Despite all the criticism Elon Musk gets, Twitter didn’t crash and burn when it was bought in 2022 and rebranded as the X platform in 2023. At the end of March 2024, X announced the official release of its large language model and AI chat, Grok-1.5. It’s been in development under X’s sub-corporation xAI since early in 2023.
Marketed as an AI search assistant with a witty sense of humor and a dash of rebellion, Grok is currently only available to X Premium and Premium+ subscribers. So far it’s limited to only a small list of countries. If you haven’t signed up for an X Premium account, formerly known as Twitter Blue, then you won’t get to try out this new feature. Unlike ChatGPT, Bing Chat, and Google’s Gemini, Grok doesn’t have a free version or demo.
What does the word Grok mean?
Elon Musk has cultivated a public persona that aligns with being “one of the geeks.” This image is both self-professed and generally recognized by the public as a significant aspect of his public image. For fans of science fiction, it’s probably no surprise he would use a term from Robert Heinlein’s 1961 novel Stranger in a Strange Land.
How to see sensitive content on X or Twitter
X has made it very easy to remove the sensitive content warning messages
The book is about Valentine Michael Smith, a human born and raised by Martians on the planet Mars, who comes to Earth when he is 20 years old. In the book, the term Grok is used to describe not just a thorough understanding of something, but one that is so deeply understood that it becomes part of yourself, and your natural empathetic intuition.
What makes Grok unique is its deep integration with the X platform
X describes Grok-1.5 as being like JARVIS from Iron Man, mixed with the wit and humor of Hitchhiker’s Guide to the Galaxy. Just like basically every other large language model chat AI, Grok utilizes next-token model weights, predicting what the next most likely word or symbol should be. What makes Grok unique is that it integrates deeply with X, accessing real-time data from the platform.
Most Chat AI’s used to rely entirely on pre-existing datasets, but now have the ability to search the web for you, and provide references to the information they give you. Before becoming X, the platform Twitter was widely regarded as a place you could go to find out real-time news. If you heard about an event that was happening, anywhere in the world that heavily used Twitter, you could search the platform to find out what was happening directly from someone close-by.
In a way, Twitter was considered by many to be a major news source, and on-the-ground reports on current events were usually easy to find. With the rebranding and changes under Musk’s direction, the platform’s importance as a news source has diminished. This can be seen through the decrease in traffic from news publishers, and a general decline in news-based engagement on the platform.
That poses a question, if there is really any reason to pay a subscription fee to access Grok, instead of using the free version of any other popular Chat AI platform.
Groks-1.5’s technical abilities show notable improvements over Grok-1
According to X’s blog announcing Grok-1.5, with the update from version 1 to 1.5, Grok’s performance is getting a lot closer to Gemini Pro 1.5 and GPT-4 in all benchmarks. Notably, Grok has achieved a MATH benchmark score of 50.6%, up from 23.9% in version 1. GPT-4 score of 52.9% isn’t far from that, but Google’s Gemini definitely is far ahead of both, with a score of 58.5%. Grok also performed better in HumanEval benchmarks than GPT-4, but still has a large gap between its capabilities and Gemini’s. HumanEval benchmarks show how well a large language model does with code generation and problem-solving.
OpenAI debuts GPT-4 Turbo with access to world events as recent as April 2023
You can also create custom-tailored versions of ChatGPT now
A major advantage Grok has when compared to GPT-4, another paid subscription, is its capability to handle large amounts of text and long contexts. GPT-4 supports 8,192 tokens, while Grok-1.5 can process 128,000 tokens in a single context window. ChatGPT already has a Turbo version in development that uses the same number of tokens though, so that advantage won’t last for long.
The way you experience 128k tokens when using Chat AI, is that there’s a better memory of what you’ve been talking about within a single chat window. The Chat AI can refer back to something you may have talked about much earlier in the conversation. That allows for better content analysis, more complexity in your interactions, and a better ability to integrate multiple types of data simultaneously. Even with a higher level of tokenization, Grok still doesn’t really stand apart. Gemini 1.5 also supports 128,000 tokens, and ChatGPT is just around the corner.
Grok is still learning and in the early stages of development
Elon Musk’s usual strategy of assembling highly skilled and experienced team members has certainly paid off again. Musk’s selection of AI innovators who have worked with DeepMind, OpenAI, Google, and Microsoft, really accelerated its development. It went from Grok-1’s alpha testing, to Grok-1.5’s publicly accessible beta launch in less than a year. It’s only been in development since early 2023, and has already reached respectable benchmarks.
Copyright: CC BY-NC 3.0
There’s a truth that never changes in the tech industry. Put together a group of highly skilled geeks who are experts in their field, provide them with a lot of money and resources, and they will always succeed. Musk’s direct involvement can’t be ignored though. Grok’s “personality” appears to reflect his public persona, and you have to use his X platform to gain access. One thing is for sure, Grok does have a degree of uniqueness because of this.
Grok is still learning though, and no one should be surprised if it quickly rises to be a direct competitor to the other most well-known platforms, like ChatGPT and Gemini. For now, early testing reports are coming back with mixed results. Right now, Grok isn’t anything special, and has all the same problems that every other autoregressive large language model has. The type of machine learning that uses previous inputs to predict future ones will always suffer from the tendency to deliver hallucinations or factually incorrect information.
New AI developments are moving so fast, it’s hard to keep up. Grok certainly seems like one to keep an eye on. With the backing and massive resources that come from an Elon Musk project, Grok is sure to be contributing to the advancement in Chat AI technologies.
