“It loves sarcasm. I have no idea who could have guided it that way,” jokes the entrepreneur
In the novel ‘Stranger in a Strange Land’, published in 1961 by Robert A. Heinlein, the word “Grok” is used as a kind of synonym for “assimilate”, a way of understanding a concept clearly and intuitively. Now that old expression, which made its fortune in the US youth slang of the 1960s, has added a new meaning: it is the name chosen by Elon Musk for the chatbot developed by his company xAI, which gives an eloquent idea of what the tool Musk wants to fight for with ChatGPT aspires to.
From science fiction to AI. Until now, “Grok” was a word from science fiction literature that managed to make its way into the hearts of young Americans half a century ago. Until now, that is. Musk has decided to rescue it to baptize the chatbot created by his artificial intelligence (AI) company, xAI, a project announced in the spring but officially launched in July.
Elon Musk himself has been in charge of sharing the name and some of the chatbot’s peculiarities through his favorite loudspeaker, the social network X, formerly Twitter, where he has posted comments and screenshots of its abilities. Few yet, but significant. “Tomorrow xAI will launch its first AI to a select group. In some important respects, it’s the best there is now,” tweeted last Friday the also responsible for Tesla or SpaceX.
“Respond with some ingenuity.” The expression is from xAI and reflects one of the peculiarities of the chatbot on which both the company and Musk himself have insisted the most, who on Saturday took it upon himself to share some screenshots with the answers generated by the new AI to answer certain questions. “It relies on and loves sarcasm. I have no idea who has been able to guide it that way,” tweeted the tycoon next to the emoticon of a smiley face. And for proof, a button.
When asked for news about SBF, an acronym for Sam Bakman-Fried, the former crypto entrepreneur and founder of FTX, who has just been convicted on seven charges, the xAI chatbot started with a sneer: “Oh, dear human, I have juicy news for you!” Something similar happened when asked for information on how to make cocaine. In that case the chatbot would deploy a four-step list that started with an invitation to “achieve a degree in Chemistry and a DEA license,” though it ended by warning that such a practice would be illegal and dangerous. “The Grok system is designed to have a bit of humor,” Musk acknowledges.
Access to X… and with a well-defined roadmap. These are not the only hints left by the entrepreneur in his profile. Beyond his penchant for sarcasm, Musk has detailed some important features of the chatbot, such as that it will have “real-time access to information through the X platform”, a peculiarity that, he claims, will offer it “a great advantage” over other models. The tycoon also specified how Grok will be deployed and, most importantly, who will be able to use it and how.
Interested? Checkout. On Friday, Musk said that xAI would launch its first AI for “a select group” the following day, but on Saturday he clarified that “as soon as the initial beta is released, xAI’s Grok system will be available to all X Premium+ subscribers”. The tagline is key because it indicates that the tool will be available to those who pay for the network’s most ambitious mode, the one that allows them to do without advertising for a monthly payment of $16.
Musk was in charge of confirming hours later that Grok will be included in the X Premium+ package, and launched a warning to navigators: “I recommend signing up.”
“Answer almost anything.” The expression is again from xAI, which in the last few hours posted on its website a much longer article in which it presents Grok, an AI “capable of answering almost any question and, what is more difficult, even suggesting what questions to ask!” Beyond the rhetoric, the publication is interesting for the technical data it slips about the chatbot, a tool, it acknowledges, that is still in a “very early” beta phase and hopes to improve.
“Grok’s engine is Grok-1, our state-of-the-art LLM, which we have been developing for the past four months,” says xAI, which explains that following xAI’s announcement a prototype LLM (Grok-0) was trained with 33 billion parameters.
Improving capabilities
“This first model approaches the capabilities of LLaMA 2 (70B) in standard LM benchmarks, but only uses half of its training resources,” the company continues in its release. In the last two months we have made significant improvements in reasoning and coding capabilities that have led us to Grok-1, a next-generation language model that is significantly more powerful, achieving 63.2% in HumanEval coding and 73% in MMLU.”
xAI even provides a table of some test results in which, it claims, Grok-1 “outperformed all other models in its class of computation, including ChatGPT-3.5 and Inflection-1.” “It is only outperformed by models that were trained with significantly more data and computational resources, such as GPT-4. This demonstrates the rapid progress we are making at xAI in training exceptionally efficient LLMs.”
It’s not all bragging, of course. The company explains that it gives Grok access to real-time search tools and information, but also acknowledges that, “as with all LLMs capabilities in predicting the next token,” its model can still generate “false or contradictory information.” “Achieving reliable reasoning is the most important research direction to address the current limitations,” he ditches.
The AI pie. Elon Musk’s move adds more competition and pressure to the AI arena, in which some of the most important multinationals in the sector have already made a move: Microsoft has given a “multi-million dollar” backing to OpenAI, the company behind ChatGPT; and a few days ago Google made a similar move, giving $2 billion to Anthropic, which had already won Amazon’s support.
Musk is not a newcomer to the AI race. He was already involved in the founding of OpenAI and his plans to launch his own personal project have been known for some time. Months ago the entrepreneur even threatened to sue Microsoft claiming that the company was using data from its X platform, then still known as Twitter, to train its artificial intelligence.