Highlights of AI technology 2024

---

AI in 2024 is considered to be approaching "artificial super intelligence", but also causes many concerns, thereby promoting regulations on responsibility and ethics.

Artificial intelligence has become a mainstream technology, permeating every aspect of life. A series of large companies are spending billions of dollars on poorly trained systems, creating a fierce race.

The world is getting closer to AGI superintelligence

Since the beginning of the year, experts have been debating "when AI will be smarter than humans ". Most predict this will take a few years, but new advances show that superintelligence AGI may appear sooner.

ChatGPT logo on smartphone. Photo: Reuters

While many big players have released large language models (LLMs), OpenAI has always been at the forefront. In the second half of 2024, Sam Altman’s company has released a series of improvements such as GPT-4, GPT-4o, especially o1 and o3.

Among them, o1, codenamed Strawberry, was announced by OpenAI in September with the ability to process complex questions, and can "create a long chain of internal thoughts before responding to the user". At that time, Business Insider assessed that the appearance of o1 made "the boundary between human intelligence and artificial intelligence increasingly narrow". One study even showed that o1 was capable of rebelling, plotting and lying when it thought it was about to be replaced .

On December 20, OpenAI continued to introduce the o3 and o3 mini models that can "reason " with more power than the o1. The company called this an "impressive leap in performance", and experts also reacted positively to the information, predicting that AGI could appear sooner.

The AI ​​Video Generation Wave

Following the AI ​​image-generating craze of 2023, this year a series of companies launched tools to turn input text or images into short videos. The wave started in February when OpenAI announced Sora .

Image cropped from video created with Sora AI. Photo: OpenAI

In May, Google unveiled an AI called Veo that can create 1080p videos longer than a minute. In July, Runway, the AI ​​VFX company behind the Oscar-winning film Everything Everywhere All At Once , announced its Gen-3 Alpha model, which “significantly improves fidelity, consistency, and motion over its predecessor.” In October, Meta released an AI called Movie Gen that creates video and audio from text, and Adobe introduced its own AI tool. Meanwhile, several Chinese companies have also launched models to compete with their American counterparts.

However, apart from Sora, which was launched commercially in early December, most are still in limited testing. In addition, the video quality is considered not really perfect because the AI ​​does not fully understand complex physical movements.

AI chip fever

The hunt for AI chips, primarily Nvidia GPUs, began last year but has accelerated this year. In January, Meta CEO Mark Zuckerberg announced that he would buy 350,000 H100 GPUs in 2024, bringing the total number of chips they own to 600,000.

Meanwhile, according to Business Insider , Microsoft aims to own 1.8 million AI chips, and plans to invest $100 billion in GPUs and data centers by 2027. Analyst firm Omdia estimates that Microsoft has purchased 485,000 "Hopper" (H100) chips as of the end of November, double Meta's 224,000 chips.

The HGX H200 AI chip uses eight Nvidia H200 GPUs. Photo: Nvidia

However, the person who actively collects AI chips the most is Elon Musk. Since announcing the construction of the Colossus supercomputer in May, he has accelerated the progress, completing the project in 122 days , which Nvidia CEO Jensen Huang called " unbelievable ".

Colossus uses 100,000 Nvidia H100 GPUs and is expected to increase to one million GPUs in the near future. In mid-September, Oracle founder Larry Ellison revealed that he and Musk had to " beg " Jensen Huang to agree to sell them the latest GPUs.

Meanwhile, in mid-December, Broadcom announced that “three major data center customers” are also on track to buy its AI chips, called XPUs. By 2027, these customers will deploy clusters of one million XPUs on a single interconnect architecture.

China cannot directly buy advanced GPUs due to the US embargo. However, many Chinese companies are still trying to build AI training machines with domestic chips, mainly supplied by Huawei, even " going around " to own the most powerful chips. According to Omdia, ByteDance and Tencent have ordered about 230,000 Nvidia chips this year, including the H2o model customized to comply with US export controls.

AI makes its mark at the 2024 Nobel Prize

On October 8, two scientists, Geoffrey Hinton and John Hopfield, were awarded the Nobel Prize in Physics for their pioneering research that laid the foundation for modern AI. On October 9, three experts, David Baker, John Jumper and Demis Hassabis, received the Nobel Prize in Chemistry for their work in decoding the secrets of proteins through AI.

From left: Demis Hassabis, John Jumper and Geoffrey Hinton. Photo: Google/AP

However, some experts believe that the AI ​​craze sweeping the world is affecting the Nobel Prize. The work of the laureates is fully worthy and deserves recognition, but the lack of a Nobel for mathematics or computer science has skewed the results, according to Professor Dame Wendy Hall, a computer scientist and AI adviser to the United Nations. "The Nobel committee did not want to miss out on the AI ​​trend, so they were very creative in having Geoffrey win the prize through physics," she told Reuters .

Agreeing, Noah Giansiracusa, a professor at Bentley University, said: "What he did was extraordinary, but was it physics? I don't think so. Even if he was inspired by physics, he didn't develop a new theory in physics or solve a long-standing problem in physics."

Some other experts say AI is so pervasive in academic research that the boundaries between fields are blurring, and "there is such admiration for computer scientists that they are classified in any field."

OpenAI's Internal Turmoil

After a failed CEO ouster in late 2023, OpenAI’s internal turmoil continued. In May, co-founder and chief engineer Ilya Sutskever , who was behind the firing of Sam Altman, announced that he was leaving OpenAI along with machine learning researcher Jan Leike.

Ilya Sutskever (right) and Sam Altman. Photo: Abigail Uzi/YnetNews

Then, two founding members, Andrej Karpathy and John Schulman, left. In October, Mira Murati , OpenAI's chief technology officer and the "woman behind ChatGPT," unexpectedly resigned. Hours later, Altman announced that director of research Bob McGrew and vice president of research Barret Zoph were also leaving.

At that time, citing internal sources, Business Insider said a series of power struggles within the company caused OpenAI to lose its brainpower . From non-profit activities and developing AI for humanity, Altman directed OpenAI to become a for-profit company, making many employees feel dissatisfied.

OpenAI and Altman are also embroiled in a lawsuit with Elon Musk, who accuses the company he co-founded of straying from its original mission. However, in the latest document, OpenAI responded that Musk had wanted them to be a for-profit company from the beginning.

13,500 creators oppose AI

In mid-October, more than 13,500 creative people from around the world in the fields of literature, music, film, theatre and television signed a letter protesting the illegal exploitation of their work to train AI. The letter was published by former CEO of Stability AI and British musician Ed Newton-Rex, stating: "The illegal use of creative work to train AI is a huge and unjust threat to the livelihoods of the people behind the work. It should not be allowed."

From left: ABBA singer Björn Ulvaeus, actress Julianne Moore and 2017 Nobel Prize winner in Literature Kazuo Ishiguro are among the signatories of the letter against AI. Photo: AP/ Vanity Fair

Previously, in May, actress Scarlett Johansson also threatened to sue OpenAI for copying her voice for the virtual assistant Sky in the GPT-4o model. On Instagram in August, actor Tom Hanks posted a warning to fans not to believe the advertisement for a miracle drug associated with his image and voice on the Internet, emphasizing that it was an AI product.

In June, three record labels—Sony Music, Universal Music Group, and Warner Music Group—filed a lawsuit against two AI music creation platforms, Suno and Udio, accusing them of using more than 1,000 recordings as training data. To date, many lawsuits between artists and companies that own AI tools continue.

Responsible AI Regulation Series

In March, the United Nations General Assembly adopted its first resolution to protect human rights, protect personal data, and monitor risks posed by artificial intelligence. It called on member states and stakeholders to refrain from using AI that is inconsistent with international human rights law.

In August, after European Union (EU) lawmakers voted in favor, the AI ​​Act was passed, considered the most comprehensive piece of legislation, ushering in an era of safety and responsibility for artificial intelligence in the region. The law requires ensuring safety and security, personal information and preventing risks related to AI, enhancing the development and application of AI in various fields as well as their legal responsibilities.

European Union lawmakers vote on the AI ​​Act at the European Parliament in Strasbourg, France. Photo: AP

Outside of Europe, this year alone, several regulations have been issued to mitigate concerns about privacy, safety, and discrimination related to AI, such as those from the US National Institute of Standards and Technology (NIST), the United Nations Educational, Scientific, and Cultural Organization (UNESCO), and the International Organization for Standardization (ISO).

In the US alone, nearly 700 AI laws were introduced in 45 states in 2024, a sharp increase from 191 last year. At the end of October, US President Joe Biden also passed an executive order outlining the task of developing AI safety.

Vietnam promotes AI research and development

In April, the first AI factory in Vietnam was announced through a partnership between Nvidia and FPT with the aim of developing artificial intelligence applications for Vietnam. In December, the Vietnamese government and Nvidia established the AI ​​Research and Development Center, as well as the AI ​​Data Center.

The two agreements are considered "important milestones" in making Vietnam become Asia's leading center for artificial intelligence research and development in the future, creating breakthroughs for key technology industries and opening up career opportunities for thousands of domestic talents.

Meanwhile, according to Cisco’s November study of 14 Asian markets, Vietnam is one of the bright spots, with 100% of businesses participating in the survey affirming that the demand for AI is increasing this year. Vietnamese businesses also plan to invest more in the field, with 48% saying they will spend 10-30% of their IT budget on AI deployment.

The rate of AI application in Vietnamese and world work according to a survey by Microsoft.

According to a survey of 31,000 employees from 31 countries by Microsoft in May, 75% are using generative AI at work. In Vietnam alone, the rate is 88%, higher than the world average. "Vietnam needs to promote more mechanisms to encourage organizations, businesses and leaders to not only experiment but also take advantage of AI as one of the main means to achieve business goals and optimize organizational efficiency," said a Microsoft representative.