Technology

Grok: Musk’s xAI ships a chatbot four months after founding — reads X in real time, answers with “a rebellious streak”, and costs $16 a month via Premium+

Elon Musk announced that Grok, built by xAI in two months of training, will roll out to X’s new $16-a-month Premium+ tier after a limited beta. Its pitch: live access to X posts, web search, and a willingness to answer “spicy” questions other chatbots refuse, with humour. xAI says it beats GPT-3.5 on benchmarks and trails GPT-4. What it is, what it is not, and why the data source is the point.

Grok: Musk’s xAI ships a chatbot four months after founding — reads X in real time, answers with “a rebellious streak”, and costs $16 a month via Premium+

In July Elon Musk founded xAI with a website and a mission statement. On Saturday it announced a product. Grok — the word is from Robert Heinlein’s Stranger in a Strange Land and means to understand something completely — is a chatbot that will be offered inside X, first to a small group of American testers and then to subscribers of a new Premium+ tier at $16 a month, the top of the three paid plans Musk introduced last month.

What xAI says it is

Three claims. First, real-time knowledge: Grok reads X as it happens, so a question about this morning’s news gets this morning’s answer, where ChatGPT’s knowledge stops at its training cutoff — and it also searches the web. Second, personality: it is “designed to answer questions with a bit of wit”, has “a rebellious streak”, and will take on the “spicy” questions that other chatbots decline. Musk’s example was a step-by-step recipe for cocaine, which Grok answered with a joke and a refusal — the point being the tone, not the chemistry. Third, competence: xAI says the model behind it, Grok-1, was trained in two months, has 33 billion parameters, and beats GPT-3.5 and Meta’s Llama 2 on standard benchmarks while trailing GPT-4 — a fair and, for two months’ work, remarkable position.

What it is not

Not available outside the United States, not free, and not yet independently tested; every number above is xAI’s. It is also not the first “uncensored” chatbot, only the first with a distribution channel of half a billion users. Where it will actually differ from ChatGPT is not the wit but the data: Grok is the only frontier model with licensed, live access to X’s posts, the corpus that Musk closed to everyone else when he ended free API access this year. A chatbot that knows what people are saying right now is a genuinely new thing, and a chatbot that treats X as the truth about what people are saying is a genuinely worrying one.

The company

xAI is led by Musk with a founding team from DeepMind, OpenAI, Google Research, Microsoft Research and Tesla, and advised by Dan Hendrycks of the Center for AI Safety. It is legally separate from X but, as Grok shows, built to feed it.

From Bangladesh

X is small here and Premium+ smaller; Grok will be a curiosity in Dhaka for a year at least. Its Bangla, based on what X’s Bangla corpus looks like, will be an interesting experiment in learning a language from its angriest speakers. The thing to watch is the tier: Musk has said Grok will be a reason to pay for X, and the price of a chatbot that lives in a social app is the price of the app.

What happened next

Two developments are worth recording against the announcement, because both cut against how it was read at the time.

In March 2024 xAI released the weights of Grok-1 under an open licence, making it one of the largest openly available models of the period — an unusual move for a company whose product was sold as a subscription perk.

And the paywall did not hold. Grok was later opened to free X users with usage limits, which is the normal trajectory: a chatbot bundled to sell subscriptions becomes a chatbot given away to hold users, once competitors give theirs away too.

Source: TechCrunch, via Prothom Alo.

Source: Prothom Alo

Written by

Tech BD

Tech BD is an independent technology publication based in Dhaka, published in English and Bangla. Articles appear under the masthead rather than under individual bylines; corrections and source questions reach the editor at info@tech-bd.com.