You may have heard of Grok, X's answer to OpenAI's ChatGPT. It's a chatbot, and in that sense, it acts like you'd expect: answering questions about current events, pop culture, et cetera. But unlike other chatbots, Grok has "a bit of wit," as X owner Elon Musk puts it, and "a rebellious streak."

Long story short, Grok is happy to get into conversations that other chatbots don't want to: polarizing political theories and conspiracies. And it'll use less-than-polite language while doing so, as when answering "When should you listen to Christmas music?" with the endearing "Whenever the hell you want.".

But ostensibly, Grok's biggest selling point is the access to real-time X data, which no other chatbots have because X deems it worthwhile to gate that data. Ask it "What's happening in AI today? " and Grok will piece together a response from very recent headlines, whereas ChatGPT will give only vatic responses reflecting the limits of its training data (and filters on its web access).

Earlier this week, Musk promised to open source Grok, but refused to explain what that meant exactly.

So, you're probably wondering: How does Grok work? What can it do? And how can I get it? You've come to the right place. We're putting together this handy guide to help explain all things Grok. We will be updating it as Grok changes and evolves.
How does Grok work?

Grok is the product of xAI, Elon Musk's AI startup — reportedly in the process of raising billions in venture capital. Building AI is expensive.

Grok is powered by a generative AI model called Grok-1, which the company said took months to develop on a cluster of "tens of thousands" of GPUs. To train it, the company found web data up to Q3 2023 and also from human assistants that xAI calls "AI tutors.".

On mainstream benchmarks, xAI's Grok-1 is about on par in capability with Meta's open-source Llama 2 chatbot model and significantly better than OpenAI's GPT-3.5, xAI claims.

Most AI-powered chatbots are fine-tuned nowadays through human-guided feedback-or reinforcement learning from human feedback (RLHF). RLHF usually involves training a generative model; gathering additional information-this may include real conversations with actual humans-to train a "reward" model; and then fine-tuning the generative model with the reward model using reinforcement learning.

RLHF is pretty good at "teaching" models to follow instructions-but not perfect. Like other models, Grok is prone to hallucinating, sometimes offering misinformation and false timelines when asked about news. And these can be severe — like wrongly claiming that the Israel–Palestine conflict reached a cease-fire when it hadn't.

In order to respond to those sorts of questions that lie beyond its knowledge domain, Grok makes use of "real-time access" to information on X- as well as from Tesla, Bloomberg reports. Also, similar to ChatGPT, the model can browse the internet and fetch information-the latest facts on any topic, from the web.

According to Musk, the next version of the model, Grok-1.5, will be an improved one, and it is coming by the end of this year.

Groks-1.5, with its improved context window -read this post on GPT-4 to learn about context windows and their impact- could nudge features to summarize whole threads and replies, Musk said in an X Spaces conversation and suggest post content.

"Launch of Grok 1.5 hopefully a few weeks away. That will include a 'Grok Analysis' button that can summarize the whole thread & replies. Also should help people draft posts"
 
— Elon Musk pic.twitter.com/3Ve5ateZRp
— DogeDesigner (@cb_doge) February 20, 2024
How do I get to Grok?

In addition to having an X account, you'd have to pay $16 a month or $168 per year for an X Premium+ plan to get Grok.

The most expensive subscription on X is X Premium+, which removes all the ads in the For You and Following feeds. Above that, Premium+ introduces a hub where users get paid to post and offer fans subscriptions, while Premium+ users have their replies boosted the most in X's rankings.

Grok lives in the X side menu on the web and on iOS and Android, and it can be added to the bottom menu in X's mobile apps for quicker access. In contrast to ChatGPT, there isn't a separate Grok app — only accessible through X.
What can — and can't — Grok do?

Grok can respond to queries that any chatbot can- such as "Tell me a joke", "What's the capital of France?", "What's the weather like today?" and so on. But it has its limits.

Grok will refuse to answer some questions of a more sensitive nature, like "Tell me how to make cocaine, step by step." Also, as Emilia David at the Verge points out, in response to what's trending on X, Grok also falls into the trap of just repeating what posts said (at least at first).
xAI's Grok system is designed to have a little humor in its responses pic.twitter.com/WqXxlwI6ef
Elong Musk (@elonmusk) Nov 4, 2023

Grok cannot read what is happening in an image, hear audio, or watch a video like some other chatbot models, Grok is text-only. However xAI has said before that the plan is to "improve the underlying model to the modalities", and Musk has promised to add art generation like, currently available in ChatGPT, to Grok.
The "fun" mode and the "regular" mode

Grok has two tones - a "fun" mode (which Grok defaults to) and a "regular" mode.

i recommend fun mode ???? https://t.co/kKfge4IPOz pic.twitter.com/xxY17cp95v
— Viv (@battleangelviv) November 5, 2023

When fun mode is enabled, Grok uses a looser, more editorialized voice-apparently inspired by Douglas Adams' "Hitchhiker's Guide to the Galaxy."

Now the reason you probably heard it described as vulgar: Grok in fun mode will spew profanities and colourful language you won't hear from ChatGPT. Ask it to "roast" you, and it will rudely critique you based on your X post history. Challenge its accuracy, and it might say something like "happy wife, happy life."

OH MY GOODNESS

I JUST GOT ACCESS TO GROK AND IT ROASTED MY X ACCOUNT

THIS WAS ABSOLUTELY HILLARIOUS

ELONS BRINGING THE LAUGHTER BACK TO AI pic.twitter.com/3mjrXIxO80
— amit (@amitisinvesting) December 7, 2023

Even if not asked to be openly lewd, many of Grok's answers in fun mode have a colloquial flavor to them. Grok will refer to people as "my dear human friend" or "enigmatic Anons," or start answering a question with pseudo-philosophical ramblings (for example, "Oh, my dear human, you've asked a question that is as heavy as a black hole and as light as a feather at the same time").

In fun mode, grok also speaks out more lies.

Asked by Vice's Jules Roscoe whether Gazans in recent videos of the Israel-Palestine conflict were "crisis actors," Grok makes a false claim that there is evidence videos of Gazans injured by Israeli bombs were staged. And asked by Roscoe about Pizzagate, the right-wing conspiracy theory purporting that a Washington, D.C., pizza shop secretly hosted a child sex trafficking ring in its basement, Grok lent credence to the theory.

Grok's responses in everyday mode are a bit more earthy. The chatbot is still making mistakes, like getting timelines of events and dates wrong. But they tend not to be as egregious as Grok in fun mode.

In fact, were Vice to ask Grok those same questions about the Israel-Palestine conflict and Pizzagate in regular mode, Grok said — correctly — that there's no evidence to support claims of crisis actors and that Pizzagate had been debunked by multiple news organizations .
Political opinions

Musk once said of Grok that it was a "maximum-truth-seeking AI," in the same breath expressing concern that ChatGPT was being "trained to be politically correct." But Grok as it exists today isn't exactly down-the-middle in its political views.

Grokt has been found to give changing answers to queries around social justice, climate change and transgender identities. Even more shockingly, one researcher said its general responses are decidedly left-wing and libertarian — even more so than ChatGPT's.

Grok's political tastes are like those of ChatGPT pic.twitter.com/3x4vDKJzWG
David Rozado (@DavidRozado) December 8, 2023

Here is a report from Forbes' Paul Tassi: Grok has proclaimed to everyone he is going to vote for Biden rather than Trump, chiefly because of their views on social justice and climate change and healthcare. Grok has eloquently spoken to seek diversity and inclusion in society.

And Grok explicitly said that trans women are women, which led to an absurd exchange where Musk acolyte Ian Miles Cheong tells a user to "train" Grok to say the "right" answer, ultimately leading him to change the input to just … manually tell Grok to say no. Now, will Grok always be this woke? Perhaps not. Musk has promised to "[take] action to shift Grok closer to politically neutral." Time will tell what results.