You might\u2019ve heard of Grok<\/a>, X\u2019s answer to OpenAI\u2019s ChatGPT<\/a>. It\u2019s a chatbot, and, in that sense, behaves as as you\u2019d expect \u2014 answering questions about current events, pop culture and so on. But unlike other chatbots, Grok has \u201ca bit of wit,\u201d as X owner Elon Musk puts it, and \u201ca rebellious streak.\u201d<\/p>\n Long story short, Grok is willing to speak to topics that are usually off limits to other chatbots, like polarizing political theories and conspiracies. And it\u2019ll use less-than-polite language while doing so \u2014 for example, responding to the question \u201cWhen is it appropriate to listen to Christmas music?\u201d with \u201cWhenever the hell you want.\u201d<\/p>\n But Grok\u2019s ostensible biggest selling point is its ability to access real-time X data \u2014 an ability no other chatbots have, thanks to X\u2019s decision to gatekeep that data. Ask it \u201cWhat\u2019s happening in AI today?\u201d and Grok will piece together a response from very recent headlines, while ChatGPT, by contrast, will provide only vague answers that reflect the limits of its training data (and filters on its web access). Earlier this week, Musk pledged<\/a> that he would open source Grok, without revealing precisely what that meant.<\/p>\n So, you\u2019re probably wondering: How does Grok work? What can it do? And how can I access it? You\u2019ve come to the right place. We\u2019ve put together this handy guide to help explain all things Grok. We\u2019ll keep it up to date as Grok changes and evolves.<\/p>\n Grok is the invention of xAI<\/a>, Elon Musk\u2019s AI startup \u2014 a startup reportedly in the process of raising<\/a> billions<\/a> in venture capital. (Developing AI\u2019s expensive.)<\/p>\n Underpinning Grok is a generative AI model called Grok-1, developed over the course of months on a cluster of \u201ctens of thousands\u201d of GPUs (according to an xAI blog post<\/a>). To train it, xAI sourced data both from the web (dated up to Q3 2023) and feedback from human assistants that xAI refers to as \u201cAI tutors.\u201d<\/p>\n On popular benchmarks<\/a>, Grok-1 is about as capable as Meta\u2019s open source Llama 2<\/a> chatbot model and surpasses OpenAI\u2019s GPT-3.5<\/a>, xAI claims.<\/p>\n Image Credits:<\/strong> xAI<\/p>\n<\/div>\n Human-guided feedback, or reinforcement learning from human feedback (RLHF), is the way most AI-powered chatbots are fine-tuned these days. RLHF involves training a generative model, then gathering additional information to train a \u201creward\u201d model and fine-tuning the generative model with the reward model via reinforcement learning<\/a>.<\/span><\/p>\n RLHF is quite good at \u201cteaching\u201d models to follow instructions \u2014 but not perfect. Like other models, Grok is prone to hallucinating<\/a>, sometimes offering misinformation and false timelines when asked about news. And these can be severe \u2014 like wrongly claiming that the Israel\u2013Palestine conflict reached a ceasefire when it hadn\u2019t.<\/p>\n For questions that stretch beyond its knowledge base, Grok leverages \u201creal-time access\u201d to info on X (and from Tesla, according<\/a> to Bloomberg). And, similar to ChatGPT, the model has internet browsing capabilities, enabling it to search the web for up-to-date information about topics.<\/p>\n Musk has promised improvements with the next version of the model, Grok-1.5, set to arrive later this year. This new model could drive features to summarize whole threads and replies, Musk said<\/a> in an X Spaces conversation, and suggest content for posts.<\/p>\nHow does Grok work?<\/h2>\n
![\"Grok](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2024\/03\/Screenshot-2024-03-13-at-1.44.56\u202fPM.png\")
<\/p>\n