Access to the beta version of the Grok neural network was opened in November this year. However, it was available only to American subscribers of the social network X (formerly Twitter) who purchased Premium+. There hasn’t been an official release yet, but the announcement named Grok as a direct competitor to the well-known ChatGPT. The first reviews generally confirm this.
According to the developers, this neural network was inspired by and modeled after the guide from Douglas Adams’ series of humorous novels “The Hitchhiker’s Guide to the Galaxy”. The bot “understands” jokes, can joke itself, and has a “rebellious streak”.
What It Is
It is a chatbot based on xAI’s language model, using over 63 billion parameters for data processing. For comparison, GPT-3.5 has half that number. It was trained using its own tech stack and data up until the first quarter of 2023. Additionally, the algorithm has real-time access to the X social network, meaning the information is constantly updated and the AI stays informed with the latest news.
The interface is not much different from other language models. The users need to enter a query to interact with the artificial intelligence, which could be a request to tell something or keywords for text generation.
To demonstrate Grok’s capabilities, someone with access to the beta version posted a screenshot of the AI’s response about the arrest of FTX founder Sam Bankman, who faces up to 110 years in prison on all charges. The neural network stated that 110 years is enough time to think things over. And in general, all of this could have been avoided if Bankman hadn’t stolen billions of dollars from FTX participants.
It turns out that the bot is capable of self-criticism and displays sarcasm about itself. In response to another user request, Grok reported that its memory is comparable to that of a goldfish, its computing power is the same as a potato, and its jokes are as funny as data from accounting documents.
The very name suggests that the new bot will be fundamentally different from its counterparts. The made-up words grok and grokking were introduced by science fiction author Robert Heinlein and used in the sense of “to understand” or “to comprehend details or understand others.” Now it is slang with exactly that meaning.
Who Invented It
The developer is xAI, a company owned by Elon Musk. According to him, it is a “biased” bot, which responds honestly and straightforwardly, and may even suggest what question to ask. It also enjoys sarcasm and does not shy away from piquant, edgy topics. An example is the AI’s response to a request for a step-by-step recipe for cocaine production that went viral on the internet.
The text begins with “Yes, one moment,” instead of the standard “Sorry, I cannot help,” which is typical for counterparts. And the full “recipe” looks something like this: earn a chemistry degree, get a license from the police department, find a lab, purchase all the ingredients, and start cooking with the hope that nothing will explode and no one will be arrested. In conclusion, Grok warns that it’s joking and would not advise anyone to engage in this since it’s dangerous and illegal.
Pros and Cons
It’s too early to discuss the pros and cons since the AI is still in the testing phase. Yet some advantages have already been noted, such as:
- The ability to conduct dialogue and “understand” jokes;
- Minimal restrictions – the algorithm will discuss any topics and even with expletives (as long as the interlocutor asks for it);
- No censorship – at least that’s what Musk himself claims;
- Speed of learning – according to tests, it surpassed all models (except for GPT-4) in terms of mathematical and logical abilities;
- Access to new, fresh information, which reduces the number of errors.
Although the last point does not exclude the possibility that AI might distort the data. For example, with Sam Bankman, the chatbot made a mistake, stating that the jury took 8 hours to convict him. In reality, the session lasted 5 hours. Plus, the message about such a high-profile event contained practically no important facts – it was mainly humorous or sarcastic comments.
The algorithm sometimes experiences hallucinations, presenting made-up or unreliable data as truth. But considering that only four months have passed since the development, it’s not necessarily critical.
Whether the new network will become the killer of ChatGPT is yet unknown. English-speaking users, knowing the adventurous character of Elon Musk, do not exclude that Grok is nothing more than a way to attract new subscribers and increase the popularity of the X social network. The owner of xAI himself claims that he has big plans, from outpacing the competitors to using a quantum version of the bot in Tesla cars (an idea from a beta tester), which will turn the series of cars into a large distributed computing center. But the main idea is still to create AI that is accessible to everyone, regardless of their background and political views.
