AI is changing faster than most people realize

Perhaps it’s still too soon to be able to explain in a reasonably accessible way the transition underway in AI; it may be a new area for many people, but it’s undoubtedly one that will play an increasingly important in our collective future.

The first stage after the appearance of the mythical paper that gave rise to transformers and generative AI saw the development of large language models, an era that for the general public — as opposed to researchers — began with the launch of Dall·E by OpenAI, then ChatGPT, almost exactly two years ago; a game changer and the fastest technological adoption phenomenon in history.

The launch of these models generated both expectation and competition. Other companies, some long-standing, others newer, immediately launched their own products. Microsoft, thanks to its privileged relationship with OpenAI Azure, put Copilot out there, while Google attempted to get in the game with first Bard and then Gemini. Among the newer entrants were Anthropic’s Claude, Perplexity, Meta’s Llama, and a whole ecosystem that includes Mistral’s open-weight model, as well as several Chinese competitors. Using a chatbot as an interface is attractive, reasonably simple and highly user-friendly.

All those companies were working on the basis of scaling laws, which describe how the performance of a neural…