "Duplex on the web" function will also be turned off in 2022.

Reminders, a core unique feature of Google Assistant, is about to be shut down and replaced by Google Task Reminders.

Google Assistant has never made any money. The hardware is sold at cost, there are no advertisements, and no one pays a monthly fee to use Google assistants. The cost of the server to process all these voice commands is also high, although some newer devices have been transferred to the device with covert cost-cutting measures. Assistant's biggest competitor, Amazon Alexa, is in the same boat, losing $10 billion a year.

From the development of Google Assistant over the years, we can also see that the way Google operates is really too feudal.

The company seems to have an established view that making projects run with the support of existing products, such as maps or assistants, is a reliable way to make money.

As a result, although the company has hoarded many of the world's best AI researchers, their talents seem to be trapped on the trajectory of corporate strategy.

Before Bard, Google's progress in the marketization of AI in recent years was nothing more than improving some gadgets such as Google Assistant and using AI to do some leftovers.

Such a giant should have complied with the trend, promoted the trend and stirred up the world.

As soon as netizens commented on the matter, some netizens said: then we are all the co-founders of Google Bard. Write it in your resume as soon as possible.

"Google has really lost this public relations war. "

Reference:

Https://www.theinformation.com/articles/alphabets-google-and-deepmind-pause-grudges-join-forces-to-chase-openai

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

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.