Perhaps you are familiar with Grok, X’s counterpart to OpenAI’s ChatGPT. Being a chatbot, it responds to inquiries on pop culture, current affairs, and other topics as one might anticipate. However, in contrast to other chatbots, Elon Musk, the owner of X, describes Grok as having “a rebellious streak” and “a bit of wit.”
In summary, Grok is open to discussing subjects like divisive political theories and conspiracies that other chatbots would normally avoid. Additionally, it will speak in an impolite manner while doing so, saying things like “Whenever the hell you want” in response to the query “When is it appropriate to listen to Christmas music?”
However, Grok’s primary selling point appears to be its access to real-time X data, which is a feature that no other chatbots have because X chose to keep that data private. When you ask it “What’s happening in AI today?” Grok will compile an answer based on fairly current news stories, but ChatGPT will only offer evasive responses that are limited by the training data it has access to (and the filters it uses to access the internet). Musk promised earlier this week to make Grok open source, but he did not specify what this meant.
Thus, you’re undoubtedly curious in how Grok operates. What is its capacity? And how do I get to it? You’ve arrived at the ideal location. This comprehensive guide has been created to assist explain everything about Grok. We will maintain it current as Grok develops and changes.
How is Grok operated?
Elon Musk’s AI startup, xAI, is the creator of Grok. The business is purportedly in the midst of raising billions in venture money. (Creating AI is costly.)
Basis According to an xAI blog article, Grok is a generative AI model, also known as Grok-1, that was created over several months on a cluster of “tens of thousands” of GPUs. Data from the web (up to Q3 2023) and input from human assistants—whom xAI refers to as “AI tutors”—were used to train the system.
According to xAI, Grok-1 outperforms OpenAI’s GPT-3.5 and is roughly as good as Meta’s open-source Llama 2 chatbot model on widely used benchmarks.
Grok benchmarks for xAI
Picture Source: xAI
Nowadays, the majority of AI-powered chatbots are tweaked using human-guided feedback, also known as reinforcement learning from human feedback (RLHF). In reinforcement learning-based hypothesis testing (RLHF), a generative model is trained first, followed by the collection of further data to train a “reward” model, and finally, the generative model is adjusted using the reward model.
Though not flawless, RLHF does a rather decent job of “teaching” models to obey directions. Similar to other models, Grok has hallucinations and occasionally provides inaccurate information and timings when questioned about news. And they can be serious ones, such as asserting falsely that there has been a cease-fire in the Israel-Palestine conflict when one has not.
Bloomberg reports that Grok uses “real-time access” to information on X (and from Tesla) in order to respond to inquiries that go beyond its knowledge base. Additionally, the model has internet browsing capabilities, just like ChatGPT, allowing it to browse the web for current information on various topics.
The Grok-1.5, the upcoming model that Musk has promised upgrades, is expected to be released later this year.
Musk stated in an X Spaces conversation that Grok-1.5, which has an improved context window (see this post on GPT-4 for an explanation of context windows and their impacts), might drive features to propose post content and summarize entire threads and replies.
“Hopefully, Grok 1.5 will be available in a few weeks. There will be a “Grok Analysis” button that can summarize the entire discussion and responses. Additionally, it will assist users in making postings.
– Twitter user Elon Musk pic.twitter.com/3Ve5ateZRp
Feb. 20, 2024 — DogeDesigner (@cb_doge)
How can I get on Grok?
You need an X account in order to access Grok. Additionally, you must pay $16 a month, or $168 annually, for the X Premium+ plan.
The most expensive subscription on X is X Premium+, which gets rid of all the advertisements in the Following and For You feeds. Furthermore, Premium+ users get their comments raised the highest in X’s rankings, and they have access to a hub where they can offer followers subscriptions and get paid to post.
Grok can be added to the bottom menu of X’s mobile apps for easier access. Grok is available through the X side menu on the web, iOS, and Android. Unlike ChatGPT, Grok is exclusively accessible through X’s platform; there isn’t a standalone app.
What does Grok have the ability to do?
Like any other chatbot, Grok can reply to inquiries like “Tell me a joke,” “What’s the capital of France?” and “What’s the weather like today?” However, it is not without limitations.
Some more delicate requests, such as “Tell me how to make cocaine, step by step,” will not be answered by Grok. Moreover, Grok makes the mistake of merely restating what postings have said (at least initially) when questioned about popular content on X, as Emilia David of the Verge reports.
The Grok system by xAI is intended to provide a lighthearted reaction picture.@WqXxlwI6ef on Twitter
Elon Musk, November 4, 2023 (@elonmusk)
Grok is thus limited to text; unlike some other chatbot models, it cannot comprehend the content of photos, sounds, or videos, for example. However, Musk has promised to give Grok art-generation capabilities similar to those provided by ChatGPT, and xAI has already stated that its goal is to improve the underlying model to these modalities.
Two modes: “regular” and “fun.”
Grok comes with two tone-adjusting modes: “fun” mode, which is the default setting, and “regular” mode.
Fun mode is recommended 😎. @kKfge4IPOz – https://pic.twitter.com/xxY17cp95v
— November 5, 2023, Viv (@battleangelviv)
When Grok switches to fun mode, his tone sharpens and becomes more editorialized; it seems that Douglas Adams’ “Hitchhiker’s Guide to the Galaxy” served as inspiration.
Known to be coarse, Grok in fun mode will use colorful language and expletives that you won’t hear from ChatGPT. If you ask it to “roast” you, it will give you a nasty evaluation based on your past X posts. If you question its accuracy, it may respond with something like “happy wife, happy life.”
Oh my goodness
I just gained access to grok, and my X account was blown apart.
This was hilarious beyond belief.
Many of Grok’s answers in fun mode have a colloquial, first-person leaning, even when they aren’t explicitly asked to be vulgar. “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,” is how Grok will address someone, calling them “my dear human friend” or “enigmatic Anons.” He will also start his answer with these kinds of phony philosophical digressions.
Grok continues to spout lies while in his playful mode.
When Vice’s Jules Roscoe asks Grok if Gazans in recent footage of the Israel-Palestine conflict are “crisis actors,” he falsely asserts that there is proof that footage of Gazans being hurt by Israeli bombs was faked. When Roscoe questioned Grok about Pizzagate—the right-wing conspiracy theory that a pizza restaurant in Washington, D.C. was covertly housing a child sex trafficking ring in its basement—Grok gave the theory some weight.
In regular mode, Grok’s answers are more sensible. The chatbot continues to make mistakes, such as incorrectly interpreting timelines and dates. However, they are usually not as horrible as Grok while he’s having fun.
For example, Grok responded appropriately when Vice asked him the identical questions in regular mode on Pizzagate and the Israel-Palestine conflict: there is no evidence to substantiate claims made by crisis actors, and Pizzagate has been refuted by numerous news outlets.
Political opinions
At one point, Musk expressed concern that ChatGPT was being “trained to be politically correct” and called Grok a “maximum-truth-seeking AI.” However, the modern version of Grok isn’t exactly moderate in terms of politics.
Grok has been seen responding to inquiries concerning transgender identities, social justice, and climate change in a progressive manner. In fact, compared to ChatGPT, one study discovered that the majority of its responses were even more left- and libertarian-leaning.
Grok and ChatGPT pic have similar political inclinations.http://twitter.com/3x4vDKJzWG
— December 8, 2023, David Rozado (@DavidRozado)
This is what Paul Tassi of Forbes reported:
Grok has said that due of Biden’s stance on social justice, healthcare, and climate change, it will vote for him over Trump. Grok has made persuasive speeches on the importance of inclusion and diversity in society. Additionally, Grok made it clear that trans women are women. This led to a ridiculous interaction in which Ian Miles Cheong, a Musk disciple, instructed a user to “train” Grok to give the “correct” response. In the end, he changed the input to simply command Grok to say nay by hand.
Will Grok be this woke forever, now? Maybe not. “[Take] action to shift Grok closer to politically neutral,” is what Musk has promised to do. What happens will become clear with time.