Since this year, there has been a hot word in the field of artificial intelligence: gpt-3. It was developed by openai, a non-profit organization of artificial intelligence, at a cost of 12 million US dollars, covering 175 billion parameters. It has reached the best SOTA at present, and its writing level is comparable to that of human beings. Thanks to various auras, gpt-3 has become an AI artifact. In September, openai decided to open the gpt-3 commercial API, and the pricing scheme was also announced. Just a few days ago, openai and Microsoft announced that,microsoft gets GPT-3 exclusive authorization.
The decision, however, has sparked anger among OpenAI co-founders Ilon Musk (Elon Musk) and well-known artificial intelligence scholar Gary Marcus (Gary Marcus), among others.
"Artificial intelligence research will move towards democracy"
On September 22, 2020, Microsoft posted a blog on its official website, saying that Microsoft cooperated with openai to obtain the exclusive license of gpt-3 language model.
Kevin Scott, executive vice president and CTO of Microsoft, said in his blog:
With the exclusive authorization of gpt-3, Microsoft will develop and deliver advanced artificial intelligence solutions for customers through technological innovation, and create brand-new solutions with the amazing power of gpt-3. Microsoft will make the most cutting-edge AI research move towards democracy, so as to help achieve secure general artificial intelligence.
Kevin Scott believes that the gpt-3 model has huge commercial and creative potential, and many possible developments are beyond our imagination. To really stimulate the potential of gpt-3 in a large-scale, responsible and economic and fair way, it is not only a technology company that can achieve it, but also requires more human resources.
The whole post is like a sentence - we see the great value of openai.
Kevin Scott also hinted that AI translation tools will be built in the future, and even gpt-3 will be used to help human writing, but the specific commercial use is not clear.
So what does openai say about exclusive licensing?
OpenAI made it clear in his blog post 22 that "exclusive authorization" does not affect users' continued access to GPT-3 models through API, and that future users will still be able to build applications through the API.
However, according to the verge, a Microsoft spokesman said the decision gave Microsoft the exclusive right to use the underlying code of gpt-3, including advanced technologies that Microsoft hopes to add to its products and services.
Looking back on the cooperation between openai and Microsoft, July 2019 is a landmark time node.
At the time, Microsoft announced a $1 billion investment in openai. As the exclusive cloud service provider of openai, Microsoft will cooperate with openai for a long time to build new azure AI supercomputing technology and further develop AI capabilities.
In exchange, openai licensed some of its intellectual property to Microsoft.
After that, Microsoft commercialized these intellectual property rights and sold them to partners. In the process of openai developing the next generation of computing hardware, Microsoft trained and ran AI models through azure.
Less than a year later, the collaboration came to fruition —— on may 19,2020, microsoft announced a supercomputer designed for OpenAI AI models at its developer conference.
Microsoft says,This supercomputer ranks among the top 5 in the world in terms of performance.
The supercomputer, with 285000 CPU cores, 10000 GPUs and 400gbps network connections, is dedicated to training large-scale AI models, enabling them to access billions of pages of text from published books, teaching manuals, history lessons, human resources guides, and other open sources.
From the start of investing, announcing overstating, to exclusive licensing, OpenAI seems a step closer to its $1 billion general artificial intelligence dream.
"Openai might as well be renamed closed AI"
And on the other side, many Twitter netizens have shown dissatisfaction.
For example, whole Mars catalog believes that openai should make gpt-3 widely used, and should not give exclusive license to Microsoft. After that, whole Mars catalog also cue Elon Musk, co-founder of openai.
Elon Musk, who is active on Twitter every day, of course, also replied, with a clear point:
Just like another netizen, disrupt or distort said:
Openai was renamed closed AI.
Elon Musk as a co-founder, talking to OpenAI so unkind, what is the reason behind it?
The view of foreign media Business Insider is that Elon Musk may be targeting Microsoft co-founder Bill Gates (Bill Gates).
Indeed, Elon Musk and Bill Gates two technology leaders have not been less resentful of each other. On June 30, Elon Musk also "refuted rumors" that Bill Gates and I are lovers can rest.
In a recent tit for tat between the two, Bill Gates blogged at the end of August that pickup trucks (such as Tesla semi) and electric jet planes are likely never to appear. Elon Musk, of course, was not happy. He said, "he doesn't understand.".
After all, it's not the first time Elon Musk has publicly criticized openai.
As early as 2015, Elon Musk, the founder of LinkedIn, the president of Y combinator and the founder of pay pal, jointly announced the establishment of open AI, with the goal of building an open AI belonging to all mankind.
OpenAI officials announced Elon Musk withdrawal from the board in February 2018 to avoid conflicts of interest due to Tesla. Nevertheless, Elon Musk will continue to serve as OpenAI adviser and provide financial support.
At that time, it was speculated that Elon Musk would occasionally ask openai scientists to give Tesla advice. Openai became Tesla's "foreign aid".
And the root cause lies in the inherent contradiction —— the two If OpenAI is ideal, then Tesla is reality, and ideal and reality can not coexist.
In February 2020, Elon Musk tweeted that he had "low confidence" in openai's security measures in artificial intelligence, and also mentioned that he had "no control" and "limited insight" in openai recently.
Except for Elon Musk, Robust.AI Gary Marcus, the founder and CEO of artificial intelligence, also came to know the news that Microsoft had been granted exclusive authorization. It turned out that democracy is the only way to have unique power.
What is worth mentioning is that Gary Marcus GPT-3 of OpenAI has always been very stereotyped.
Gary Marcus tweets at the top of the Twitter are GPT-3： spray
The two generations of language model before gpt-3 are "shit artist". Although gpt-3 is better, it is still "shit artist" in essence.
But one of the peculiarities of gpt-3 is that it is regarded as dog's excrement while being praised to the sky.
In terms of performance, the larger the model is, the better the performance of gpt-3 is in terms of total number of training, number of layers and learning ratio.
Answering questions, writing articles, translating, generating code, doing mathematical reasoning, data analysis, drawing charts, making resumes, playing games are omnipotent, and the effect is very good. Gpt-3 has become the most popular topic in AI field, and netizens exclaim that gpt-3 has become excellent.
However, even if you can play more than 50 tasks, gpt-3 can't even pass some basic common sense problems and mathematical logic.
Sam Altman, co-founder of openai, also believes that gpt-3 is over hyped:
There is too much hype about gpt-3. It's still seriously flawed and sometimes stupid.
What's more, gpt-3 has sparked a heated debate about the moral issue of powerful AI programs that can be used for evil purposes.
At first, openai refused to publish research on gpt-3 because of concerns about its abuse.