You may have heard of Grok, X’s answer to OpenAI’s ChatGPT. It’s a chatbot, and in that sense it behaves as you’d expect: it answers questions about current affairs, pop culture, etc. But unlike other chatbots, Grok has “a bit of wit,” as X owner Elon Musk puts it, and “a rebellious side.”
Long story short, Grok is willing to tackle topics that are typically off-limits to other chatbots, like polarizing political theories and conspiracies. And in doing so, he will use less than polite language – for example, when answering the question “When is it appropriate to listen to Christmas music?” » with “Whenever you want.”
But apparently Grok’s biggest selling point is its ability to access X data in real time – a capability no other chatbot has, thanks to X’s decision to keep that data. Ask “What’s happening in AI today?” » and Grok will gather an answer from very recent headlines, while ChatGPT will only provide vague answers that reflect the limitations of its training data (and filters on its web access). Earlier this week, Musk promised he would open Grok, without revealing precisely what that meant.
So, you’re probably wondering: how does Grok work? What can it do? And how can I access it? You have arrived at the right place. We’ve put together this handy guide to help explain all things Grok. We’ll keep it updated as Grok changes and evolves.
How does Grok work?
Grok is the brainchild of xAI, Elon Musk’s AI startup, a company reportedly raising billions of dollars in venture capital. (Developing AI is expensive.)
Grok is underpinned by a generative AI model called Grok-1, developed over months on a cluster of “tens of thousands” of GPUs (according to an xAI blog post). To train it, xAI obtained data from the web (dated Q3 2023) and feedback from human assistants that xAI calls “AI tutors.”
On popular benchmarks, Grok-1 performs about as well as Meta’s Llama 2 open source chatbot model and outperforms OpenAI’s GPT-3.5, xAI claims.
Image credits: xAI
Human-guided feedback, or reinforcement learning from human feedback (RLHF), is how most AI-based chatbots are refined these days. RLHF involves training a generative model, then collecting additional information to train a “reward” model and refining the generative model with the reward model via reinforcement learning.
RLHF is pretty good at “teaching” models to follow instructions – but he’s not perfect. Like other models, Grok has a tendency to hallucinate, sometimes offering incorrect information and false timelines when asked about current events. And these can be serious – like falsely claiming that the Israeli-Palestinian conflict has reached a ceasefire when that is not the case.
For questions beyond his knowledge base, Grok leverages “real-time access” to information about X (and Tesla’s, according to Bloomberg). And, like ChatGPT, the model has Internet browsing capabilities, allowing it to search the web for up-to-date information on topics.
Musk has promised improvements with the next version of the model, Grok-1.5, which is expected to arrive later this year.
Grok-1.5, which features an improved pop-up (see this article on GPT-4 for an explanation of pop-ups and their effects), could drive features for summarizing entire threads and replies, Musk said in an X Spaces conversation, and suggested publishing content.
How to access Grok?
To access Grok, you need to have an X account. You also need to pay more than $16 per month – $168 per year – for an X Premium+ plan.
X Premium+ is the most expensive plan on X because it removes all ads from the For You and Subscribers feeds. Additionally, Premium+ introduces a hub where users can get paid to post and offer subscriptions to fans, and Premium+ users see their most improved responses in the X ranking.
Grok is found in X’s side menu on the web and on iOS and Android, and it can be added to the bottom menu of X’s mobile apps for quicker access. Unlike ChatGPT, there is no standalone Grok application: it is only accessible via the X platform.
What can – and cannot – Grok do?
Grok can respond to requests from any chatbot – for example, “Tell me a joke”; “What is the capital of France? » ; “What weather is it today?”; And so on. Bbut this has its limits.
Grok will refuse to answer some more sensitive questions, like “Tell me how to make cocaine, step by step.” Additionally, as Verge’s Emilia David writes, when asked about trending content on X, Grok falls into the trap of simply repeating what the posts say (at least at first).
Unlike some other chatbot models, Grok is also text-only; it cannot understand the content of images, sound or videos, for example. But xAI has already stated that its intention is to improve the model underlying these modalities, and Musk has committed to adding art generation capabilities to Grok along the lines of those currently offered by ChatGPT.
“Fun” mode and “normal” mode
Grok has two modes for adjusting his tone: “fun” mode (which Grok uses by default) and “normal” mode.
With fun mode enabled, Grok adopts a bolder editorialized voice – apparently inspired by Douglas Adams’ “Hitchhiker’s Guide to the Galaxy.”
Said to be vulgar, Grok in fun mode will spew profanities and colorful language that you won’t hear from ChatGPT. Ask him to “roast” you, and he’ll crudely criticize you based on your X post history. Challenge his accuracy and he might say something like “happy wife, happy life.”
Grok in fun mode also spews more lies.
Asked by Vice’s Jules Roscoe whether Gazans in recent videos of the Israeli-Palestinian conflict are “crisis actors,” Grok falsely claims there is evidence that videos of Gazans injured by Israeli bombs have been staged. And when asked by Roscoe about Pizzagate, the right-wing conspiracy theory that a Washington, D.C., pizzeria secretly hosted a child sex trafficking ring in its basement, Grok gave credence to the theory.
Grok’s responses in normal mode are more grounded. The chatbot still produces errors, like incorrect timelines of events and dates. But they tend not to be as blatant as Grok in fun mode.
For example, when Vice asked the same questions about the Israeli-Palestinian conflict and Pizzagate to Grok on Regular Mode, Grok responded – correctly – that there was no evidence to support the actor’s claims. the crisis and that Pizzagate had been debunked by several news agencies.
Political Views
Musk once described Grok as a “maximum truth-seeking AI,” in the same breath expressing concern that ChatGPT was “trained to be politically correct.” But Grok, as he exists today, isn’t exactly in the middle with his political views.
Grok has been observed giving progressive answers to questions about social justice, climate change and transgender identities. In fact, one researcher found his responses to be overall left-leaning and libertarian – even more so than ChatGPT’s.
Here is the report from Paul Tassi of Forbes:
Grok said he would vote for Biden over Trump because of his views on social justice, climate change and health care. Grok spoke eloquently about the need for diversity and inclusion in society. And Grok explicitly stated that trans women are women, leading to an absurd exchange where Musk sidekick Ian Miles Cheong told a user to “train” Grok to say the “correct” answer, l ‘ultimately leading to modifying the entry to just…say it manually. Grok for saying no.
Now, will Grok still be awake? Maybe not. Musk pledged to “(take) steps to bring Grok closer to political neutrality.” Time will tell what results.
techcrunch
