Artificial intelligence: what Sam Altman thinks about hallucinations, ChatGPT’s biggest flaw

Sam AltmanCEO of OpenAI and main reference of the boom generated in recent months around the generative artificial intelligenceresponded to the greatest criticism that language models such as ChatGPT: the “hallucinations”.

In conversation Marc Benioff -executive director of Salesforce– during the event Dreamforce 2023Altman discussed some key points about the current state of artificial intelligence and provided some predictions regarding its future evolution.

Questioned about “hallucinations” – information presented coherently as if they were real events, but with incorrect, biased or outright false data-, Altman stated that they are not a problem, but part of the “magic” of these AI systems.

“Much of the value of these systems is strongly related to the fact that they hallucinate. They are more of a feature (characteristic) that a bug (mistake). If you want to search something in a database, there are already useful things to do it“explained the CEO of OpenAI.

And he added, in that sense: “The fact that these artificial intelligence systems can generate new ideas and be creative It’s a big part of his power.. We are working on being creative when we want and objective when we want.”

For Altman, ensuring that these new AI platforms only generate content when they are absolutely sure it is correct would be “naive” and “contrary to its fundamental nature”.

“If you just decide to do something naive and tell the AI ​​’never say anything you’re not 100% sure about,’ you can get everyone to do the same thing. But it won’t have the magic that people love so much.”he added.

In artificial intelligence models, hallucinations occur when a chatbot provides false informationin many cases easily verifiable, as if it were objectively correct.

The topic is being widely discussed in the environment tech for a year, and even led to OpenAI being sued for ChatGPT’s false claims. Googletherefore, was ridiculed during the world presentation of bardtheir own chatbot, by incorrectly answering a simple question.

Critics and industry leaders They also expressed concern in recent weeks. In fact, a few hours before his talk with Altman, when delivering his keynote speech at dream force -the self-proclaimed “world’s largest AI event”- Benioff was more than eloquent: “I don’t call them hallucinations, I call them lies”.

In another part of his conversation with Benioff, Altman revealed what his favorite science fiction movie is. And, to the surprise of few, the plot has to do with artificial intelligence.

“I like Her (“She”). The things the movie got right, like all the interaction models and how people use AI, They were incredibly prophetic.“Altman said.

Her, released in 2013, directed by Spike Jonze, stars Joaquin Phoenixa heartbroken man going through a divorce and falls in love with an artificial intelligence virtual assistant named Samantha, with the voice of Scarlett Johansson.

Although he is optimistic about the future of his technology, the CEO of OpenAI acknowledged that artificial intelligence will not save you in a crisis from the real world.

And he sentenced: “I’m under no illusions that any of this is going to protect us or anyone else if AI goes wrong.”