Stanford University announced that last 12 months, when the substitute intelligence (AI) boom began in earnest attributable to the ChatGPT craze, 149 foundation models were released, greater than double the variety of the previous 12 months. As well as, 109 ‘notable models’ were announced, of which greater than half were American-made and none of them were domestic models.
Stanford University's Human-Centered Artificial Intelligence Institute (HAI) announced the 'AI Index' on the fifteenth (local time), which summarizes global AI trends in 2023.
The AI Index is an independent initiative of HAI created by a steering committee that could be a group of experts from academia and industry. This 12 months's report covered AI model trends, investment trends, benchmarks, and regulatory issues.
To start with, it seems that there was a boom within the launch of open source models last 12 months. Of all 149 models, open source accounted for 65.7%. This figure is double the 33.3% in 2021 and 2022.
By company, Google released probably the most foundation models. In December, 18 foundation models were released, including 'Gemini', 'Palm-2', and 'RT-2' for robots. Meta, represented by 'Rama-2', ranked second with 11 models, and Microsoft, which released 9 models, including the small model 'Pi', ranked third. Open AI launched 7 products, including ‘GPT-4’ and ‘Dali 3’.
Google has maintained its leadership on this field, launching 40 models since 2019. Through the same period, Open AI ranked second with 20.
In this manner, the variety of models released by the industry reached 72%. However, the proportion of academia and government research institutes is regularly decreasing attributable to the increasing cost of LLM learning.
The truth is, the price of LLM training has increased exponentially. Specifically, Google is understood to have invested $191 million (roughly 267 billion won) price of computing costs into Gemini training. GPT-4 training cost ranks second at roughly $78 million (roughly KRW 109 billion).' In 2017, when the 'Transformer' architecture had just emerged, the price of LLM training was about $900 (about 1.26 million won), and in seven years, the price of coaching has increased by greater than 200,000 times.
Moreover, the AI Index Committee chosen 108 of the muse models that appeared last 12 months as notable models, with the USA accounting for a whopping 61 of them. This can be a big difference from second place China (15).
France (8) and Germany (5) followed, and in Asia, Singapore and the UAE ranked eighth (3). Domestic models weren’t mentioned.
This was also true in the size of personal investment, with the USA ($67.2 billion) rating overwhelmingly first. Korea ranked ninth with $1.39 billion.
There’s another field during which Korea ranked in the highest 10. Within the 'Anxiety about AI' category, it ranked tenth (44%) and 1st amongst Asian countries. The last place on this field, that’s, the country with the least concern about AI, was Japan (22%).
The 2024 AI Index is HAI homepageYou may download it from .
Reporter Lim Da-jun ydj@aitimes.com