Last Updated on
Grok AI is now accessible to all X Premium+ subscribers in the US. xAI first announced Grok on the 4th of November, 2023. Initially touted as a witty alternative to OpenAI’s ChatGPT with a “rebellious streak”, xAI’s Grok system is powered by the Grok-1 frontier LLM in the same way that ChatGPT is powered by GPT-4. With different proprietary large language models, the capabilities (and humor) of these flagship AI systems vary. So what can we expect from Elon Musk’s AI chatbot?
What is Grok AI?
Elon Musk founded xAI in 2023, in response to the global technological (and economic) boom of artificial intelligence. Its first software product, Grok, is an AI chatbot powered by an LLM (large language model) called Grok-1. Grok joins a luminary class of artificial intelligence comprised of Google Bard, Bing Chat, Anthropic Claude, and Meta AI. Other tech giants such as Amazon and NVIDIA do own proprietary LLM technology, but do not have a public-facing AI chatbot, and are instead focusing on B2B technology.
Grok is inspired by The Hitchhiker’s Guide to the Galaxy, which first appeared on a 1978 BBC Radio show of the same name. Most will know of this piece of well-loved media from the 1979 science fiction novel, or by having watched the 2005 feature film.
When and where can you use Grok AI?
Currently, only verified users of the 𝕏 platform can access Grok AI. Musk’s new AI firm and the social media platform formerly known as Twitter are inextricably linked, with xAI training its AI chatbot on public data from X.
Public rollout to X Premium+ subscribers is complete in the US, with UK users coming next.
This access will include using Grok as a standalone tool (presumably via web browser) and also via the X social media platform, where Grok will be an integrated AI chatbot.
Essential AI Tools
7-in-1 AI Content Checker – One-click, Seven Checks
Winston AI detector
Originality AI detector
How is Grok AI different to ChatGPT?
The “journey to Grok-1” began with Grok-0, the earliest iteration of the xAI foundation model. A foundation model is an AI model that will be used as the basis of other, more specialised models. It’s both expensive and scientifically challenging to get an AI model trained and weighted appropriately, and so the efficient strategy being adopted by big tech is to get it right once, and then adapt that successful model to perform specific tasks afterwards.
|GPT-3.5 or GPT-4
Speaking in greater depth about this technology, the xAI blog explains the progress it has made since August:
“After announcing xAI, we trained a prototype LLM (Grok-0) with 33 billion parameters. This early model approaches LLaMA 2 (70B) capabilities on standard LM benchmarks but uses only half of its training resources. In the last two months, we have made significant improvements in reasoning and coding capabilities leading up to Grok-1, a state-of-the-art language model that is significantly more powerful, achieving 63.2% on the HumanEval coding task and 73% on MMLU.”
HumanEval and MMLU (Massive Multitask Language Understanding) are two of several specialist benchmarks that AI firms are using to objectively compare software products with each other. This is not dissimilar to how graphics card manufacturers market the results of benchmarking software such as Heaven UNIGINE, 3D Mark, and PassMark.
What is Grok-1? — The Elon Musk LLM
Grok is an AI chatbot. Grok-1 is an LLM (large language model). The most common analogy you’ll find for this dynamic is that Grok is the car, and Grok-1 is the engine. You, the end user, interface with the car, which provides all the ergonomic controls you need. However, it’s pretty useless without the engine. The difference between being an end user and a developer could be compared to being a driver and a mechanic respectively — far more people are drivers, because being a mechanic requires a high level of technical expertise in a specific field. The analogy breaks down when we start talking about how developers use an API (short of a mechanic taking the engine out of one car and hotwiring it into another), because now we’re into hardware and software concepts. Still, it’s a useful analogy, if an imperfect one.
In the same vein, ChatGPT is powered by GPT-3.5 or GPT-4, Bing Chat is powered by GPT-4, Google Bard is powered by PaLM 2, and Anthropic Claude is powered by Claude-2. This is the most likely-to-be-out-of-date sentence in this article, because each of these AI bots are updated regularly and asynchronously, so bear with.
While still in early beta, Grok-1 has surpassed “all other models in its compute class, including ChatGPT-3.5 and Inflection-1. It is only surpassed by models that were trained with a significantly larger amount of training data and compute resources like GPT-4.”
Grok 1.5 predicted release window
xAI in confirmed to be working on the next version of Grok AI. Grok 1.5 “should be out next month with substantial improvements across the board,” according to CEO Elon Musk. The new update should solve the issue of the AI chatbot citing scientific sources that do not exist — a fault to which it is apparently “just as prone as ChatGPT”.