2023 has undoubtedly become the "Big Model Year", and OpenAI's GPT is a leader in the wave. With the support of investment and financing funds and customer groups, the company continues to upgrade its capabilities in large models, driving a new round of technological revolution towards a larger scale.
At 2am Beijing time on November 7th, OpenAI held its first developer conference in San Francisco, USA. At the meeting, the company updated the GPT big model to the GPT-4 Turbo version and further improved the business architecture of big model development, including launching "store" products that attract software developers, launching copyright shields, and providing API development assistants.
Some software developers have stated that they have seen the "iPhone moment" and the scale of AI engineers is expected to rapidly increase.
GPT model major upgrade
The GPT-4 Turbo launched by OpenAI this time focuses on three major features: "volume increase, speed increase, and price reduction". In addition, the company's business architecture has been further improved, with a greater emphasis on copyright and ecological development.
The three major characteristics of increasing volume, increasing speed, and reducing price reflect that OpenAI has undergone another major model upgrade in just 8 months, with an accelerated speed. In June 2020, the company launched the GPT-3 model; In March 2023, it launched GPT-4. Perhaps users are urging the company to accelerate technological updates, as Sam Altman, CEO of OpenAI, introduced at the meeting. As of now, the company has attracted approximately 2 million developers worldwide to develop applications around GPT, with over 92% of Fortune 500 companies becoming users of its products. The weekly active users of ChatGPT, a dialogue robot, have reached 100 million.
Specifically, the GPT-4 Turbo text interpretation length has been upgraded from 32k to 128k, which is equivalent to 100000 English words or 300 pages of English text content, and the processing speed per minute has also doubled. In addition, the input unit price of the new model will be reduced from 3 cents to 1 cent, and the output unit price will be reduced from 6 cents to 3 cents. For example, uploading a 1080x1080 pixel image to the GPT-4 Turbo for interpretation will cost $0.00765 (approximately RMB 0.06), while the starting price for each image generated using the cultural image function is $0.04 (approximately RMB 0.29).
OpenAI has also established a more comprehensive business architecture, with the newly launched "GPT Store" reminiscent of Apple's "App Store" launched in 2008. This "store" will also provide API assistants, with a pattern very similar to that of Apple, and may also attract more developers to develop applications around the GPT model.
In addition, OpenAI combines multiple star products under its umbrella in the GPT-4 Turbo, including the graphic tool DALL · E and speech recognition tool Whisper. This means that this large model can not only "speak", but also "see" and "listen".
Further differentiation in the large model industry
The competitive situation of the "Hundred Model Battle" may also enter the next stage with the aforementioned press conference. Will the situation where the strong win accelerate? At the above conference, OpenAI and Microsoft once again demonstrated their collaborative muscle of "big model development+cloud infrastructure services".
As a 'leader', OpenAI has achieved significant growth in revenue and valuation in 2023. In October, Altman revealed that OpenAI's annual revenue had reached $1.3 billion, an increase of over 45 times compared to 2022. At the same time, OpenAI is highly favored by market investment institutions, as it sold some equity to Sequoia and a16z in April. In a financing negotiation in October, its overall valuation had risen to $80 billion.
OpenAI has also gained the favor of technology giants, as artificial intelligence may become one of the most important driving forces for the future growth of cloud services. As the most well-known technology "partner" of OpenAI, Microsoft appeared at the aforementioned developer conference. Microsoft CEO Satya Nadella appeared around the 20th minute of the press conference and talked about continuing to collaborate with OpenAI in both infrastructure and application products. Among them, Microsoft Cloud Azure is an infrastructure that supports OpenAI for training, and many of Microsoft's products have already been embedded with AI functionality.
The partnership between big model developers and cloud infrastructure providers is further dividing the big model development industry. Many industry insiders have expressed the belief that there are already too many large models available, and people even need advantageous products to win as soon as possible.
According to statistics, as of July this year, American companies have developed a total of 114 large models, while Chinese companies have developed 130 large models. In addition, "one country, one model" has also become an industry phenomenon. India, Japan, the United Arab Emirates, South Korea and others have joined the team to independently build large models.
Software developers welcome the 'iPhone moment'
Although "competitors" big model developers, "sellers" cloud infrastructure suppliers, and "buyers" mass users are all paying attention to the OpenAI developer conference, "programmers" are the protagonist of this conference. Software developer Lin Yi (pseudonym) told reporters: "Programmers pay attention to OpenAI's launch event, just like the public pays attention to mobile phone launches. The updates to software development tools make the industry very excited." He also introduced that the group of AI engineers is rising, and the industry is ushering in the "iPhone moment".
About 900 software developers arrived at OpenAI's conference in San Francisco. Although the scale was not comparable to the developer conferences of companies such as Google, Apple, and Meta, the industry's influence was almost on the same level.
The performance and functionality of the latest model are the focus of developers' attention, while copyright protection and ease of development are also the main attractions. At the meeting, OpenAI also introduced the built-in copyright protection measure "Copyright Shield" and will provide API assistants in the "GPT Store" that will be launched next month. This indicates that OpenAI hopes to provide software developers with a more secure and convenient development environment.
Just like Apple launched an application platform based on the iPhone back then, will the development around GPT also create 'popular' applications in the future? "Lin Yi said. He said that from the initial blockbuster touchscreen games to later social and e-commerce software, and now to short video platforms, the "innovation" of an application platform will bring a "revolution" to half of the technology industry.
The emergence of large models such as GPT is also fundamentally changing the process of software development. Although large model development is currently very popular, AI engineers should be more needed in the future, "said Lin Yi.
AI engineers refer to developers who specialize in researching AI applications and applying emerging technology stacks. This group of people will work together with data engineers, analysis engineers, and operation and maintenance engineers to form sub positions. Unlike other engineers, AI engineers may prefer the latter in terms of research and engineering abilities, and require the ability to "talk" with AI, while the programming threshold may be lower.
Although there is currently no large-scale recruitment demand in the recruitment market, as more companies enter the development of AI application stores, relevant opportunities will be worth paying attention to, "said Lin Yi.