DEGGENDORF, GERMANY - 2022 has been an exciting year in the field of artificial intelligence (AI). Key developments include novel models and applications, the widening China-United States AI research gap and increased political pressure to regulate AI. Some of the major developments have been summarized in the State of AI Report 2022.¹ How will AI develop in the coming year? This article provides various predictions of likely AI developments in 2023.

GPT-4

The Generative Pre-trained Transformer 3 (GPT-3) released in mid-2020 by OpenAI is a high-end natural language processing AI based on a transformer network. GPT-3 offers various functionalities, including text classification, question answering and text generation, and is available in different sizes, while the largest one has about 175 billion parameters. In the last two and a half years, a number of competitors have been released.² One of these is Wu Dao 2.0, which has 1.75 trillion parameters and has shown extremely promising results.³ In recent months, the release of the next version, GPT-4, has been increasingly rumored for early 2023, and could possibly have up to 100 trillion parameters.⁴ Given the increased pressure on OpenAI to soon release a novel competitive model after more than two and a half years since the last one, it seems very likely that GPT-4 will appear in early 2023. The new model will also likely be applicable to computer vision tasks to solve tasks similar to the recently popularized DALL·E 2 and Stable Diffusion models. OpenAI will, however, thoroughly test GPT-4 before its release and not rush into releasing it. Part of the motivation for this is recent news about Meta’s much-hyped Galactica model, which was supposed to help ‘organize science.’ Instead, it quickly started to spread misinformation and was subsequently turned off after 48 hou