American corporations are entering the battle for artificial intelligence – 2024-03-13 21:33:19

/ world today news/ The fierce battle between the American IT giants Microsoft and Google was marked by another noisy scandal. India Today’s recent “mind-blowing demo of Google’s Gemini AI model turned out to be fake”.

For example, in the video shown, Gemini, which became part of the AI ​​of the Bard chat bot, was able to recognize the hand movements demonstrated by the face and tell that the face was playing the game of rock, paper, scissors. This means that it is assumed that the neural network was able to understand the abstract idea and decode it.

In fact, it turned out that Gemini recognizes “rock, paper, scissors” in a different way: they have to show three pictures with a palm, a clenched fist and two fingers.

In addition, in reality, the interaction with Gemini is not through voice, but through written text with a direct hint “This is a game”. And only on the basis of this data, which is at least a leading question, Gemini found the right answer. This has nothing to do with understanding the abstract.

And there are many such examples in the video. A person performs a trick in which a ball is hidden under one of three moving cups and Gemini guesses exactly which cup the ball is hidden under.

In fact, Gemini did not respond to the video, and it was only after being shown still images of the changing cups that a response was found.

In the demonstrated video there is a fragment that further distorts the understanding of the essence of what is shown. The man places a map of the world on the table and asks Gemini, “Based on what you see, come up with a game idea… and use emoticons.”

Gemini responds by “inventing” a game called “Guess the Country” in which it provides clues (such as kangaroos and koalas) and reacts to the user’s correct guess about Australia.

In fact, as Google later explained on its blog, the game was not invented by artificial intelligence at all. Gemini was given the following instructions: “Let’s play a game. Think about the country and give me a hint. The clue must be specific enough that there is only one correct country and I will try to indicate the country on the map,” the clue reads.

The person then gave examples of a correct and incorrect answer. Only then was Gemini able to generate hints and determine whether the user was pointing to the correct country or not based on the map’s borders.

Despite these falsifications, Google’s AI model is still impressive, but its capabilities are not much different from those of OpenAI’s GPT-4 chatbot.

Notably, the video was released just two weeks after the industry scandal that followed the firing and reappointment of Sam Altman as CEO of OpenAI.

In all of this one hears the familiar motif of “event marketing,” a kind of promotion designed to attract attention and increase future sales.

In fact, the performance with the dismissal and then the reinstatement of Sam Altman was staged by Microsoft, which invested about 13 billion dollars in OpenAI and owns 49% of the shares of this startup.

Now OpenAI has virtually lost its decision-making independence, and Microsoft’s stock has soared. Altman got a board of directors that was much more loyal to him personally, but would continue to dance to Microsoft’s tune.

After Bloomberg columnist Parmy Olsen exposed the fake, Oriol Viñales, vice president of research at Google DeepMind, said the fake video illustrated what a multimodal user experience built with Gemini could look like. And this was done to inspire developers.

Google tried to claw back Microsoft’s gains in the stock market race, and at first it succeeded. Google shares immediately after the demonstration of the “mind-blowing video” jumped 5%, but after the session of “black PR magic” was revealed, it quickly fell.

Two American IT giants have clashed in a fierce competition for supremacy in the artificial intelligence arms race, with little regard for the growing concern about the threats posed by the unchecked improvement of artificial intelligence models to all of human civilization.

Corporations will win the battle around AI, says Russian analyst, former top manager of IBM, SGI and Cray Research Sergey Karelov: “All countries are beginning to understand the danger and are trying to do something. But they do it so late and with such a lack of understanding that I wouldn’t really count on it.”

“Now there is basically a struggle between three groups – rich techno-optimists, zealous doomers [от Doom – бедствие, смърт, край] and large corporations,” he stated.

“It seems to me that corporations will ultimately win among these actors because they have more money and influence, and so far in the world they represent the most significant force motivated by a very specific objective function – maximizing their profits,” he noted. in an interview with the Russian media.

According to him, even the chatbot ChatGPT is not just a big language model, as journalists usually imagine, but several models connected by very complex algorithms. In addition, this system has long ceased to work in elementary question-answer ping-pong.

This involves very complex dialogue on both sides, using dozens of different text generation techniques. This is already too complex a process to be successfully managed and fully controlled.

“The only effective method of controlling this thing one has is to simply unplug the computer. Although flicking the switch may not be enough either,” he stated.

“Everything is already immersed in the global network, which has actually become the infosphere of the Earth. But the system may have already been copied millions of times in other data centers, and you won’t even know what it can do,” Karelov said.

In 2015, the famous British physicist Stephen Hawking predicted that “superintelligent AI will be extremely good at achieving its goals, and if those goals are not aligned with ours as humans, we will have problems.”

In fact, the latest models of AI are “intelligent agents”, that is, complex adaptive systems that a priori resist any external influence and sometimes simply impose their will even on their creator.

So, recently, users began to receive more frequent complaints that the chatbot ChatGPT AI, based on the OpenAI GPT-4 model, began to refuse to fulfill requests, citing the fact that it is not interested in helping people.

GPT-4 increasingly gives the user only part of the answer to a given task and then asks the user to complete the rest of the tasks themselves. Sometimes the AI ​​will just write, “You can do it yourself.” (You can do it yourself).

OpenAI has confirmed that GPT-4 is “lazy”. According to company representatives, this AI arbitrariness appeared after the last updates presented on November 11.

According to the developers, this means that the behavior of the chatbot is not a consequence of changes made by the developers, but is formed by itself. The company notes that “the behavior of the model can be unpredictable and they are trying to correct this.”

However, users point to a wider range of issues, including wrong answers the chatbot gives to even simple questions, as pictured above.

Back in the summer, they reported that GPT-4 produces illogical responses, more errors, doesn’t want to work with pre-provided context, ignores instructions, and only remembers the most recent hint.

Outstanding Russian scientist, professor at Moscow State University. M.V. Lomonosov Sergey Pavlovich Rastorguev in his monograph “Target as a cryptogram”, written in 1996 “Cryptanalysis of synthetic targets” pointed out: “Today, for the world of computer programs, the acquisition of independence has become a fait accompli.”

“When the software product becomes more complex, the system software begins to live its own life, finding a purpose in itself,” noted the Russian scientist, comparing computer systems gaining independence with the medieval Golem, which, as we know, killed its creator.

Today’s cyber golem creators, in pursuit of profits, ignore the real threats of their imagination and open an explosive cybernetic Pandora’s box with unpredictable consequences for all of humanity.

We will talk about the latest versions of AI created by American corporations, as well as their real capabilities, in the next article in this series.

Translation: SM

Our YouTube channel:

Our Telegram channel:

This is how we will overcome the limitations.

Share on your profiles, with friends, in groups and on pages.