This webinar covers the generative-AI developments that defined 2023, along with their economic and legal consequences.
The year opened with Microsoft's 10-billion-dollar investment in OpenAI. Microsoft integrated OpenAI's technology across its product line and became the exclusive cloud provider for OpenAI's APIs.
GPT-4 launched in March and outperformed its predecessor across nearly all benchmarks, scoring notably on standardized tests like the GRE and the US bar exam.
Meta accidentally released its LLaMA model. What looked like a disaster became a catalyst: researchers worldwide used LLaMA as a base for their own work, producing open-source alternatives like Stable Vicuna. By year-end, the French startup Mistral AI shipped a model with a novel architecture that matched much larger systems in efficiency.
ElevenLabs pushed speech synthesis to the point where it could convincingly replicate nearly any voice. Midjourney's image generation reached a quality level that won photography competitions.
Goldman Sachs estimated that GPT-4-class systems could affect 300 million jobs. The New York Times sued OpenAI for using copyrighted content as training data, a case with potentially broad implications for future model development.
ETH and EPFL launched the Swiss AI Initiative, combining vertical projects in health and education with horizontal basic-research programs. Switzerland also secured NVIDIA GPUs early for the Lugano supercomputer, at a time when prices were still moderate, giving it a head start in European AI compute.
I expected the pace of breakthroughs to slow as training data limits were reached (GPT-4 had already consumed most available data). Progress would shift to improved multimodality, more efficient architectures, and specialized industry applications. The gap between proprietary and open-source models would continue to narrow.
The most important developments were the release of GPT-4, the Microsoft-OpenAI partnership, the LLaMA leak and its impact on the open-source community, and breakthroughs in multimodal AI such as ElevenLabs and Midjourney.
Meta's accidental LLaMA leak led to a democratization of AI development. This enabled the creation of open-source alternatives like Stable Vicuna and later Mistral AI, which set new standards in efficiency and architecture.
Switzerland positioned itself strategically in 2023 through the Swiss AI Initiative by ETH and EPFL and the early procurement of NVIDIA GPUs for the supercomputer in Lugano, giving it an edge in European AI research.
For 2024, advances are expected in three areas: improved multimodality (integration of text, image, audio, and video), more efficient models with equal performance using fewer resources, and specialized applications for specific industries.
Goldman Sachs projected that AI systems like GPT-4 could potentially affect 300 million jobs. Additionally, there were significant legal developments, such as the New York Times lawsuit against OpenAI over copyright infringement.
.
Copyright 2026 - Joel P. Barmettler