"If you take together the invention of electricity, computers, the internet and mobile, you fall short of what we are going to see," said Siqi Chen, a San Francisco-based entrepreneur.

"All these things were created by intelligence. But for the first time, we are able to create intelligence itself," he continues. "It's a double-edged sword, but if it goes well, it can solve all the problems (...), like global warming."

Like many other tech players, Siqi Chen is convinced that he is witnessing a historic paradigm shift.

Especially since OpenAI's presentation on Tuesday of GPT-4, a new, even more powerful version of the natural language model that operates ChatGPT, the generative AI interface used by millions of people in recent months to write essays, poems or lines of computer code.

ChatGPT will be able to process not only text, but also images, and produce more complex content, such as legal complaints or video games.

GPT-4 thus represents a step forward in the direction of so-called "general" artificial intelligence, that of programs "smarter than humans in general", according to Sam Altman, the boss of the Californian start-up.


On Thursday, Microsoft, OpenAI's lead investor, vowed that "soon we won't be able to do without" generative AI-based assistants that can interact with humans in their languages and perform everything from meeting summaries to creating a website or ad campaign.

These tools will free humans "from the chores that stifle creativity" so they can reconnect with "the soul of their work," said Jared Spataro, an executive at the IT group.

"I used GPT-4 to code 5 microfeatures for a new product. A (very good) developer wanted $6,000 and two weeks. GPT-4 did it in 3 hours for $0.11. Mind-boggling," tweeted Joe Perkins, a British entrepreneur.

Siqi Chen acknowledges that new technology may replace him one day. But it relies on the ability of humans to adapt, with solutions like universal income.

OpenAI wants to build so-called "general" artificial intelligence, i.e. AI systems with human © cognitive capabilities Stefani Reynolds / AFP

Beyond the threat to the intellectual and artistic professions, general AI is generating insurmountable societal debates.

What will remain authentic, when the slightest photo on Instagram or the slightest review of a restaurant will have been produced with or by an AI? What will become of learning, when it will be enough to formulate requests to machines? Who should make the decisions to define the algorithms?


"General AI is coming faster than we are able to digest it," said Sharon Zhou, founder of a generative AI startup.

"This is going to pose existential questions to humanity. If it is more powerful and intelligent than us, do we exploit it? Or is it exploiting us?" asks the former Stanford University researcher.

OpenAI says it wants to build general AI gradually, with the goal of benefiting all humanity. It relies on the large-scale use of its models to detect and rectify problems.

But the company itself seems overwhelmed by events.

Greg Brockman, one of the co-founders, acknowledged in an interview with The Information that ChatGPT wasn't as value-neutral as they would have liked.

Ilya Sutskever, the chief scientist, would like "there to be a way to slow down the rate of release of these models with unprecedented capabilities," according to an interview with MIT Technology Review.

And the start-up, whose name means "open AI", is criticized for its lack of transparency. The release of GPT-4 marks "its transformation from a nonprofit research laboratory into a capitalist enterprise," said Will Douglas Heaven, an expert in the scientific journal.

But despite the criticisms, concerns, and real or fantasized risks, the industry remains convinced that mainstream AI is coming, inexorably.

Because the race between companies is on, explains Sharon Zhou, but also between countries, especially the United States and China.

"The power is in the hands of those who know how to build all this," she said. "And we can't stop, because we can't afford to lose."

© 2023 AFP