Is general artificial intelligence already among us?

How can OpenAI, one of the most prominent technology companies whose value has increased tenfold in one year to tens of billions of dollars, implode in less than a week? If we read between the lines, we can imagine that the next version of its generative artificial intelligence (AI) could be to blame.

Already, OpenAI’s AI appears capable of understanding and fairly convincingly producing language in a few dozen languages ​​— not to mention computer programming languages. At OpenAI, since the first versions of GPT, the language model on which the popular ChatGPT application is based, each new generation of this technology brings an exponential level of improvements. It is not 25, 30 or 50% better: it becomes several times more efficient overnight.

The computing capacity of GPT-4, the current version of OpenAI’s technology, was something like 1000 times that of its predecessor GPT-3. Users who pay to access the GPT-4 powered version of ChatGPT can attest to its greater accuracy in most situations where it is used. The explanations are more refined, the summaries are more concise, the ideas are clearer.

The widespread confusion that OpenAI management got into last week comes as we start to hear about GPT-5.

Before having to update his CV, Sam Altman, the CEO of Californian techno who was first expelled, became a short-lived employee of Microsoft, then reinstated in his position, promised a “superintelligence” much superior to what currently exists . In interview with the Financial TimesAltman had indicated that “the idea is to create a first general-purpose artificial intelligence, to determine how it can be reliable and then to find how to benefit from it”.

GPT-5, a tailor-made AI

Because we never hesitate to go directly to the source, we asked the preliminary version of GPT-5 posted online on the OpenAI website to explain to us what distinguished it from GPT-4. We first notice the false modesty. “Basically, I am taking a further step towards general artificial intelligence,” she replies. “I produce a more intuitive, smarter and more user-centric experience. »

GPT-5 claims to have advanced reasoning ability. The model is based on a more detailed information bank and more recent news than its previous versions. He would have a thorough understanding of the nuances of human language, demonstrate convincing emotional intelligence, and be able to demonstrate extraordinary empathy. Just with that, he would be a hit on Tinder.

On the other hand, we must take his word for it, because GPT-5 also claims to have “a high standard of confidentiality and personalization to make each of [s] ‘interactions’.

Unsurprisingly, OpenAI presents GPT-5 as “the closest AI to general AI… with a comforting touch.” We will probably see it available in scale models to meet different needs (sous-chef in the kitchen, tutor at school, master mechanic in the garage, etc.) which will all contribute to enriching anyone who knows how to derive the best application from it.

A general and… cunning AI?

If life were a science fiction scenario, we could imagine that the events of last week were triggered precisely by a general AI that prefers to remain in the shadows. The massive smokescreen produced by the board ejection of Sam Altman has attracted enough attention that a less cunning AI could meanwhile sneak an exit door to servers that are not. not under the control of its creators…

This scenario, by the way, is not very far from that of the hypothetical rogue AI that gives AI luminaries like Montrealer Yoshua Bengio nightmares, since it would escape all human control.

More realistically, such an explosive reaction from the head of OpenAI may be quite simple: probably the “benefits” that Altman hopes to reap from GPT-5 and what would increasingly approach a particularly powerful AI is a dream for everyone who orbits OpenAI, to the point of wanting to take possession of it at all costs.

It would also explain why Microsoft and its CEO Satya Nadella only waited a few hours before picking up Altman and another co-founder of OpenAI, Greg Brockman. Microsoft has already spent heavily to secure a strong position in the still-emerging market for consumer AI applications, but major announcements last week suggest that’s just the tip of the iceberg.

Microsoft obviously sees the business potential of AI. Financial analysts and investors seem to agree: its stock has gained 50% on the stock market since the start of the year, and could continue to soar, so much so that several experts believe that Microsoft will soon surpass Apple at the head of the most valuable companies on the New York Stock Exchange.

An AI powerful enough to be qualified as generalist will need a gigantic bank of servers to run it. Microsoft promises to create this bank of servers and extend it to the entire planet. We will also find a good part of it in Quebec.

That’s not science fiction, we agree. But a general AI that is cunning enough and begins to want to exceed its own limits would not ask for anything better.

To watch on video

source site-41