Listen to the audio version of the article
«The biggest surprise was realizing that ChatGpt really worked». He said it so seriously that he got a real laugh out of the audience at Dreamforce in San Francisco. The founder of OpenAi Sam Altman gave himself a long interview with Marc Benioff, CEO of Salesforce, on the opening day of what was defined as the largest event dedicated to artificial intelligence. Altman spoke about himself, explaining that the hallucinations and errors that his chatbot makes are not necessarily errors but in the future they could become a new way of seeing things. Appearing particularly prophetic and cryptic, the young entrepreneur however reiterated that in the short term the objective remains that of “having the most intelligent, most capable and most customizable models on the market”. Answering a question about his favorite science fiction film, he mentioned “Her”, the 2013 Spike Jonze film starring Joaquin Phoenix in love with a form of artificial intelligence. «Inside the film there are many intuitions – he explained – on models of human-machine interaction that turned out to be incredibly prophetic».
Altman speaks in a low voice giving the impression of measuring every single word. Compared to a long-standing visionary like Marc Benioff who asked concrete questions, the father of ChatGpt wanted to bring the discussion to the highest levels, avoiding precise answers. Until the moment we talked about the AI’s mistakes.
“One of the things that isn’t obvious is that much of the value of these systems is strongly related to the fact that they hallucinate,” Altman underlined. If you want to search for something in a database, we already have useful material for that.”
It is precisely on the hallucinations, and therefore on the errors, that these large linguistic models such as ChatGpt and Bard owe perhaps one of the most striking lines of the interview. If we move away from the logic of punctuality, let’s say quantitative of the response, “the hallucinations – he said – are more a feature than a bug”. In other words, when we talk about generative artificial intelligence, let’s not expect to have correct answers. These chatbots take existing information and present it in a different and new way based on the user’s request and creating a different perspective. AI, Altman implied, is not a calculator. What are considered errors today could tomorrow become suggestions or the starting point for new theories. As said before, if the Ai ever wanted to look for a prophet, today they found one.