The “mastermind” behind OpenAI Gongdou is ChatGPT

Just when I (naively, again and again) thought that the OpenAI turmoil was about to settle, it always turned around in a way that even the screenwriters of Succession couldn't have written.

After more than 90% of OpenAI employees completely broke up and jointly demanded the dissolution of the board of directors and the return of Sam Altman, the board of directors has not given up resistance. They even approached rival Anthropic and tried to merge the two companies.

The board of directors hopes to persuade Anthropic co-founder and CEO Dario Amodei to take over as CEO to complete the historical feat of replacing four CEOs in four days.

Dario Amodei was also a former OpenAI executive. Anthropic, which he founded after leaving his job, focuses on promoting safe AI, and its chatbot Claude is a competitor to ChatGPT.

Analysts believe that if OpenAI and Anthropic really merge, it will be the beginning of another chaos, because behind the two star startups are several technology giants' positions in the AI ​​field.

Microsoft and OpenAI team up, and Anthropic has signed multi-billion dollar cloud computing agreements with Amazon and Google. Both giants have also invested in Anthropic.

Currently, OpenAI is on the verge of disintegration. Competitors are beginning to recruit a large number of outflow talents. At the same time, OpenAI's major customers are also considering switching to companies such as Anthropic, Microsoft, or Google.

As we continue to track developments, we also look back at the starting point of this upheaval—OpenAI’s internal conflicts. In its latest report, The Atlantic Monthly unearthed a long-buried internal conflict of ideas.

OpenAI’s chaos started the day ChatGPT was released

1. OpenAI did not originally plan to release GPT-3.5. In the fall of 2022, OpenAI was originally preparing to release GPT-4. However, there were rumors that Anthropic was also going to release a chatbot, so the OpenAI management team announced in November that the product would be released a few weeks later, choosing GPT-3.5, which was relatively complete at the time.

2. No one expected the success of ChatGPT. The release of ChatGPT was very low-key, so low-key that many employees in the company did not know that OpenAI had released it. At that time, employees who knew about it were still betting on how many users it could attract, and they only guessed 10,000 users at most. As a result, the number of users exceeded 1 million within five days of launch.

3. ChatGPT not only brings down the server, but also renders detection tools ineffective. The team built a traffic monitoring tool that can track user behavior, understand how users use this tool, and coordinate potential crises and abuse. The popularity of ChatGPT caused the tool to crash repeatedly.

4. OpenAI’s security team advised the product team to slow down product updates. After the release of ChatGPT, the number of users grew rapidly, and the security team failed to keep up to deal with security risks and malicious uses. Subsequently, the product team ignored the reminders of the security team and persisted in pursuing the victory, launching the paid version of ChatGPT and the open API interface one after another.

5. Fraud crisis. With the widespread application of ChatGPT and GPT-4, some people have also found vulnerabilities. Some users began to create new accounts on a large scale to get the $20 discount each new account brings. Stopping these scams became OpenAI's top priority at the time, and a small team of security team members originally assigned to different areas were assigned to focus on solving the problem.

6. Under pressure, some employees even develop mental health issues. Communication within the company was poor, and employees only found out they were being fired when they noticed their co-workers disappeared on Slack.

7. "Feel the AGI". Technology is advancing rapidly. Chief scientist Ilya Sutskever is becoming more and more convinced that AGI will eventually be realized, and his behavior is becoming more and more like some kind of "spiritual leader." His mantra is "Feel AGI," which means that everyone is very close to achieving the goal of AGI. . Sutskever once posted on social media, "If you can feel AGI, you should apply for OpenAI." At OpenAI's 2022 holiday party, Sutskever even led employees to shout "Feel AGI! Feel AGI!" Employees also made "Feel AGI" emojis on Slack.

8. The other side of the belief is the concern that OpenAI does not pay enough attention to security. In July this year, Sutskever and Jan Leike led the establishment of the "Superalignment" team to develop more upstream AI security technologies and monopolize 20% of the company's existing computing chips to prepare for the upcoming AGI in this decade. "Alignment" means letting AI do only what people want them to do.

9. At an internal event this year, Sutskever even customized a wooden statue to represent "unaligned AI" and then burned it on fire to represent OpenAI's determination to align AI.

10. In October this year, conflicts escalated. People familiar with the matter said Altman aimed to reduce Sutskever's influence in the company, triggering a conflict with the board of directors.

11. At the developer conference on November 6, Altman announced the launch of customizable GPT functions. Sutskever was completely angry and went to the board of directors with concerns.

12. On November 17, Altman was fired by the board of directors during an online meeting.

13. Faced with many rumors that Altman will return, Musk said on X that he would be very worried if Altman really returned. Because "Ilya has a very good moral code and will not pursue power. If he didn't feel that he had to take action, Ilya would not have such extreme behavior."

14. On November 20, Sutskever told employees that Altman would not return to OpenAI as CEO. Emmett Shear, co-founder of Amazon's live streaming platform Twitch, will serve as OpenAI's interim CEO.

Behind the Chaos: Sutskever and Altman’s Philosophical Clash

1. Chief scientist Ilya Sutskever, who leads the development of OpenAI’s cutting-edge technology, studied under the “Godfather of AI” Geoffrey Hinton and advocates cautious handling of the security risks of AI and AGI.

2. Sutskever’s performance has been stellar. In the early days of his joining the Hinton team, Hinton gave him a very complex task. Sutskever was tired of the calculations required in the writing process and told Hinton to write a programming language himself to complete the simpler task. Hinton got a little angry and warned him that it might take a whole month and lead to nothing. "I actually finished it this morning," Sutskever confessed.

3. High status in the industry. Sam Altman called Sutskever one of the most respected researchers in the world and believed that bringing him in would attract more AI talents. Musk frankly believes that he is a key figure in the success of OpenAI.

4. In order to lure Sutskever from Google to OpenAI, Musk had a falling out with Google co-founder Larry Page.

5. Sutskever’s view on AI: Technology is like a force of nature. It is crucial to control the initial birth of AGI, otherwise it will follow a law similar to "survival of the fittest" and prioritize its own existence. Just like humans and animals, humans do not hate animals, but when humans want to develop, we will put the need to build highways over the survival of animals.

6. When that historic paper "Attention Is All You Need" came out, Sutskever was one of the few people who could see its importance: "This is what we're waiting for."

7. Sutskever advocates "engraving" specific values ​​into super AI, so that AI has some "imprint" similar to "nature" like animals: "Eventually, AI will become very, very, very powerful. We will not be able to understand it. They. They will be much smarter than us. At that time, a strong imprint will be crucial, so that they will feel for us the same way we feel for our own children."

8. The fight for AGI is futile. Sutskever believes that if there are multiple teams competing to make AGI first, there will be less space for them to work to ensure the security of AGI. In contrast, Altman calls himself more of a "centrist" who wants to ensure both speed and safety.

9. The first batch of AGIs in Sutskever's mind looked like super huge data centers with built-in proprietary neural networks, which were extremely energy-consuming and consumed approximately the same amount of electricity as tens of millions of homes. These AGIs will certainly profoundly impact society, but we’re not sure who will benefit.

10. Former OpenAI employees say Sutskever has many followers internally, calling him a well-respected and hands-on leader who is crucial to building OpenAI's cutting-edge technology. Greg Brockman, former president of OpenAI, is good at transforming Sutskever's research results into usable AI products.

Timeline of Sam Altman's ouster:

Thursday night, November 16th local time

Altman received a message from Sutskever requesting a conversation at noon on Friday. Sutskever tells Sam he's fired and the news will be out soon.
Friday, November 17, local time

OpenAI announced that CEO Sam Altman will leave the company, and Chief Technology Officer Mira Murati has been appointed as interim CEO, effective immediately.

The OpenAI board of directors stated in a blog post that "Mr. Altman has not been candid in his communications with the board of directors, which has hindered the board of directors' ability to carry out its responsibilities. The board of directors no longer has confidence in him to continue to lead OpenAI."

Altman posted on the X platform that I loved my time at OpenAI. It was transformative for me personally and I hope it will also change the world. Most of all, I love working with so many talented people.

OpenAI President and Board Chairman Greg Brockman announced his resignation. He said he was shocked and saddened by the board's decision.

Three senior researchers at OpenAI resigned on Friday night local time.
Saturday, November 18th local time

Sam Altman told investors he plans to launch a new AI venture. Greg Brockman is expected to join the venture, which is still under development and the exact nature of the venture is unclear at this time.

OpenAI's board of directors is discussing a return to the CEO role with Altman. He is "ambivalent" about the return and wants major changes to governance.
Sunday, November 19th local time

Sam Altman and Greg Brockman arrived at OpenAI's San Francisco headquarters to meet with executives.

OpenAI interim CEO Mira Murati plans to rehire Sam Altman and Greg Brockman, while board members continue to seek a new CEO.

The OpenAI board of directors chose to hire Emmett Shear, the former CEO of Amazon's live game streaming platform Twitch, as CEO, and Altman's return failed.

Upon learning that Altman would not be returning, dozens of OpenAI employees announced internally that they were quitting the company.

Microsoft CEO Satya Nadella announced that OpenAI founders Sam Altman and Greg Brockman and their colleagues will join Microsoft to lead a new advanced artificial intelligence research team.
November 20, local time

More than 710 OpenAI employees asked the board of directors to resign, otherwise they may defect to Microsoft collectively

To be continued…

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo