ChatGPT, the frontier of artificial intelligence between potential and dark side
Economy & innovation — di Pietro Minto
Trained to answer any question: ChatGPT is a new conversational model based on AI. The risk is that it produces disinformation.

Trained to answer any question: ChatGPT is a new conversational model based on AI. The risk is that it produces disinformation.
This time it's for real, this is not a drill. Artificial intelligence has arrived. In recent weeks, more and more people have discovered the potential of the GPT-3 model, which underpins systems such as Dall-E and ChatGPT — services produced by the startup OpenAI. And then Stable Diffusion and MidJourney AI, more accessible and open to the public, whose ability to generate images based on written descriptions has astonished many.
ChatGPT is a special case. The moment it was released to the public has been described as "the iPhone moment for the AI sector" — the event through which the mass public becomes aware of what the future holds for it, according to Bloomberg. But is that really the case? As the name suggests, ChatGPT presents itself to the user as a simple chat interface through which one can interact with GPT-3, a language model that has existed since 2020 and has been adapted for conversational use. For years, OpenAI "trained" GPT-3 on vast archives of text drawn from the web, Wikipedia, newspapers, magazines and books, as well as from social media, in order to teach the machine how humans communicate. To do so, it was programmed to break down every text and sentence into basic units of text (known as "tokens"), which are then analysed individually.

Creating a model as functional and astonishing as the ones in question requires computational power, however. A great deal of it. According to data cited by the podcast of The Verge, OpenAI burns through three million dollars a day responding to user requests. All those screenshots we've seen in recent days, in which users ask ChatGPT the most absurd things? They've cost the company a fortune — a company that has long been receiving (and spending rather quickly) investment from leading Silicon Valley venture capitalists and entrepreneurs, Reed Hastings of Netflix, the ever-present Elon Musk and Microsoft above all. Costs aside, the results have helped cement AI as the next big thing, after a couple of years spent chasing the ghosts of crypto. The potential applications of this kind of technology are many, and it is still too early to venture predictions: some swear that a service like this could replace Google and search engines, for example. The idea goes like this: since ChatGPT can answer any question, why should we use Google in the future the way we have for the past twenty years? The answer seems clear — until, for a moment, we set aside the fact that similar AIs often give bizarre and incorrect answers, while maintaining a very authoritative tone and style. ChatGPT's function is to generate credible text, not to become an oracle to which one entrusts one's life. For that, I think, we'll have to wait for GPT-4 — who knows.

In recent days you have likely read or seen content that was written or created by an artificial intelligence. In most cases, this was done by one of OpenAI's services — a startup founded in 2015 by Sam Altman and others, with investment from Netflix founder Reed Hastings to Elon Musk, as well as Microsoft. At the centre of recent discussions is ChatGPT, an AI that works precisely like a chat interface to which you can put any kind of question and receive consistently plausible answers. The chat can compose poetry, imitate writers and respond to creative questions. It follows instructions, and does so well. This also means it can produce disinformation with extreme speed and effectiveness; you need only ask it to. OpenAI has of course placed restrictions on the topics the chat will address, but circumventing them is not difficult. The potential the technology holds for propaganda and fake news is enormous — something CEO Altman himself has acknowledged, admitting that ChatGPT is capable of producing "content that can give a misleading impression of quality." Imagine a button for generating conspiracy theories and disinformation, in bulk, in a matter of seconds. Well, we're there now. ChatGPT is a defining moment for the AI sector, of which it represents the dark side.