Speaking to the students ofStanford Universitya few days ago, Sam Altman said that “ChatGpt is not phenomenal”. Indeed, the OpenAI chatbot – according to Altman – “It’s embarrassing”.
Criticisms of the most famous generative artificial intelligence in the world, used every month by more than one hundred million userscome from the very man who represents it in the world.
Altman co-founded OpenAI in 2015 and since 2019, except for the brief period in which he was surprisingly ousted from the board, he has held the role of CEO of the company. Yet Altman held back harsh words to the technology developed by OpenAI: “Gpt-4 is by far the stupidest model you will ever find yourself using” he said during the talk he was the protagonist of at Stanford.
Artificial intelligence The creators of ChatGpt: “We didn’t think it would work” by Pier Luigi Pisa 02 November 2023
It is striking that Altman has discredited it a cornerstone of generative AIcapable of writing and producing images like a human being would.
In fact, Gpt-4 is the most advanced AI model from the San Francisco company. It was only introduced a year ago and is based on approx 1.76 trillion parameters. The parameters of an AI are, in short, the value that broadly describes the complexity and performance of a model.
It is true that ChatGpt has been accused several times in recent months that I have become particularly “lazy”. OpenAI itself recognized, at a certain point, a deterioration of the Gpt-4 model, writing on the social network X that it had combated the problem encountered by users.
But it is also true that Gpt-4 – still today represents the highest point of reference for the big tech companies that develop LLM, i.e. the AI models on which tools such as ChatGpt are based, precisely, but also Gemini (Google) e Claude (Anthropic).
So what interest does Sam Altman have in saying that his “creature” isn’t all that great?
There could be at least two reasons.
The first is what fans of OpenAI, ChatGpt and more generally of generative AI hope for: the company led by Altman is about to launch a new model of artificial intelligence, the highly anticipated Gpt-5, which will far surpass any previously developed AI. And it will even make Gpt-4 pale to the point of making this technology “embarrassing”.
The second reason, however, would have nothing to do with Gpt-5.
By shooting at Gpt-4, Altman may have put into practice one of the marketing operations that OpenAI, in the opinion of many, succeed particularly well. Like when, a few months ago, someone from the company leaked the existence of a phantom “Project Q”, that is, an artificial intelligence so advanced that it constitutes a possible threat to humanity. No one has ever been able to ascertain whether this AI really exists. But in the meantime the hype around OpenAI’s technology has grown further.
The case The discovery of OpenAI: what is the Q* project that “puts humanity at risk” by Pier Luigi Pisa 23 November 2023
Today as then, therefore, the real objective of the CEO of OpenAI would not have been to impress Stanford students or AI enthusiasts, rather throw bait to investors.
We don’t know if OpenAI will really be able to launch Gpt-5 by summeras many expect, but in the meantime Altman – as a person informed of the facts – suggests that his company can take an important step forward.
After all, OpenAI has no shortage of means and data to do something great.
Microsoft – OpenAI’s main investor – is ready to spend $100 billion in a supercomputer network which will give a huge boost to the development of AI in the coming years.
The indiscretion OpenAI and Microsoft are thinking of a 100 billion supercomputer for AI by Pier Luigi Pisa 29 March 2024
Altman’s company will combine computing power with the other fundamental ingredient for putting together an LLM: huge databases, this time coming from the most important media in the world. Several newspapers, from Financial Times a The world, have signed million-dollar deals to license their archives to OpenAI. They will end up being fed to ChatGpt, which will thus become even more “intelligent”.
Perhaps able, one day, to think like a human being.
We are talking about breaking latest news, theartificial general intelligence that OpenAI has been pursuing since its foundation. There is no univocal definition of breaking latest news, but for many it is the AI that will one day develop cognitive abilities similar to or superior to those of peopleand who like them will be able to face problems and make decisions.
At Stanford, Altman also expressed his opinion on this. He said OpenAI intends achieve breaking latest news at all costs.
“I don’t care if we burn 500 million dollars a year doing it,” Altman said. I don’t even care if it’s 5 billion a year. Or 50. I don’t care as long as we’re on a trajectory that will ultimately allow us to create something of value for society.”
“And as long as we have money to pay the checks – added Altman – it will be worth it. Even if it will be expensive.”
But the price, as Altman himself has been telling us for some time, may not have anything to do with it only with money.
“One thing that needs to be kept in mind – said the CEO of OpenAI at Stanford – is that we cannot pretend that this technology, like all technologies, is only good. I believe it will be for tremendous good, but I also believe that someone could misuse it. It’s like a hammer: you can do great things with it or you can use it to kill someone.”
This is why, according to Altman, big tech companies should release their new models in the future in a more “responsible” way, that is, gradually and through small steps, avoiding large leaps forward. To give users – and more generally to society – the opportunity to metabolise AI.
And therefore it would be better for OpenAI, at this point, to release an intermediate model, Gpt-4.5 for examplee non Gpt-5.
Who knows, it could actually go like this in the coming months.