Despite several warnings by some of the most prominent names in AI, people considered the fathers of AI and tech companies are powering on and are in a race to achieve AGI or artificial general intelligence.
In a new report published by Stanford University, a team of researchers they concluded that some AI companies may have already taken the first step to AGI and have a working model that can perform at the human level.
According to the ‘AI Index 2023’ report by the Stanford Institute for Human-Centered Artificial Intelligence, 2023 saw a remarkable increase in the release of foundation models, a total of 149, which indicates that models have more than doubled compared to the previous year.
Among these models, over 65.7 percent were open-source, a significant rise from 2022 and 2021.
Artificial intelligence (AI) is witnessing a surge in open-source initiatives, with more than 65.7 percent of newly developed large language models (LLMs) being made accessible in 2023.
Despite this trend, a recent study conducted by the Stanford Institute suggests that closed-source models outperform their open counterparts across various benchmarks.
However, the study found that closed-source models exhibited a median performance advantage of 24.2 percent over open-source models across ten selected benchmarks.
This indicates that while open-source models are growing in number, they could be delivering better performance.
Leading the charge in AI development, Google emerged as the frontrunner by releasing 40 foundational models, followed closely by OpenAI with 20 models.
Among the costly endeavors of certain tech companies, Google’s Gemini Ultra topped the list with a training cost of $191 million, followed by OpenAI’s GPT-4, which cost the company $78 million.
Regarding global AI development, the United States developed 61 models in 2023, followed by China with 15 models and France with eight. Additionally, the US attracted the lion’s share of investments in AI, totaling $67.2 billion, significantly surpassing China’s investment by 8.7 times. India secured tenth place globally, with $1.39 billion invested in AI initiatives.
While AI has made significant strides, achieving human-level performance on various benchmarks such as testing, reading, comprehension, and visual reasoning, it still needs to improve in areas like competition-level math, where it falls just short of human capabilities.
Overall, the landscape of AI development is evolving rapidly, with open-source initiatives driving accessibility but closed models demonstrating superior performance across critical benchmarks.
