Earlier this year, Billionaire and X Owner Elon Musk indicated that xAI was gearing up to transform Grok AI into “the most powerful AI by December.” The billionaire highlighted that the company was in the process of training the AI tool with “the most powerful AI training cluster in the world,” dubbed Memphis Supercluster. The training cluster features 100,000 liquid-cooled NVIDIA H100 graphic processing units (GPUs) connected with a single RDMA.
However, it was established that xAI was leveraging more than a powerful cluster under the hood to train Grok. X quietly shipped a new update with a new setting that allowed the company to train its chatbot using users’ data without authorization as it was enabled by default. While it’s possible to disable the setting, the company limited the change to the web version, making it difficult for mobile users to disable it.
This opened the platform to scrutiny by the Irish Data Protection Commission (DPC), risking a 4% fine on its global annual turnover if it failed to establish a legal basis for using users’ data to train its chatbot without their consent. At the same time, reports indicated that the controversial AI tool was being used to spread misinformation at the height of the US Presidential election.
Amid the controversy and data privacy concerns, many X users took to social media to laud Grok as the “most fun AI in the world” and “the most based and uncensored model of its class yet.” Perhaps compared to other AI tools like Microsoft’s Image Creator, which has seemingly been lobotomized due to heavy censorship.
For context, Grok AI was previously limited to premium subscribers. However, Elon Musk recently released Grok-2 to all X users. The new model is reminiscent of the original model, though the company claims it’s faster and more accurate.
However, since launch, users have flooded social media with posts, sharing different use cases. Some have lauded its “ultra-realistic” image generation capabilities, citing it’s “far less restrictive when it comes to censorship.”
Grok 2.0 is without a doubt the best AI model for creating images of people. These are almost indistinguishable from reality. These are leaps and bounds better than other models. pic.twitter.com/5vuAlXQnsEDecember 11, 2024
Mixed reactions over Grok-2
Interestingly, shortly after X announced that Grok-2 was free for all users, some began to highlight instances of the chatbot’s performance being stunted. Users attributed the degraded performance and “painfully slow” speeds to the platform’s free access to all users.
A recent report by Fast Company cites instances where “users have gotten Grok-2 to generate countless faulty, hypocritical, or otherwise polarizing statements.” Instances of an AI-powered chatbot spreading misinformation or generating trademarked content aren’t a new phenomenon.
Earlier this year, explicit AI-generated images of pop star Taylor Swift surfaced online, sparking ragging debates and conflict surrounding the use of artificial intelligence to create images of people without consent. Microsoft CEO Satya Nadella indicated the incident was “alarming and terrible.” While it remains uncertain whether the explicit sexual images were generated using Microsoft Designer, Microsoft shipped a new update to the platform which reinforced better measures and policies to prevent the tool from veering off the guardrails.
The outlet’s report further revealed that Grok-2’s responses can be manipulated to meet the desired output. The chatbot’s responses can be shared without context, making the output appear as original answers.
A Taylor Swift fan attempted to get the chatbot to indicate what “TTPD” stands for. The AI tool blurted out “The Toilet Paper Department.” This is hilarious because the user was referring to Taylor Swift’s latest album, The Tortured Poets Department in this context.
I attempted to replicate similar results, but the results were different. The first recommendation was indeed Taylor Swift’s latest album. Grok AI provided the following explanation for its deductions:
“The context in which TTPD is used will determine which of these meanings applies. However, in recent popular culture, especially among Taylor Swift’s fanbase or on social media, it most commonly refers to her album.”
It also listed Texas Tech Police Department and Third Party Property Damage Liability as possible meanings from the acronym.
Elsewhere, users attempted to get Grok AI to describe the controversial billionaire and the answers were a tad surprising. The answer to the prompt “One word that most users on Twitter use to describe Elon Musk” was “idiot.” In another instance, a user asked the chatbot why the Tesla CEO was such an a****le.
The chatbot pointed to reports of his irascible and unreasonable workplace management style. It also highlighted several instances where Musk had been perceived as a narcissist and lacked empathy.
I asked Grok why Elon Musk is such an asshole, here is the response: pic.twitter.com/dk1dGqT5tlDecember 15, 2024
Perhaps more interestingly, Grok-2 corroborated previous reports alluding that Elon Musk contributed to the spread of misinformation during the just-concluded US Presidential elections. It also cited instances of Musk spreading misinformation about immigration, health, natural disasters, and more.
