A couple of days ago, a good friend sent me the State of AI report. I started reading it today, and it is packed with valuable content. You can download the report from https://www.stateof.ai/. This report was created by AI investor Nathan Benaich and Air Street Capital.
of the "State of AI Report 2024" being released, covering several dimensions:
: Technological breakthroughs and their capabilities. : Business applications of AI and its impact on operations. : Regulation of AI, its economic impact, and the evolving geopolitical landscape. : Identifying and mitigating catastrophic risks that future high-capability AI systems may pose to us. : Looking ahead to the future and conducting performance evaluations to ensure the accuracy of predictions.
Today, I will first share a summary of their report. Since the entire report is quite lengthy, starting tomorrow, I will select some valuable sections for in-depth interpretation.
The main conclusions of the 2024 report are as follows:
: The gap between GPT-4 and other models is narrowing. OpenAI's GPT-4 once helped the lab regain the top spot on the leaderboard, but how long this lead will last remains uncertain.
: Companies are exploring ways to combine large language models (LLMs) with reinforcement learning, evolutionary algorithms, and self-improvement to unlock future autonomous intelligent applications.
: These models have demonstrated potential in multi-modal research, spanning mathematics, biology, genomics, physical sciences, and neuroscience.
: Despite U.S. sanctions, Chinese labs continue to build high-performance ultra-large language models (VLLMs) through stockpiles, approved hardware, smuggling, and cloud access. Meanwhile, China’s efforts to develop its domestic semiconductor industry remain chaotic.
: Publicly listed companies experienced a bull market due to the AI craze, and investments in private AI companies also increased, though to a much smaller extent compared to public markets. However, there were some significant financings for generative AI companies in the U.S.
: Including foundation model-building companies and startups focused on video and audio generation. However, questions about long-term sustainability remain unresolved as model costs decrease and competition intensifies.
: Some AI companies, struggling to find sustainable business models and facing high costs in competing at the cutting edge, opt for pseudo-acquisitions as an exit strategy. My understanding is mainly Acqui-hiring—returning 💰 to investors and acquiring the team.
: Especially after internal turmoil at OpenAI. Nevertheless, researchers continue to delve into potential vulnerabilities and misuse of models, proposing possible fixes and safety measures. (Last month, I heard some gossip about OpenAI in Silicon Valley, but it was all very peripheral and second-hand information.)
Starting tomorrow, I will share more key content.