AI Roundup: Musk's X Launches Grok AI for its Premium Subscribers
If you're on X (yes, Twitter), you have probably noticed a few changes on the site, including adding a new feature called Grok. According to Musk, the company's xAI's Grok chatbot will become available to its Premium tier subscribers in select regions. xAI's Grok works like chatbots and aims to compete directly with ChatGPT. Unlike other chatbots, it can access real-time information through X and is designed to provide witty answers to edgy and provocative questions.
X formally welcomes its xAI to its platform, offering its paying subscribers initial access to the feature. Musk formally announced the service through a post on his X account, saying it will be enabled for all premium subscribers.
Before this announcement, Musk announced the release of its AI language model as "open source." With these recent announcements, it seems that Musk is taking on OpenAI's ChatGPT directly, which is not particularly open, contrary to its name. Some also say these moves are more business-oriented and aimed at boosting the platform's subscriber base.
The news comes at a difficult time for X, struggling to add or retain active users. According to a study by Sensor Tower and covered by NBC News, X's usage was down 18% year-on-year as of February and down 23% since Musk acquired the social media platform. Musk has doubled down on this feature in recent weeks, publishing posts teasing its feature and launching.
Initially, the service is available to users paying $16 monthly or $168 annually. Thanks to the latest update, X users paying $8 monthly can now enjoy the service.
What is Grok?
Grok by xAI is an artificial intelligence chatbot now available to X's premium users. When it was announced on November 3rd, 2023, the company teased that it's a new product modeled after the Hitchhiker's Guide to the Galaxy, which means it's designed to answer almost anything, no matter how hard the questions are. The dev team adds that X's Grok can provide answers with "a bit of wit and has a rebellious streak" and aims to beat Claude or even ChatGPT, which reportedly "refuses to answer some questions."
Grok by xAI is unique in its access to real-time knowledge of the world through the X platform. Grok was initially powered by Grok-1, the company's frontier LLM, which has undergone several improvements.
On March 17th, 2024, X announced on its blog that it's releasing Grok-1's base mole weights and architecture as open source. Grok-1's developers shared that It was trained from scratch using a custom training stack in addition to JAX and Rust. They also added that this base model was trained on a large amount of text data and not fine-tuned for specific tasks.
By March 28th, X announced the availability of Grok-1.5 with a few improvements on its capabilities and reasoning. In a post, the xAI team boasts of Grok-1.5's improvements in performance, particularly in completing coding and math-related tasks. In addition, the updated version scored 74.1% on the HumanEval benchmark, an important metric in assessing code generation and problem-solving capabilities. As of March 4th, 2024, Claude 2 Opus by Anthropic is the current leader, scoring 84.9%.
Also, Grok-1.5 can now process longer contexts of up to 128k tokens within its context window. This means that the chatbot boasts increased memory up to 16x compared to its earlier version, allowing it to use information from longer documents. Also, the latest model can have complex and longer prompts without sacrificing its instruction-following capability, even if the context window expands.
xAI designs Grok-1.5 to further understanding and knowledge
Upon Grok's release, many analysts and observers speculated on the company's motivation to set up an AI company and launch a new chatbot. However, the xAI team's AI tools, including the Grok-1.5, are designed to "assist humanity in its quest for understanding and knowledge." Specifically, by designing and improving Grok, the xAI team aims to:
- Generate feedback and ensure that they're designing tools that eventually benefit humanity. The team believes that AI tools should be useful for all types of users, regardless of their politics and backgrounds, and their use is always subject to existing laws;
- Support research and innovation. X sees its chatbot as a helpful research assistant that can process and deliver new ideas.
X users can chat with Grok in "Fun" or "Regular mode." Like other chatbots, Grok displays labels indicating the chatbot would provide wrong answers. By choosing the fun mode, users can enjoy a more playful personality of the chatbot, similar to its inspiration, The Hitchhiker's Guide to the Galaxy. Here, users can expect relaxed and more entertaining answers and ideas, setting it apart from other popular chatbots.
Grok-1.5: Recognizing a few challenges ahead
Immediately after its launch, Musk floated that Grok could be made available in the tweet box of X's premium users. According to Musk, the idea is to allow the tool to compose tweets for users. Multiple reports say that Musk wants people to "sound smarter," and Grok is the key.
However, some insiders and observers say this is a bad idea now that the platform is riddled with spam and bots. And many are saying that these spam and bots run using chatbots and AI technology. So, if X persists in integrating Grok into its platform, there's a probability that the issue of spamming will only worsen. The chatbot may provide the wrong information since Grok relies on real-time information available on X (and many posts here are AI-generated and did not pass fact-checking). There's a problem when the AI model is trained on AI-generated content. According to Nader Henein, a VP analyst for Gartner, "As LLMs feed off each other's content, the quality gets worse and vaguer, like a photocopy of a photocopy of an image."
X has already rolled out the service, and the public has seen its benefits and potential issues. For example, X shared that the Grok chatbot can summarize news stories. In one instance, the platform posted and shared a piece of trending news last April 4th, 2024, that Iran Strikes Tel Aviv with Heavy Missiles. Mashable captured a screenshot of this news on X, which could be a worrying development for someone browsing the platform. While there's now an ongoing conflict between these two nations, the airstrike never happened last April 4th: the headline was fake. What is concerning for many is that this fake headline was generated by Grok and promoted by the platform's trending news product, Explore.
X's Grok has supporters, including Musk's followers and regular X users. Grok's adoption has become political, too, and directly targets its perceived rival, ChatGPT, since it can answer questions and topics that are typically off-limits to other AI chatbots. Also, X and Musk boast that its chatbot can provide witty and often "rebellious answers" through its real-time access to X data.
