ChatGPT CEO criticizes AI that simulates friendship with people

Sam Altman, the CEO behind ChatGPT, said he is very skeptical of AI models being produced to show empathy for humans. This is the case of Meta AI, owned by Facebook, and the startup Character.AI, co-founded by Brazilian Daniel de Freitas, the former director of Google.

These two technologies promise to put well-known personalities into online conversations with users. Character.AI’s portfolio includes Taylor Swift, Napoleon Bonaparte, Tony Stark and the fallen angel Lucifer. Meta, in turn, has characters inspired by Gisele Bündchen’s ex-husband, Tom Brady (also a former American football player), rapper Snoop Dogg, and model and influencer Kendall Jenner.

“I personally am very skeptical of this vision of the future in which people are very close to friends made by artificial intelligence and not so many human friends,” Altman said on Tuesday (17) at a technology event promoted by the American Wall Street Journal. magazine.

However, it recognized the right of companies to operate in this area. “On the one hand, progress in personalization using AI is positive, but it is important to distinguish between interactions with humans and machines.”

“We were very intentional in naming our product ChatGPT and not after a person. OpenAI makes it clear that the client is not talking to a human,” he added. GPT is an English abbreviation for “Generative Pre-Trained Transformer,” which is the technical name given to the model behind a text generator bot.

In the case of the model from Facebook owner Meta, chats are named after characters, in the form of a famous person. Snoop Dogg, for example, inspires the program Dungeon Master, which talks to the user about the medieval world and makes up rhymes in the style of a rapper.

Facebook founder Mark Zuckerberg introduced this character at the end of September at a Meta event. The relaxed tone set the conversation between man and machine.

The model is still in the testing phase in the United States and is not available in Brazil. “At the moment there is no information about expansion into other markets,” Meta tells the report.

It is also scheduled to take place at a Wall Street Journal event, where the CEO in charge of Meta AI, Chris Cox, responded that their technology has a neutral name. His presentation came shortly after a speech by the CEO of OpenAI, the creator of ChatGPT. “But I think there’s more room for fun and expression.”

Chatbots will be the future of entertainment, according to Character.AI co-founder and CEO Noam Shazier. He left Google with Brazilian CEO Daniel de Freitas to found the startup, which is currently worth R$25 billion, according to Bloomberg estimates.

Since the launch of the beta version in September 2022, more than 20 million people have registered on the platform. Forbes magazine tested it and said the technology is still primitive.

However, Daniel de Freitas states that Character.AI’s characters are enough to distract the startup’s employees from work. Freitas encourages his employees to play Ship AI, a text-based space adventure game narrated by the company’s chatbot.

Today, new entertainment markets such as social networks and video games generate much more money than traditional sectors such as music, cinema, theater and literature. This is evidence of the power of these products to capture consumer attention, attracting advertisers, buyers and sponsors.

Other chatbots competing with ChatGPT, such as Pi and Claude, have also been trained to show more elements of human conversation in their responses, even if they are not entertainment-oriented. Pi, for example, tends to provoke his interlocutor with questions, in a Socratic rhetorical strategy.

With less investment and at a more rudimentary stage of development, other platforms promise embodied AI friends. The experience reminds us of the fictional narration of the film “Her” (2013) by director Spike Jonze.

For example, the AI ​​chatbot Replika was able to act out rousing speeches until June. When the company responsible for the technology blocked these responses, users complained that the artificial intelligence was “fragmented,” according to the specialized website Business Insider.

People tend to see the humanity in AI platforms through the fluidity of word sequences provided by the technology. This process is called the Eliza effect, in honor of the first chatbot created in the 1960s.

However, AI models like ChatGPT act like a statistical parrot: they are trained to guess the next most likely word to continue a block of text.

“They are able to reproduce the language they have been trained in, but they are not thinking about what we are asking them to do,” Patty Mays, a professor at the Massachusetts Institute of Technology (MIT) said in an interview with Folha.