The AI revolution of chain games (2): How does OpenAI become a god?

Original title: "The AI Revolution of Chain Games (2) How OpenAI Conquers Gods?" by W Labs AI Chain Games Research Group"

Original source: W Labs Guadain Lab

How does OpenAI become a god?

Givel a shot! The book continues from the above, let’s talk about the legendary history of OpenAI, which is now exploding all over the world. The starting point of the story is a dinner in Menlo Park, California in July 2015. In addition to the well-known global top-ranking V Elon Musk, the participants of the luxurious dinner were also gathered, including YC at that time. President Sam Altman (YC is a well-known startup incubator in Silicon Valley, incubating about 1,900 companies), Greg Brockman, chief technology officer of the top unicorn Stripe, and in 2019 for his contribution to deep learning and outstanding Joshua Bengio who won the Turing Award together with Free Hinton and Yang Likun. These people constitute the founding team of OpenAI.

(Sam Altman and Elon Musk)

(Greg Brockman and Ilyasutskevi)

Bengio gave a list of the best researchers in the field of deep learning, and Brockman contacted the list one by one persuade. Although there is a $1 billion investment commitment from the benefactor’s father, Microsoft, as a start-up organization, OpenAI still cannot offer a salary comparable to that of big companies such as Google and FB. In the end, it almost relies on great vision and feelings Convinced all the top talents to join us, it is the development of this direction (General Artificial Intelligence AGI) that can benefit all mankind, let's become figures that can be engraved in the century-old history of science and technology together!

It is not enough to have an amazing talent density (some departments of many large companies also have a high-density talent reserve). Thanks to OpenAI's firm belief in AGI from top to bottom, as well as the experimental mechanism to promote innovation from bottom to top, it is possible to achieve positive results only after taking everything from top to bottom.

How strong is OpenAI's belief in AGI? OpenAI employees vote every year to predict the arrival time of general artificial intelligence AGI. Even within this matter, there are relatively large differences of opinion, and different employees have their own ideas on this. But as early as February 2020, which was still in the GPT2 period (four months before the release of GPT3, and two years and nine months before the release of ChatGPT), half of the employees bet that AGI will be realized within 15 years—that is, 2035. Let us remember this point in time and witness history together.

Early OpenAI was like a revolutionist, facing the already laid-out front, with sufficient human and financial resources and various resources Google and other giants, but it still hopes to be the first organization to achieve AGI-because it knows how a large company will be driven by interests, be held hostage by stock prices, and be dragged down by internal factional struggles on hilltops. How will it consolidate its position and try to monopolize, OpenAI's vision is to achieve general artificial intelligence and distribute the resulting benefits fairly to all.

OpenAI has never wanted to rule the world, but only to ensure that technology can develop safely and ultimately benefit all mankind. I once read an in-depth report a few years ago. The author once went to OpenAI to do a lot of first-hand interviews, and talked about many secrets of the OpenAI organization that were not known to outsiders. The article mentions:

The organization's first statement said it would "create value for everyone, not shareholders"; its charter - the The document is so sacred that employee salaries depend on how well they adhere to its terms—further declaring that OpenAI’s “primary duty is to humanity to place the overall interest above the individual’s own”;Moreover, Achieving AGI safely is so important that OpenAI will stop competing and partner with another organization if it comes close to achieving it first.

I was deeply moved when I read these. This kind of thinking has begun to evolve into religious values. I used to think that AI and Crypto were completely opposite things. When Crypto tried to return power to the people and empower individuals, AI was a symbol of monopoly, implying a greater centralization of power. But I didn't expect the early OpenAI charter to read a bit like Satoshi Nakamoto when he invented Bitcoin (An interesting timeline coincidence is that deep learning appeared in 2007, and Bitcoin appeared in 2008 coin, the great invention of front and rear feet).

I very much believe that the drafting of this charter contains the best idealism in the hearts of the first batch of founding members, including Musk, when they wrote When these words are spoken, they are sincere enough and believe them. In fact, Musk himself has always been a representative of this kind of willingness to share the fruits of human wisdom. He has achieved maximum open source and collaboration in both Tesla and Twitter.

When OpenAI was pressured by reality and eventually compromised and betrayed its own revolutionary ideals to a certain extent, Musk also left OpenAI directly board of directors. Perhaps because he has seen the capabilities and potential of AI at the forefront, he has never stopped calling for AI safety over the years, and has made a lot of voices by using his public influence, even as an enterprise Home to take the initiative to ask the government for supervision.

Unfortunately, most people (including some members of our AI chain game research team) narrowly understand that his appeal and stop are "to give yourself time catch up with rivals”—in fact, the direction of Tesla’s early research and development has never been AGI, and Musk’s appeal did not start after the appearance of ChatGPT—this is somewhat “humans never thank Luo Ji” ( A famous sentence from the novel "Three-Body Problem", friends who haven't read the Three-Body Problem just skip it).

Altman, who stayed and took over as CEO, is also related to Crypto. His other project besides OpenAI is the encryption project Worldcoin. Compared with OpenAI’s hope to realize the liberation of all mankind through AGI, Worldcoin wants to achieve human fairness and inclusiveness at the financial level from the way of Crypto.

Back to OpenAI. In fact, OpenAI chose three for its early development direction, one is robot (artificial intelligence in physical or hardware form), one is game artificial intelligence (playing games), and the third is language model, so that artificial intelligence Learn to understand the world with text language, which is the GPT series. It can be seen that the exploration at this stage is relatively scattered. The team did not directly understand how to go about AGI with the script of the Chosen One, but finally completed the focus after various attempts.

The route of the robot may be better understood by everyone, and it is also the most common imagination of artificial intelligence that is very intuitive.

The road of games is recommended by Musk. There are many books about games that like to compare anything humans do in the world to some kind of game , In fact, the process of human learning and understanding of the world, as well as the resulting behaviors, do have many similarities with games. I don’t know whether to say that the world is a game or the game is a world. The previous AI has successfully challenged sports such as chess and Go, which are actually considered games. The OpenAI team chose Dota2, and in the end the challenge was successful, all the way to the world championship. In the next article, we will gossip about how games promote the development of technological productivity.

The turning point happened in 2019. This year, GPT-2 with 1.5 billion parameters trained on 8 million Reddit forum posts and a total of 40GB of text made the entire company really realize that it was on the right path that no one had traveled. The direction is confirmed to be correct, and the next step is to increase the size vigorously. In order to concentrate resources, the team aggressively cut down the teams that had already made some achievements in the other two directions, and went all out to make large-scale models. This time, we used hundreds of billions of English words (English Wikipedia only accounts for 3% of the data volume) and tens of millions of dollars in computing resources to develop GPT-3, and then developed another order of magnitude larger GPT without stopping. -4.

It may be hard for you to imagine that ChatGPT was only temporarily developed by OpenAI in mid-November 2022, and the whole process only took 13 daysbetween. I can’t help but feel that sometimes the course of history is rewritten inadvertently like this—the eagerness to release ChatGPT is because I am worried that Google will release similar products first, which will bring threats to myself. posed a truly massive threat and completely rewrote the industry landscape.

Many people describe that the significance of OpenAI's journey to large-scale models is no less than that of Columbus's arrival in the New World. Companies that enter the market after that can use relatively low Develop your own ChatGPT-like large language model with much cost and little risk.

Today, when we look back on OpenAI’s magnificent development in just a few years,on the one hand, we feel that the pressure of competition and the business environment have forced it to deviate from itself The original oath, on the one hand, can see that its compromise with reality is still within sufficient limits, and it is still doing its best to promote the realization of a great vision of general artificial intelligence for the benefit of all mankind.

An era of artificial intelligence that is as important as the emergence of computers and the Internet back then, and is about to profoundly change human society again, will start from the last month of 2022 officially opened. Everyone excitedly played with the latest model and discussed the future of human beings and artificial intelligence. Some people are optimistic, thinking that they have the strongest productivity helper so far; some people are pessimistic, thinking that human beings have already walked on the road of self-destruction. And no matter which point of view you hold, the only certainty about the present is how lucky we are to be born at the right time, compared with the very uncertain future.

To be continued. This series is collectively created by the W Labs "AI Chain Game Research Group". Thanks to the team members Gua Ge, Jia Ran, Bao Bao, Brian, Xiao Fei, and Hua Ge for their hard work! The main author of this article is the bag.

This article is from a contribution and does not represent the views of BlockBeats

欢迎加入律动 BlockBeats 官方社群：

Telegram 订阅群：https://t.me/theblockbeats

Telegram 交流群：https://t.me/BlockBeats_App

Twitter 官方账号：https://twitter.com/BlockBeatsAsia