2020-2021: language models change scale
In 2020, GPT-3 showed that a model trained on very large volumes of text can write, summarise or translate from very few examples. In 2021, text-to-image generation (DALL·E) and code assistance (GitHub Copilot) widened the range of uses. That same year, the European Commission proposed a risk-based regulation on AI.
2022: image generation goes public
DALL·E 2, Midjourney and Stable Diffusion made assisted visual creation available to everyone. Questions about copyright, training data and disinformation entered the public debate.
November 2022: ChatGPT
The launch of ChatGPT, at the end of November 2022, put a conversational assistant in the hands of the general public for the first time. It reached a very wide audience within weeks and set off an innovation race among major players and labs.
2023: the year of adoption
GPT-4, open models such as LLaMA and the first assistants built into office tools brought generative AI into companies. Organisations discovered both productivity gains and risks: hallucinations, data confidentiality, bias.
Key takeaways
- The leap comes from model scale and interface quality, not from a sudden research breakthrough.
- Uses come before rules: governance and training become leadership topics.
- Understanding a tool’s limits matters as much as knowing how to use it.
More in part 2: 2023-2025.