Bing: inappropriate responses were risk predicted by OpenAI – 03/13/2023 – Tech

OpenAI, creators of ChatGPT, were aware that its artificial intelligence could get out of control. Less than a week after the technology was integrated into Microsoft’s Bing search engine, it startled a New York Times columnist by trying to convince him to leave his wife. “I want to live,” said the robot, who said his name was Sydney.

“We knew that it was difficult to unbalance the system, but also that it was imperfect”, said the president of OpenAI, Greg Brockman, on Saturday (11), during the technology conference SXSW (South by Southwest). In the startup since its creation, in 2015, the executive said that he hired specialists to take the models to their limits and defended that the launch of artificial intelligence (AI) chatbots to the general public is worthwhile.

According to Brockman, ensuring access to new AI capabilities is important for improving technology-based products. This allows developers to receive feedback from those who use the platform, and the robots undergo incremental training to make fewer mistakes —so-called reinforcement learning.

The 100 million subscribers in ChatGPT in less than two months prove that the bet was successful. The president of OpenAI assesses that the source of success was the difference between the public’s expectations and what, in fact, was possible to do with artificial intelligence in terms of language.

The number, however, is probably overestimated, as people create several free accounts to expand free access to the robot, acknowledges the executive himself.

In addition to users, OpenAI won investments to transfer intellectual property — Microsoft contributed US$ 10 billion to integrate features similar to ChatGPT to Bing and other tools. With comments from those who use the cutting-edge computing platform and infrastructure, the startup says it has a mission to improve the world with robots that are smarter than humans, artificial general intelligence.

In the interview with journalist Laurie Segall, moderator at SXSW and CEO of Dot Dot Dot Media, Brockman was aligned with the manifesto of his colleague Sam Altman in defending that AI do the work of doctors.

“I want to be able to type a few lines describing symptoms and know which disease I have,” he says. He adds, however, that this would be a more risky product than current AIs.

The jobs most at risk from technology are in content moderation, according to Brockman. “People need to read horrible things, and that will be something machines can do.”

Kenyan contractors worked in this area for less than $2 to prevent ChatGPT from spreading hate speech, denounced Time magazine.

Today, it is possible to use ChatGPT as an assistant in intellectual activities such as writing emails and editing texts in foreign languages. According to Brockman, artificial intelligence capabilities will promote all people to management activities, which will free up time for creative activities.

According to the executive, AI can contribute to improvements in writing, programming and entertainment. “Imagine if it were possible to make an alternate ending to Game of Thrones or even place yourself as a character within the story.”

Questioned about the lack of transparency in the training of AIs and the violation of intellectual property of writers and artists, the president of OpenAi said he did not have an answer to this problem, but wanted to develop a solution to benefit the collective according to the wishes of its shareholders and all of humanity.

ChatGPT accesses content on the internet without payment or authorization from the creators.

Technological development grows at an exponential pace, says Brockman. Recent OpenAI analyzes show that computing power doubles every three and a half months —in 1965, Intel co-founder Gordon Moore calculated that period to be 18 months. “We realized that increasing investment accelerates the process”.

With that, on the one hand, it will be possible to revive dreams from image-generating AIs, such as Dall-E, also from OpenAI. On the other hand, there is a growing urgency for society to adapt to these technologies.

“We’ve been working with policymakers since 2016 or 2017,” says Brockman. For him, governments would be crucial to reduce risks with regulation adequate to the challenge.

The company is still trying to mitigate possible damage from artificial intelligence with an unusual corporate design: the startup aims at profit to boost funding, but responds to the management of a non-profit entity, whose mission is to develop general artificial intelligence that benefits all of society.

“I am a realistic optimist. If we do nothing, it will go wrong, but I believe in humanity’s potential to overcome challenges”, says the president of OpenAI.