ChatGPT Rival Goes Live!

X has successfully rolled out its AI chatbot, Grok, to all Premium+ subscribers in the United States.

Elon Musk, the founder of the xAI startup, shared that while the beta version of Grok ( a ChatGPT competitor developed by xAI) may face challenges, continuous improvements are in progress. He outlined a timeline for the global expansion of Grok, indicating that all English language users with Premium+ subscriptions can expect access within “about a week or so.”

Following this, Japanese users, constituting X’s second-largest user base, are slated to gain access, with Musk expressing hopes for Grok to be available in “hopefully” all languages by early 2024.

Elon Musk is known for occasionally setting ambitious timelines, akin to experiences with Tesla’s full self-driving (FSD). However, in the case of Grok, Musk needs to catch up to his launch estimates. Although he initially mentioned a launch date of “next week” on November 22, Grok was officially launched on December 7.

The success of Grok in driving subscription revenue for X remains uncertain. Currently, Grok is a feature exclusive to the top-tier Premium+ subscription, priced at $16 per month. This contrasts with X’s Basic ($3/mo) and Premium ($8/mo) options, raising questions about its appeal to casual AI users who can access competing chatbots like ChatGPT or Google’s Bard for free.

Premium+ subscribers enjoy various perks, such as ad-free experiences in For You and Following timelines, boosted replies, ads revenue sharing for creators, ID verification, a verified checkmark, access to Media Studio, and more. However, X has traditionally relied on advertising for most of its revenue, and the impact of Grok on this dynamic is yet to be seen.

Elon Musk’s recent confrontations with advertisers leaving X over concerns of antisemitic content have added uncertainty to the platform’s ad-supported future. To sustain itself, X may need a significant increase in Premium+ subscribers to compensate for the loss of ad revenue, as major brands like Apple, Disney, IBM, Paramount, and Walmart exit the platform.

Despite these challenges, X experienced its highest-ever month for subscription revenue in November’23, reaching $6.2 million in net revenue after app store fees, according to estimates by app intelligence provider Apptopia. Yet, this figure is still less than a third of Snapchat’s in-app subscription revenue, which surpassed $20 million for the first time in the same month.

With over 500 million monthly active users, X has the potential for substantial subscriber growth. However, whether it can achieve this growth remains uncertain in the evolving landscape of social media platforms.

But now our question here is what exactly is Grok and what it’s going to do for us.

What is Grok?

Grok, developed by Elon Musk’s xAI company, is an AI bot similar to OpenAI’s ChatGPT but with the unique capability to tap into real-time knowledge through the X platform, formerly known as Twitter. Distinguishing itself from other models, Grok fearlessly engages with “spicy” questions and responds with a blend of wit and rebellion.

Despite being in beta, Grok has undergone rapid development over a short two-month training period, with xAI expressing confidence in its evolution through user feedback.

How Does Grok Work?

Grok’s development involved creating a robust infrastructure using Kubernetes, Rust, and JAX. The team emphasises the significance of reliable infrastructure in deep learning research, leveraging Rust for its performance and reliability. As Grok advances, the focus remains on scalability, efficiency, and reliability in training and inference systems.

The research goals for Grok include addressing the limitations of current AI models by emphasising reliable reasoning, scalable oversight, integration with formal verification for safety, long-context understanding and retrieval, and improving adversarial robustness. Plans to enhance Grok with multimodal capabilities, such as vision and audio, are in progress.

Does Grok Answer Uncensored Queries?

Grok stands out by answering “spicy” questions rejected by other AI systems, showcasing notable results on controversial queries. Elon Musk shared a response to a user inquiring about making cocaine, where Grok injected humour and later provided a detailed response for “educational purposes.”

This query represented Grok differently from other AI tools that avoid giving answers to such questions.

Is Grok better than ChatGPT?

In comparison to ChatGPT, Grok-1, the engine behind Grok, demonstrates impressive performance in machine learning benchmarks, surpassing models like ChatGPT-3.5 and Inflection-1 in its compute class.

Musk also shared an image on Twitter showcasing how Grok utilises current information in its responses compared to other typical GPTs.

Grok’s accuracy was tested in the 2023 Hungarian National High School Math Exam, earning a grade of C. The expectation is that, with more advanced learning algorithms, Grok will outpace the learning and adaptation speed of competitors like ChatGPT and Bard.

Delving into Grok AI’s capabilities, xAI has meticulously unveiled performance metrics that shed light on Grok-1’s formidable strengths. The company proudly reports that Grok AI excels with a remarkable 63.2% accuracy on the HumanEval coding task and an impressive 73% on MMLU. Despite not reaching the pinnacle of GPT-4’s capabilities, Grok-1’s performance showcases its prowess in various tasks. xAI underscores the substantial progress witnessed in Grok-1 compared to its predecessor, Grok-0, achieved within a remarkably short timeframe.

In a significant benchmark, Grok-1 outperformed expectations by scoring 62.9% on GSM8k, specifically designed for middle school math word problems. This achievement surpassed both GPT-3.5 and LLaMA 2, showcasing Grok-1’s adeptness in mathematical reasoning. However, it fell short of surpassing benchmarks set by Palm 2, Claude 2, and GPT-4, signalling room for potential enhancements.

Beyond mathematics, Grok AI continued to impress in various benchmarks such as MMLU for multi-choice questions, HumanEval for Python code generation, and MATH for middle and high school math. Grok-1 consistently demonstrated competitive positioning across these assessments, underscoring its versatility and promising potential across diverse domains.

However, concerns arise regarding Grok’s accuracy, as it was predominantly trained on Twitter’s user data, known for misinformation. Critics point out that ChatGPT relies on more reliable sources like published journals, websites, and books.

Elon Musk’s motivation for releasing Grok is seen as linked to his plans to monetise Twitter and transform X into a super app, having obtained a banking license for the platform and preparing to launch live streaming services.

Grok is currently in limited availability to U.S. users for feedback purposes. Elon Musk mentioned that, once out of beta, Grok will be accessible to X Premium+ subscribers, prioritised based on subscription duration. Interested users can join the waitlist for access.

Hope you like reading this article! Want to know more about Grok, stay tuned to our news section!