ChatGPT's Popularity Ignited a Fire... AI Evolution Accelerates

ChatGPT Sparks Development Competition... MS Support Also Plays a Role
"AI Cost-Effectiveness Predictable, Development Accelerates"

From 18 months to 4 months.

The evolution speed of GPT, the massive AI model developed by the US artificial intelligence (AI) startup OpenAI, is accelerating. The success of the chatbot AI ChatGPT, based on GPT-3.5, has ignited competition in AI development. Microsoft's (MS) partnership with OpenAI to provide infrastructure support has also played a role.

OpenAI released GPT-1 in June 2018. GPT-2 came out in February 2019, and GPT-3 in May 2020. GPT-3.5, which powers ChatGPT, was released in November 2022, and the latest version, GPT-4, was unveiled on the 14th of this month. Compared to the 18 months it took to release GPT-3.5, the development period for GPT-4 was reduced to 4 months.

The number of parameters, which largely determine AI performance, has also evolved. GPT-1 started with 17 million parameters, while GPT-3 and 3.5 have 175 billion parameters, an increase of over 10,000 times. Although the number of parameters for GPT-4 has not been disclosed, it is estimated to be slightly larger than before. Beyond parameters, the overall model size, considering training data and computing power, has continued to grow.

The fact that GPT-4 was released in just 4 months is thanks to the success of ChatGPT. ChatGPT surpassed 1 million subscribers within five days of its launch. This is a record that Facebook took 10 months and Netflix 3 years to achieve. Microsoft's search engine 'Bing,' equipped with ChatGPT, has exceeded 100 million daily active users. This was just one month after MS introduced the new Bing. Notably, one-third of these 100 million users are first-time Bing users.

The explosive popularity of ChatGPT has changed the atmosphere in the IT industry. At OpenAI, the mood has clearly shifted from concerns about AI risks such as ethics to seizing the opportunity while the tide is high. Competitors like Google have also become busier.

MS's support for OpenAI is another factor accelerating the pace. MS provides the computing resources needed for OpenAI to develop AI models through its cloud. Thousands of NVIDIA graphics processing units (GPUs) are connected to MS's Azure cloud computing platform. MS has stated that it is spending hundreds of millions of dollars to build the supercomputer required to run ChatGPT. Separately, in January, MS decided to invest $10 billion (about 12 trillion won) in OpenAI.

Industry experts expect AI development speed to accelerate further with the release of GPT-4. As development achievements have accumulated up to GPT-4, the 'cost-effectiveness' (performance relative to price) of AI development has become clear. Ha Jung-woo, head of Naver Cloud AI Lab, explained, "In the past, the performance achievable relative to computing resource input was inconsistent, but from GPT-4 onward, it has become somewhat predictable," adding, "From a corporate perspective, this means the effects relative to investment have become clear."