Elon Musk has entered the competitive world of artificial intelligence with Grok, a conversational AI poised to rival ChatGPT and Midjourney. Grok aims to provide a chatbot experience infused with “a sense of humor,” featuring fewer content restrictions and a candid tone, as Musk humorously noted.
This innovative chatbot is built on the X social media platform, leveraging data from its users. Here's what you need to know about Grok.
What is Grok?
Grok is a generative AI chatbot developed by xAI, Musk’s AI startup. It initially utilized the Grok-1 model, created over several months using “tens of thousands” of GPUs and enhanced by the Flux.1 model from Black Forest Labs for image generation.
Grok is trained on a diverse mix of web data (with information available up to Q3 2023) and data from X users. As of now, the chatbot operates on the Grok-2 model, which launched in August 2024. It first became available in November 2023 as a premium offering for subscribers paying $16 a month for X Premium+. In March 2024, the service expanded to those on the $8/month Premium plan, although free X users still don’t have access. The name “Grok” originates from Robert Heinlein’s sci-fi novel Stranger in a Strange Land, where it signifies the ability to understand intuitively or establish rapport.
Where did Grok come from?
In 2015, Elon Musk co-founded OpenAI with Sam Altman and others but departed in 2018, citing disagreements with the organization's direction. Following the launch of ChatGPT in November 2022, Musk expressed concern over its approach and announced plans in April 2023, on the Tucker Carlson Show, to create “a maximum truth-seeking AI” unconstrained by social norms. Originally named "TruthGPT," this AI was later renamed Grok.
In March 2024, Grok-1 was upgraded to Grok-1.5, enhancing its performance and extending context length to 128,000 tokens. That April, X integrated Grok AI into its Explore section to summarize breaking news, taking over tasks previously managed by humans. However, it faced controversy when it mistakenly reported an incorrect headline concerning Iran and Israel, which highlighted the AI's susceptibility to errors. Grok-1.5 has since transitioned to the Grok-2 model, along with Grok-2 mini, both released in August 2024.
What can Grok do?
According to its launch announcement, Grok is “modeled after The Hitchhiker’s Guide to the Galaxy," designed to respond to a wide array of inquiries. This open-source large language model features 314 billion parameters—about three-quarters the scale of Llama 3.1-405B—and has been trained on publicly available web data as well as insights from X users.
In July, X made a controversial adjustment to user settings that allowed xAI to scrape data for training, sparking backlash and swift reversal after privacy advocates voiced concerns. This prompted a lawsuit from nine European Union nations for potential GDPR violations. Grok can address user queries based on its Q3 2023 knowledge cutoff and can perform web searches and access real-time information from X for events occurring afterward. This may account for Grok's increased tendency to produce hallucinations and spread misinformation compared to competitors like ChatGPT.
Grok’s standout feature is its openness to discussing taboo topics, including politics, religion, and race. For example, when an X employee prompted it for a playful response regarding the appropriate time to listen to Christmas music, Grok responded, “whenever the hell you want,” advising detractors to “mind their own damn business.”
Additionally, Grok unveiled new image generation capabilities alongside Grok-2 and Grok-2 mini in August, featuring minimal guardrails. While Grok 2.0 claims to have certain restrictions, users can generate images of celebrities, political figures, and even copyrighted characters like Disney’s Mickey Mouse and Nintendo’s Mario.
Given the stringent IP protections enforced by companies like Disney and Nintendo, expect potential lawsuits and significant changes to Grok’s image generation features in the future.
Ranking
