OpenAI is caught in the biggest public relations crisis in history, and the head of Altman, who is in charge, donated half of his net worth to help the company tide over the difficulties

36 Krypton

2024-05-29 18:05Posted on the official account of Beijing 36Kr

Text: Chen Sida, Li Ran

Editor|Li Ran

OpenAI may have had its biggest PR crisis since its founding!

And Sam Altman's response is so fast and effective that it is definitely worth learning from all company owners and publicists on a frame-by-frame basis.

Let's take a look at OpenAI's latest action first - the official tweet suddenly announced on May 28:

Source: X

OpenAI董事会成立安全与安保委员会（Safety and Security Committee），将就OpenAI项目和运营的关键安全决策，向全体董事会提出建议。

At the same time, it also released high-profile news that it has begun training its next cutting-edge model, which is expected to "go to another level on the road to AGI".

But if you just see that OpenAI is just carrying out daily AI safety work and showing off its muscles by the way, many people who eat melons may have questions - why the establishment of a safety committee and the start of training of new models, two things that were originally completely unrelated, should be put together?

What is the subtext of the announcement?

Source: X

The backbone of the security team defected, and the former board members issued an attack

Although the "OpenAI Zunyi Conference" finally established a leadership structure with Altman as the core, the company's senior management still has not resolved the differences on AI security issues.

The dark line of "AI security" has been laid since 2015 when Musk pulled people to establish OpenAI.

Recently, Ilya and Leike, the directors of the "super alignment team" in charge of AI security, both left, and put OpenAI's management on the fire.

Source: X

And Leike turned around and joined the "OpenAI Avengers" - Anthropic, which was co-founded by two former employees who later left with different values from the OpenAI team.

Before leaving, he did not forget to shout to the employees of OpenAI:

I would like to say to all OpenAI employees:

Society Sensitivity AGI.

Act with majesty commensurate with the cause you are building.

I believe you can "push" the cultural change needed.

I'm counting on you.

The whole world is counting on you.

Source: X

This remark is almost pointing at the nose of OpenAI's management and saying that you don't pay attention to AI security.

According to TechCrunch, Leike will be able to report directly to Jared Kaplan, Anthropic's chief scientific officer, and Anthropic's current research on scalable supervised researchers (techniques for controlling large-scale AI behavior in a predictable and desirable way) will report to Leike.

图源：TechCrunch

Leike and Ilya's exodus from netizens has also caused netizens to question whether OpenAI developed AGI to steal everyone's jobs, or for the well-being of all mankind?

Source: X

5月26号,OpenAI宫斗剧的始作俑者、前 OpenAI 董事会成员 Hellen Toner 和 Tasha McCauley 都站出来撰文指认Sam Altman以及他治下的OpenAI,完全不把AI安全当会事。

In the view of these two former members of OpenAI's board of directors, it is completely impossible to effectively supervise the AI company internally, and external supervision is imperative.

Source: X

On May 28, in the latest episode of the podcast The Ted AI Show, Hellen Toner revealed that the dignified OpenAI board members had to know about the release of ChatGPT through Twitter!

Toner emphasized that due to the non-profit nature of OpenAI's board of directors, it was established to ensure that OpenAI prioritizes the public good, not profit. But given Altman's repeated lies to the board, the board's work was also difficult. Host mending: OpenAI forgot its roots.

In the context of this eight-sided crusade and backlash, let's take a look at Sam Altman and OpenAI's textbook-level PR combo.

Can a combination of punches turn OpenAI into an opportunity?

The first move is to set up a new AI security agency to fill the organizational vacuum caused by the exodus of "defectors".

On the 28th, OpenAI officially announced these two things, let's translate them for you.

OpenAI董事会成立安全与安保委员会（Safety and Security Committee），将就 OpenAI 项目和运营的关键安全决策，向全体董事会提出建议。

安全与安保委员会将由董事Bret Taylor（董事会主席）、Adam D’Angelo、Nicole Seligman和 Sam Altman领导。

- We have a complete top-level architecture to ensure that OpenAI will fully consider the security issues brought by AI in the pursuit of AGI.

At the same time, a high-profile message was released:

We have already begun training its next cutting-edge model, which is expected to "go one step further down the road to AGI."

- We will further increase the transparency of the cutting-edge (hazard) model to the public, and we hope that everyone will trust us. OpenAI will not come out with a nuclear weapon behind the world's back.

The rest of the announcement:

The Safety & Security Committee's top priority is to evaluate and further develop OpenAI's processes and safeguards over the next 90 days. At the end of the 90 days, the Safety and Security Committee will share their recommendations with the full Board. After a thorough review by the Board of Directors, OpenAI will publicly share an update on the adopted recommendations in a manner consistent with safety and security.

——The committee is not a decoration, the first step is to comprehensively assess whether there are any security issues in OpenAI's development process, within 90 days!

OpenAI technology and policy experts Aleksander Madry (Head of Preparation), Lilian Weng (Head of Security Systems), John Schulman (Head of Alignment Science), Matt Knight (Head of Security), and Jakub Pachocki (Chief Scientist) will also be present on the committee.

- The committee is not all retreat leaders, and the core members are all from technical backgrounds, as described below. Although Ilya and Leike are gone, we at OpenAI are still full of talent.

In addition, OpenAI will support security efforts by retaining and consulting with other safety, security, and technology experts, including Rob Joyce, a former cybersecurity officer who provided security advice to OpenAI, and John Carlin.

-- We will also make full use of external forces to help AI security governance. AI security is not limited to OpenAI's internal oversight.

But netizens don't seem to buy it so much, and now they don't pay attention to whether GPT is 5 or 6 at all, but instead dislike the new security team: What role can you Altman play in covering the sky with one hand?

Source: X

"If you do this, you will listen to the government, and everyone will not believe you, nor will they trust the security team you have formed!"

Source: X

If Altman is a gold medal publicist, he must have expected that netizens may not buy the official initiatives, the two news have just been officially announced, and the founder IP will continue to stand up!

The second trick is to donate half of your net worth, just to gain a good reputation.

Also on the 28th, Altman and her husband, as billionaires, announced that they would join the "Giving Pledge" and donate more than half of their personal wealth to improve society.

Source: official website

In the joint open letter, dated the 18th, the two said:

We would not have made such a pledge of giving without the hard work, talent, generosity, and dedication of many people to improve the world. Because it is these people who have built the scaffolding of society that has brought us to where we are today. Our gratitude is beyond words, we can only keep this kindness in our hearts, do our best to repay the society, and build this scaffolding higher.

We intend to focus our donation on supporting technologies that help create affluence for people so they can build their scaffolding higher.

The initiators of the Giving Pledge, Warren Buffett and Gates, are well-known. By signing a giving pledge, a wealthy person is equivalent to giving at least half of their wealth to charity or charitable causes, either during their lifetime or in their will.

According to Bloomberg, Altman is worth at least $2 billion.

The third trick is to stifle all negative news in the cradle.

In mid-May, some media broke the news that in the labor contract signed by OpenAI and employees, it was clearly required that the departing employees should not make remarks that were unfavorable to the company's image after leaving, otherwise they would lose the opportunity to cash out their OpenAI "options".

Source: Vox

When this "overlord clause" was first exposed, there was an uproar on the Internet. But as soon as public opinion was about to ferment, Altman publicly stated that the company's senior management was completely unaware of such a "PR clause", and now the clause has been deleted.

Source: X

It is impossible to verify whether the senior management is aware of this "PR clause" in the contract. But if the "PR clause" is the source of the PR crisis, the easiest thing to do is to delete it outright.

This set of combination punches has prevention in advance, response during the event, and make up for it afterwards. Sam Altman and OpenAI's textbook crisis PR tactics make people applaud.

But at this point, in addition to watching the excitement of OpenAI, everyone will also be concerned, what kind of configuration can make up for the gap in the security team caused by the departure of Ilya and Leike?

The technical background is strong, look at the strength of the security committee bosses

OpenAI's technology and policy experts on the Safety and Security Committee are all heads across the company.

Aleksander Madry (Head of Preparation)

Public photos

OpenAI's technology and policy experts featured in the announcement, Aleksander Madry, are listed first.

According to his profile, Madry is a professor in the Department of EECS (Department of Electrical Engineering and Computer Science, the largest department at MIT) and a member of CSAIL (MIT's Laboratory of Computational Science and Artificial Intelligence).

On X's homepage, he describes his work as "making AI more reliable and safe to have a positive impact on society." In line with this, Madry is the Director of the Deployable Machine Learning Center at MIT, which aims to build stable, reliable, and secure AI systems for real-world deployments, and co-head of the AI Policy Forum.

Source: LinkedIn

He received his Ph.D. from MIT in 2011. However, prior to joining MIT, he also had research and work experience at Microsoft Research New England and Ecole Polytechnique Fédérale de Lausanne.

Source: LinkedIn

Lilian Weng (Chinese name Weng Li, head of security system)

Public photos

When she joined OpenAI in 2018, she was not directly responsible for safety-related work, but as the technical lead of the robotics team, focusing on training algorithms for robot tasks. But OpenAI has since abandoned research on robots.

Source: LinkedIn

It was also in 2021 that Weng Lide's job responsibilities began to clearly involve safety-related content. During this year, she established and led the Applied AI research team, and began to focus on safety and alignment issues.

Source: LinkedIn

In 2023, she officially began leading a team of engineers and researchers working on AI security.

Source: LinkedIn

Weng Li studied information management at Peking University as an undergraduate, then went to Indiana University Bloomington to study computer science, and then worked for companies such as affirm (fintech), Dropbox (cloud storage service), and also interned at Facebook.

Source: LinkedIn

John Schulman (Head of Alignment Science)

Public photos

Schulman is the co-founder of OpenAI. As one of the leaders of the post-training team, he is responsible for fine-tuning ChatGPT and models deployed in the OpenAI API.

In addition to this, language model alignment is his research interest. He also focuses on the behavior of models and is committed to improving the transparency of models in front of the public.

Previously, Schulman received his Ph.D. in computer science from the University of California, Berkeley, where he worked in robotics and reinforcement learning. He studied neuroscience briefly at the University of Berkeley, and I studied physics at Caltech.

Source: Personal homepage

Matt Knight（安全负责人）

Public photos

Knight is responsible for leading the engineering and research related to security, IT, and privacy at OpenAI.

Source: LinkedIn

Knight has a bachelor's degree in engineering science from Dartmouth College.

Source: LinkedIn

Previously, Knight specialized in the development of secure embedded systems, with expertise in both software and hardware. Due to Knight's special interest in RF networks and the physical layer, he reverse-engineered the LoRa PHY based on blind signal analysis.

Jakub Pachocki(首席科学家)

Public photos

The sudden mention of Pachocki may be a little unfamiliar, but he is Ilya's successor after she leaves.

Let's take a look at the resume of this tech mogul.

Source: LinkedIn

Pachocki holds a Ph.D. in theoretical computer science from Carnegie Mellon University, served as director of research at OpenAI, led the development of GPT-4, and participated in fundamental research on large-scale RL and deep learning optimization. As previously announced, Pachocki has played a key role in refocusing OpenAI's vision on scaling deep learning systems.

Source: LinkedIn

View original image 50K

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

•