Written by/ Zhang Linyu
Editor/Zhang Nan
Design / Ju Jia
Source: The New York Times, by Cade Metz
In the U.S., ChatGPT (or GPT-3.5) isn't the only AI product based on big language models, and its competitors include other products like Google's Bard, which began small-scale testing in March.
Cade Metz, a technology reporter for the New York Times, was invited to experience the product and concluded that Bard was more cautious than GPT-3.5 or GPT-4.0.
Kod has covered AI-related technologies for more than a decade, and he has published a book about the geniuses of AI.
In the summer of 2019, he received an invitation from Sam Altman, known today as the "father of ChatGPT," who had just received a $1 billion investment contract from Microsoft. They had dinner at a small modern restaurant near San Francisco, where Sam lived.
Between meals, he held up his phone and showed Codd the contract he had just signed with Microsoft, which stated that Microsoft would invest $1 billion to help OpenAI build Artificial General Intelligence.
Sipping sweet wine, Sam described OpenAI's plan, which he likened to the power and significance of the Manhattan Project.
The Manhattan Project was the U.S. Army's program to develop atomic bombs using nuclear fission reactions in 1942. At that time, the project brought together the best nuclear scientists in the West and mobilized 100,000 people to participate in the project, which lasted for three years. The first nuclear explosion was successfully carried out in 1945. The positive and negative deterrent effects of atomic bombs on the world are known to every scientist.
Sam told Codd that Julius Robert Oppenheimer, the leader of the Manhattan Project, had the same birthday as him, both on April 22, and he hinted at his fate.
OpenAI had just transitioned from a nonprofit to a for-profit organization that year, Sam resigned as president of Combinator Y to personally serve as CEO of the company that was only 4 years old, and it was also in this year that he fell out with donor Elon Musk, presumably because of the change in the nature of the organization or Microsoft's capital injection, but the conflict between the two was not disclosed at the time. Their grudges read
"Maskela hacked OpenAI"
It wasn't until nearly two weeks that Musk took the initiative to mention it on Twitter that the dispute between them came to light. Interestingly, a month ago, Musk also gave an interview to Sam at the Tesla factory, and the two discussed how young people can start businesses.
That night in 2019, Sam compared the impact of OpenAI's product to an atomic bomb, and he talked as lightly as if he were talking about parents, but at that time OpenAI had less than 100 people, and everyone didn't know where Microsoft's next money would come from.
Codd sat across from Sam and felt like he was listening to a science fiction novel.
At the time, Sam told Codd that OpenAI was developing products that could bring prosperity and wealth to the world, but it could also put many people out of work and destroy the world.
"I don't know if I'm doing good or bad," Sam said.
boycott
From the release of ChatGPT in the fourth quarter of last year to the launch of GPT-4 this year, different voices appeared at the same time.
The Future of Life Institute released an open letter in March signed by more than 1,000 AI experts and technology leaders calling on OpenAI and other companies to suspend work on systems like ChatGPT and stop training AI systems more powerful than GPT-4 for at least six months. They believe that there is a "great risk to society and humanity." Among those who called for it was Musk.
At the same time, the Biden administration has stepped in and invited industry experts to evaluate the product.
On March 31, the Italian Personal Data Protection Agency launched an investigation into OpenAI for alleged violations of data collection rules, and announced a ban on the use of ChatGPT and temporary restrictions on OpenAI's processing of Italian user data.
The reason for the ban by the Italian Personal Data Protection Agency is: "There is no legal basis to justify the large-scale collection and storage of personal data for the purpose of 'training' the algorithms behind the operation of the platform." The regulator also accused OpenAI of failing to effectively identify user age, and under OpenAI terms, users of ChatGPT must be at least 13 years old.
On April 5, local time, in a video conference with the Italian Personal Data Protection Agency, OpenAI promised to be more transparent in the way it handles users and verifies their age. Sam attended the meeting. They submitted a document on remedies to the Italian Personal Data Protection Authority on April 6, after which the Italian Data Protection Authority will evaluate the option.
In the mainland, Baidu's Wen Xin has been released, more Internet giants or start-ups are rapidly entering this track, and data security is also a topic of concern.
About GPT-4
Compared to GPT-3.5, the main advantage of GPT-4 is that it is already very proficient in natural language, and you will not feel like you are talking to a machine.
OpenAI's office in San Francisco▼
Not only does it produce fluent and coherent text, but it also understands and processes text in a variety of ways, such as summarizing, translating, or answering an extremely wide range of questions. In addition, the translation here refers not only to translation between different natural languages, but also to translation of tone and style, as well as cross-disciplinary translations such as medicine, law, accounting, computer programming, music, and so on. These skills clearly demonstrate that GPT-4 is capable of understanding complex thoughts.
Many may wonder if the GPT-4 really understands all of these concepts, or if it is simply much better at improvisation than previous models without any really deep understanding.
OpenAI's only test of "truly deep understanding" is whether it can generate new knowledge, such as proving new mathematical theorems, which is currently out of reach for large-language models.
GPT-4 outperforms ChatGPT (GPT-3.5) in real-world scenarios that require reasoning about the psychological states of others, and proposes cooperative actions for common goals in social settings. GPT-4 has a very advanced level of theory of mind.
While ChatGPT also performs well on basic tests, it seems that GPT-4 is more delicate and better able to reason about multiple actors and how various actions affect their mental state, especially in more realistic scenarios.
Simply understood, GPT-4 is more human-like than ChatGPT.
Some industry leaders, AI researchers, and pundits see GPT as a fundamental technological shift, as important as the invention of the web browser or iPhone, but the differences between them are mainly on the future shape of the technology and the use cases, while others are strongly opposed, and these more than 1,000 scientists and pundits have fully expressed their concerns.
Take one for ten
"The current hype around ChatGPT is out of our control in the short term, and it's completely out of control, but in the long term we hope to move in the right direction." Sam tells Cod that there will always be a stage where people will understand how these products will change the world.
Sam, 37, looks a little childish, especially in conversations with Musk. When he became the CEO of OpenAI in 2019, he told Kede that he did not own any shares, and that he was paid at OpenAI, $65,000 a year, to meet the requirements of social security.
Sam grew up in St. Louis, and in 2000 he enrolled in the local prep school, John Burroughs School, a private school named after the 19th-century American philosopher John Burroughs. The school was very anti-gay at the time.
Sam, 17, who entered the school as a gay student, began to spearhead the change in that ethos.
He alone persuaded teachers to put up "safe spaces" signs on classroom doors to show support for gay students. He also asked a particularly homophobic teacher to post a "safe space" sign just to mock the person.
He was very convincing. The teacher who taught computer science at the time thought that Sam, in addition to his talent in computers, could lead his classmates in new directions.
"He had the creativity and vision, combined with ambition and strength of character, to convince others to work with him and put his ideas into practice", the teacher told Codd.
The same is true of Paul Graham, Sam's mentor and founder of Y Combinator.
The two met in 2005, when Sam applied for Y Combinator's first start-up spots. He was the first to receive a spot, including $10,000 in seed funding, when he was a sophomore at Stanford before dropping out to start his own new company, Loopt, a social media startup that lets people share their location with friends and family.
"He was born with the ability to persuade." Graham said, "If it wasn't innate, at least he had fully mastered it before he was 20 years old." I first met Sam when he was 19 years old, and I thought he resembled Bill Gates. ”
Sam used his persuasion to single-handedly persuade Microsoft CEO Satya Nadella and Microsoft CTO Kevin Scott at an annual gathering of tech leaders in Sun Valley, Idaho, around the end of 2018, and they negotiated a deal.
Sam now says he learned more from many nights playing poker during his short stay at Stanford than he did from most other college activities, which taught him how to read people and assess risk.
He then sold Loopt, and after a small return, he joined Y Combinator as a part-time partner. Three years later, Graham stepped down as president of the company and appointed Sam, who was only 28 years old, as his successor.
Opposites and unity
Sam is not a programmer, engineer, or researcher in artificial intelligence.
He is more of a person who sets goals, agendas, brings teams together and closes deals. After his appointment as president of "YC", he launched a new investment fund and a new research laboratory, and increased the number of companies that provide consulting to hundreds per year.
During this time, he also began investing in several projects outside of the company, including OpenAI. Ultimately, he decided to focus on a project that he believed would have a real impact on the world. He had considered politics, but ultimately chose artificial intelligence.
OpenAI employees working in the coffee room▼
According to his younger brother, Max Altman, he believes he is one of the few people who can change the world through AI research.
At the same time, Sam is a man who lives in contradiction, just as the artificial intelligence he works on has the attributes of both extremes.
He lives Monday through Friday in downtown San Francisco, and on weekends he and his partner, Oliver Mulherin, an Australian software engineer, live on a ranch in California where farmworkers grow grapes to make wine and raise cows.
But Sam is a vegetarian and raising cows is strange, but he says: "Oliver loves it. ”
It is a quiet farm located between rocks and hills covered with grass. They remodeled the 25-year-old house, which looks earthy and modern on the outside, and the steel covering the exterior walls is completely rusted.
Sam lives between the city centre and the countryside, and lives in an environment where vegetarian diets are surrounded by meat products.
Cod thinks he's a product of Silicon Valley.
Grew rapidly in the mid-2010s and was ecstatic. As president of YC from 2014 to 2019, advising on a steady stream of new companies, he was savvy enough to personally invest in several household names, including Airbnb, Reddit, and Stripe.
He recognizes when a technology will grow exponentially and then follow the curve into the future.
Around the same time he arrived in Silicon Valley, AI was controversial. While he and his team are advancing artificial intelligence, they are also acutely aware of the lethal power of this technology.
He is not necessarily motivated by money. At OpenAI, he doesn't get rich with an annual salary of $65,000, and he previously generated a portion of his income through projects he invested in YC.
Graham said: "Some people stop when they make enough money, but Sam doesn't seem to be particularly interested in money. Another possibility is that he worked on OpenAI because he preferred power. ”
At a lake on a California farm, when Cod asked Sam whether a machine that mimics the human brain would eventually reduce the value of human labor to zero, he disagreed, saying he couldn't imagine the world where human intelligence would one day be useless.
Note: This article refers to the paper "GPT-4, The Spark of General Artificial Intelligence" translated by orange AI
This article was originally produced by Automotive Business Review