Exclusive reveal: The 35-year-old mysterious godmother behind ChatGPT single-handedly made this god-level AI popular

When OpenAI suddenly decided to launch ChatGPT in late November 2022, employees were shocked.

Employees who were developing GPT-4 at the time were suddenly ordered to package the already existing GPT-3.5 and make it into a chatbot as soon as possible.

A few days later, ChatGPT went online with zero publicity. The employees did not have any expectations. However, within dozens of days, ChatGPT seemed to snowball and spread around the world.

The media has been digging into the behind-the-scenes story of the birth of this top-notch AI.

Recently, they discovered Mira Murati, the 35-year-old CTO of OpenAI and the godmother of ChatGPT.

#These days, the media spotlight has shifted from Sam Altman, the father of ChatGPT, to Mira Murati.

Especially the foreign media’s frequent reports on her in the past two months are particularly eye-catching.

But she is so mysterious that she doesn’t even have a Wikipedia homepage.

It is undeniable that she has now become one of the most influential figures in the technology world. And it was no accident that she got to this point.

OpenAI’s mysterious CTO

According to information circulating on the Internet, Mira Murati was born into a family of Albanian high school teachers and later obtained a bachelor’s degree in mechanical engineering from Dartmouth College. Bachelor of Science.

In 2013, Murati officially joined Tesla and was responsible for the development of Model X.

It was during this year that she became interested in AI.

At that time, Tesla was developing an early version of the artificial intelligence driving assistance software Autopilot and was also developing artificial intelligence robots for factories. This got Murati thinking about other real-world applications.

After that, Murati served as vice president of product and engineering at Leap Motion for two years.

Leap Motion is a software and hardware company that makes controllers that enable users to manipulate digital objects with hand movements while connected to a PC or Mac.

During Murati’s tenure at Leap Motion, the company launched new software designed for hand tracking in VR.

As the vice president of product and engineering, she hopes that the interaction between users and computers can be "as intuitive as playing with a ball." But she soon realized that VR-based technology had not yet been accepted by the public.

Murati thought carefully about her next steps, and she concluded that "tremendous advances in technology" must be used to solve the world's biggest challenges.

In 2018, Murati joined OpenAI as Vice President of Applied Artificial Intelligence and Partnerships.

At this time, OpenAI has stepped up its pace and formulated a plan that is more ambitious than its current strength.

In 2019, OpenAI transformed from a non-profit organization to a for-profit organization.

Subsequently, Murati rose through the ranks and became OpenAI’s senior vice president of products and partnerships.

Ten months ago, she became chief technology officer.

Since Murati was appointed as chief technology officer, OpenAI has released some of the most popular star AI products, such as DALL-E 2 and ChatGPT.

These products have also made the whole world look at OpenAI with admiration. Everyone realizes that OpenAI’s technology can create unimaginable economic value for the world.

On February 24, Sam Altman, CEO of OpenAI and the father of ChatGPT, released the AGI roadmap to ensure that AGI benefits all mankind and aims directly at a super-intelligent world.

There is no doubt that if artificial general intelligence (AGI)—human-level artificial intelligence—is truly realized, the world will Economies can be transformed overnight, followed by fundamental changes in social structures, political systems, and even international power dynamics.

Obviously, the subversive impact that ChatGPT may have on human society that people are currently discussing is not an exaggeration.

And Murati can let himself enter OpenAI and become a person with the power to influence the world. Every step he takes is not random.

Step by step, towards the top

During his tenure at Leap Motion, Murati was deeply aware that VR had not yet been accepted by the public, and in order to find technology to solve the world's biggest challenges , she joined OpenAI in 2018.

This time, she definitely made the right choice.

In OpenAI, she found many like-minded people who share her belief that AGI will become a technology that changes the world.

Here, Murati discovered an organization that was constantly changing. The world's best artificial intelligence labs are leveraging Internet-scale data volumes and custom-built supercomputers to train their models, and their brute force methods are producing results.

At first, OpenAI was small, but the team is growing. They realize that the computing power companies need is increasingly expensive. Commercialization is both an economic necessity and a way to expose their technology to the world.

In 2019, the nonprofit restructured itself as a for-profit startup with a nonprofit parent company, naming former Y Combinator president Sam Altman as CEO. He lived up to expectations and quickly raised $1 billion from Microsoft.

As OpenAI embarks on a new journey, Murati serves as Vice President of Applied Artificial Intelligence and Partnerships.

OpenAI faces the core problem of large language models (or algorithms that process and generate text on neural networks), according to Patrick Hebron, vice president of research and development at OpenAI competitor Stable Diffusion In other words, large language models are like sealed geniuses.

For example, Mozart was not Mozart until someone handed him an instrument.

Similarly, although OpenAI’s basic language model has huge potential, it still needs the help of external partners to be exposed at a faster speed.

At the same time, other AI labs are also exploring how to use large-scale language models to develop products. For example, Google Duplex is a voice robot version of Google Assistant that can help people make phone calls and make appointments.

In comparison, OpenAI focuses more on how to make the model better match human intentions. (Google’s subsequent failure was foreshadowed here.)

At this time, Murati was also promoted to director of research supervision, and she found herself at the core of this work.

Specifically, in order to achieve a true "Chat" GPT, OpenAI needs to allow the model to understand the value of human dialogue. A lot of language is subjective: What songs count as good? What kind of restaurant is considered high-end? Without human feedback, it would be difficult for a language model to understand these concepts, let alone understand a joke.

In order to convey these human values ​​​​to GPT, OpenAI has made a series of efforts. It hires contractors to evaluate GPT's output and write better output, as well as OpenAI's own employees to do the same.

It found partners willing to share their experimental data. It establishes a security operation that attempts to predict the ways in which users may compromise the system. Data from all these processes is then fed back into an algorithm layer that can sit on top of the language model.

The last step is to create a search bar interface on the system and name it ChatGPT. This step is the easiest.

In this way, the most popular and top-notch language model was born. In Murati's words, they got a chatbot that was "more likely to do what you want it to do."

##Greylock Interview: Dr. Fei-Fei Li, co-director of Stanford HAI, and Mira Murati, CTO of OpenAI

But ChatGPT is just the beginning.

OpenAI has a diverse application portfolio. From the beginning, OpenAI has designed a broad product portfolio.

"One of the reasons we wanted to make Dall-E was to understand the world more deeply and have these models understand the world the way we do," Murati said.

In her view, artificial intelligence models trained by multi-modal learning (that is, learning that includes text, images, audio, video and even robotics) are the key to achieving long-term breakthroughs .

The trick is to connect those future plans to the product in the here and now.

##Attend the event held by GIRL GEEK X

Unlike Google, which keeps its AI research under wraps, Murati is clearly keen on testing these products with the public.

"You can make technological progress in a vacuum without real-world contact. But then the question is, are you really moving in the right direction?"

In Murati’s view, if AGI is only constructed within the scope of a laboratory, the social impact caused by its emergence will be even more disturbing.

Adjusting large language models through reinforcement learning and human feedback is the most effective way to solve the future challenges of AGI, and it is also a way to engage the public in dialogue.

In addition, although Murati is confident in OpenAI’s capabilities, she does not seem to be like CEO Sam Altman, who often talks about science fiction stories such as AGI generating consciousness to create headlines.

In Murati’s view, consciousness is a philosophical idea, not a scientific idea.

Murati wants to reimagine the Turing Test to determine whether computers can think like humans and take on a wide range of cognitive tasks.

But most importantly, she hopes OpenAI’s systems can learn from real humans.

Regarding Murati’s personal information, we currently only know that she is a fan of the science fiction classic "2001: A Space Odyssey", the poet Rainer Maria Rilke and the rock band Radiohead

Bing’s three personalities also have her contribution

Yesterday, users discovered that Bing AI suddenly has three “personalities” to choose from: creativity, balance and accuracy.

Behind this, there is also the promotion of Murati and other OpenAI employees.

According to Fast Company, at OpenAI’s San Francisco headquarters, the six-person leadership team behind ChatGPT gathers together for regular meetings every week.

On a morning at the end of January, the first question in the regular meeting was how the team could make ChatGPT balance accuracy and creativity in answers.

Murati, as the CTO, was the first to speak at the regular meeting and asked how much the team knew about the correlation between the accuracy of ChatGPT and its imagination.

Liam Fedus, a former particle physicist and Google Brain scientist and current OpenAI researcher, said, “There is growing evidence that our underlying models are talented at creative writing. , but because OpenAI has shifted to a more useful model to some extent, it may have lost some capabilities."

Murati listened carefully to the answers, took notes, and planned New experiments to test the accuracy and creativity of the model will be launched in the coming weeks.

This line of investigation affects billions of dollars and millions of jobs.

If ChatGPT’s credibility cannot be proven, OpenAI’s status as the most promising startup in Silicon Valley in decades may be threatened.

Now we can see that after the ChatGPT-like model is integrated into Microsoft's Bing, it is also testing and exploring the product route step by step based on real market feedback.

After ChatGPT became popular

Previously, Murati rarely came to the stage, but with the recent explosion of AIGC, OpenAI has become a symbol leading the trend of the times, and Murati has also appear more in the public eye.

As early as four months ago, she appeared on Cui Wa’s “The Daily Show” to discuss the impact these powerful AI tools may have on humans.

In 2021, OpenAI hired ML researcher Jan Leike, who now leads an alignment team, marking their Pay more attention to the training of reinforcement learning with human feedback (RLHF).

On Twitter, Murati has been closely interacting with Leike, and they also discussed training the AI ​​model during The importance of incorporating feedback.

In discussions with CogX as early as eight months ago, Murati explained: "We want these models to follow our explicit instructions, but we also want them to follow our implicit instructions. instructions, so we don't want them to create any stereotypes, or create toxic reactions. We want them to be roughly consistent with what we want, which can be quite vague. To achieve this, we need to Provide manual feedback to the model to make the intention clearer."

Also in that discussion, she praised AI has brought about progress, but he also frankly admits that he does not know whether AI tools will replace humans.

"We are not sure whether AGI will completely replace us or help us do what we are doing. It is still unclear how human work will change in the future and what the economic system will look like."

"But overall, I think AI will have a very positive impact on mankind, because it can solve very difficult problems in the real world, such as dealing with climate change, or treating difficult problems. Miscellaneous diseases."

On Cui Wa's "The Daily Show," Murati took a more noncommittal attitude, calling OpenAI's applications "human helpers." It will become what "society shapes them".

OpenAI CEO and founder Sam Altman’s attitude is equally vague when talking about ethical issues related to ChatGPT.

And Murati also admitted that "the question of what exact impact AGI will have on humans" is too big and too open, and she cannot answer it yet.

In a recent interview with Time magazine, Murati also called on policymakers to start regulating AI.

"It is important for OpenAI and companies like it to bring AI into the public consciousness in a controlled and responsible way. But we are a small group, We need more investment in this system, and more investment beyond technology, such as regulators, governments and all other forces."

Now, the next big thing about OpenAI The language model GPT-4 is also preparing for launch. Obviously, it may attract the same attention as ChatGPT, or even more. However, Murati believes that "less hype would be better" for this.

But there is no doubt that Murati and her team will continue to stir up the world with every step they take.

The above is the detailed content of Exclusive reveal: The 35-year-old mysterious godmother behind ChatGPT single-handedly made this god-level AI popular. For more information, please follow other related articles on the PHP Chinese website!