Around 1770, Hungarian engineer Wolfgang von Kempelen’s mechanical Turk, an automaton in a cape and silk turban, stunned the world intelligentsia with his chess brilliance. Its battery of cogs fooled the curious, while a man, cleverly concealed, disputed the games on the sly.
Posted a few weeks ago by the start-up OpenAI, ChatGPT-3 is not a trick, but the illusion is nonetheless comparable. He convincingly simulates a human conversation and can write, tit for tat, relevant content of an incredible diversity: summary note, editorial, dissertation or even lines of computer code.
But behind the screen, it is still and always the human who holds his pen. Imagine a colossal calculator, handling words and images rather than numbers, having been trained for a long time by instructors and constantly corrected, reconfigured, guided and formatted, even if we speak of “automatic learning”. Its overpowering and sprawling memory feeds on all of our writings and conversations, taken from the Net, and recomposes them, like a dizzying kaleidoscope.
However, despite its phenomenal capabilities, ChatGPT is not a “brain”, a big head. Beware of metaphors: its cogs are called “artificial neurons” because they simulate the electrical activity of ours, but that’s where the comparison ends. Nothing to do with our gray cells, it’s all about
This article is for subscribers only. Take advantage of our special offers without obligation.
WHERE
Google Exclusive: 6.99€/month
By choosing this promotional subscription path, you accept the deposit of an analysis cookie by Google.
