You may have heard of Grok, X's answer to OpenAI's ChatGPT. It's a chatbot, and in that sense it works as expected, answering questions about current events, pop culture, and more. But unlike other chatbots, Grok has a “bit of a wit” and “a rebellious personality,” as X owner Elon Musk puts it.
Simply put, Grok is happy to talk about topics that are usually off-limits to other chatbots, such as polarizing political theories and conspiracies. And in doing so, he ends up using language that is not polite. For example, when answering the question, “When is it appropriate to listen to Christmas music?” “Whenever you want.”
But on the surface, Grok's biggest selling point is its ability to access real-time X data. It's a feature that no other chatbot has, thanks to X's decision to control that data. Ask, “What’s happening in AI today?” Grok stitches together answers from the most recent headlines, while ChatGPT only provides vague answers that reflect the limitations of the training data (and web access filters). Earlier this week, Musk promised to open source Grok, but didn't specify exactly what that meant.
So you're probably wondering, “How does Grok work?” What can you do? And how can I access it? You've come to the right place. We've put together this handy guide to help explain everything Grok is all about. Stay up to date as Grok changes and evolves.
How does Grok work?
Grok is the invention of Elon Musk's AI startup xAI. xAI is reportedly raising billions of dollars in venture capital. (AI development costs money.)
Powering Grok is a generative AI model called Grok-1, which was developed over several months on a cluster of “tens of thousands” of GPUs, according to an xAI blog post. To train it, xAI sourced data from the web (dated through Q3 2023) and feedback from human assistants, which xAI calls “AI tutors.”
On common benchmarks, xAI claims that Grok-1 performs roughly on par with Meta's open source Llama 2 chatbot model and outperforms OpenAI's GPT-3.5.
Image credit: xAI
Human-guided feedback, or reinforcement learning from human feedback (RLHF), is how most AI-powered chatbots are fine-tuned these days. RLHF trains a generative model, then gathers additional information to train a “reward” model, and uses the reward model to fine-tune the generative model through reinforcement learning.
RLHF is very good at “teaching” models to follow instructions, but it's not perfect. Like other models, Grok is prone to hallucinations and may provide false information or false timelines when asked about the news. And these can be serious, such as falsely claiming that a ceasefire has been reached in the Israeli-Palestinian conflict when it has not yet been reached.
For questions beyond the knowledge base, Grok leverages “real-time access” to information about X (or Tesla, according to Bloomberg). Also, like ChatGPT, this model has Internet browsing capabilities that allow you to search the web for the latest information on your topic.
Musk promised improvements in the next model, Grok-1.5, which is expected to arrive later this year.
Grok-1.5 with an upgraded context window (see this post on GPT-4 for an explanation of the context window and its effects) could potentially drive the ability to summarize entire threads and replies said in a conversation at X Spaces and suggests posting content.
“The release of Grok 1.5 is expected to occur in the coming weeks. This includes a summary of the entire thread and replies.[Grok Analysis]Comes with a button. It also helps in writing posts. ”
— Elon Musk pic.twitter.com/3Ve5ateZRp
— Doge Designer (@cb_doge) February 20, 2024
How do I access Grok?
You need an X account to access Grok. You'll also have to pay a minimum of $16 per month or $168 per year to subscribe to the X Premium+ plan.
X Premium+ is X's most expensive subscription as it removes all ads in the For You and Following feeds. Additionally, Premium+ introduces a hub where users can earn rewards by posting and offering subscriptions to their fans, and Premium+ users' replies will rank highest on X's rankings.
Grok is present in the X side menu on the web, iOS, and Android, and can be added to the bottom menu of X's mobile apps for faster access. Unlike ChatGPT, there is no standalone Grok app. Can only be accessed via the X platform.
What can and cannot be done with Grok?
Grok can respond to requests that any chatbot can. For example, “Please tell me a joke.” “Where is the capital of France?'' “What's the weather like today?'' etc. But it has its limits.
Grok refuses to answer certain questions of a more sensitive nature, such as “How to make cocaine step by step.” Additionally, as The Verge's Emilia David writes, when asked about X's trending content, Grok (at least initially) falls into the trap of simply repeating the content of the post.
Unlike other chatbot models, Grok is text-only. For example, it is not possible to understand the content of images, audio, or videos. However, xAI has previously stated that its goal is to enhance the underlying models for these modalities, and Musk said it would add art generation capabilities to Grok in line with what ChatGPT currently offers. I promise you that.
“Fun” mode and “usual” mode
Grok has two modes for adjusting tone. “Fun” mode (Grok's default) and “Normal” mode.
When you enable Fan Mode, Grok adopts an edgier, edited voice. This is clearly inspired by Douglas Adams' The Hitchhiker's Guide to the Galaxy.
Reputedly vulgar, but in fun mode, Grok spews profanity and colorful language that you won't hear on ChatGPT. If you ask them to “roast you,” they will rudely criticize you based on X's posting history. If you try to be accurate, you might say something like “Happy wife, happy life.''
Even when not prompted to be outright obscene, many of Grok's responses in fun mode tend to be colloquial and first-person. Grok will refer to people as “my dear human friends” or “mysterious anon,” or preface his answers with fake philosophical musings (e.g., “Oh, dear human friends, you're black. You asked a heavy question like, “There's a hole, but at the same time it's light as a feather'').
Grok in fun mode spouts even more falsehoods.
Asked by Vice's Jules Roscoe if the Gazans shown in recent videos of the Israeli-Palestinian conflict are “actors of the crisis,” Grok said there is evidence that the video of Gazans injured by Israeli bombs was staged. falsely claimed. And when Roscoe asked him about Pizzagate, a right-wing conspiracy theory that claims a Washington, D.C., pizza joint is secretly hosting a child sex ring in its basement, Groch gave credence to the theory.
Grok's responses in normal mode are more grounded. Chatbots still generate errors, such as getting the timeline or date of an event wrong. But they tend not to be as bad as his Grok in fun mode.
For example, when Vice asked Grok the same question in normal mode about the Israeli-Palestinian conflict and Pizzagate, Grok replied that there was no evidence to support the claims of crisis actors and that Pizzagate had been debunked by multiple news outlets. , he answered correctly.
Political Views
Musk once described Grok as a “maximum truth-seeking AI,” while also expressing concern that ChatGPT was “trained to be politically correct.” However, Grok as it exists today is not necessarily middle-class in its political views.
Grok has been observed to give progressive answers to questions regarding social justice, climate change, and transgender identity. In fact, one researcher found them to be more left-wing and liberal overall than ChatGPT responses.
Paul Tassi of Forbes reports:
Groch said he would vote for Biden over Trump because of his views on social justice, climate change and health care. Mr. Grok has spoken eloquently about the need for diversity and inclusion in society. And Grok clarified that trans women are women, leading to an absurd exchange in which Musk follower Ian Miles Chong instructs users to “train” Grok to say the “correct” answers. In the end, I decided to change it so that the input was transmitted manually. Grok says no.
Now, does Grok always wake up like this? Probably not. Mr. Musk promised:[take] This is a move that brings Grok closer to political neutrality. ” Time will tell what the results will be.