X.ai, Elon Musk’s AI startup, has announced their latest generative AI model, Grok-1.5. According to a recent blog post, this cutting-edge technology is expected to power the social network X’s Grok chatbot in the near future – specifically, “in the coming days.” Grok-1.5 boasts significant improvements over its predecessor, Grok-1, which is evident from the benchmark results and specifications released by X.
The team at X.ai explains that Grok-1.5 has “enhanced reasoning abilities,” particularly when it comes to tasks involving coding and mathematics. This model has surpassed Grok-1’s score on a popular mathematics benchmark, known as MATH, and has also shown a remarkable improvement of more than ten percentage points on the HumanEval test – a measure of programming language generation and problem-solving capabilities.
Of course, while these results may be impressive on paper, it is difficult to predict how they will translate into real-world usage. As experts in the field have pointed out, commonly-used AI benchmarks often fail to accurately capture the average person’s experience with models today – even those that measure performance on highly specialized topics such as graduate-level chemistry exam questions.
However, one notable improvement that Grok-1.5 offers is its increased ability to consider context compared to Grok-1. This model can now process a 128,000-token context – “tokens” being smaller units of raw text, such as breaking the word “fantastic” into “fan,” “tas,” and “tic.” In AI terminology, “context” or “context window” refers to the input data (in this case, text) that a model takes into account before generating an output (more text). Models with smaller context windows often struggle to remember previous conversations, even those that took place just a short while ago. In contrast, models with larger context windows overcome this challenge and, as an added benefit, have a better understanding of the flow of information they receive.
“Grok-1.5 is capable of utilizing information from substantially longer documents,” states the X.ai blog post. “In addition, the model can handle longer and more complex prompts while also maintaining its ability to follow instructions as its context window expands.”
What sets X.ai’s Grok models apart from other generative AI models is their ability to respond to questions on topics that are typically off-limits to other models, such as conspiracy theories and controversial political ideas. Additionally, these models are known for their “rebellious streak,” as described by Musk himself, and can even use rude language if specifically requested to do so.
It remains to be seen if Grok-1.5 brings any changes in these areas. The X.ai blog post does not mention any updates in this regard.
According to X.ai, Grok-1.5 will soon be available for early testing on X, along with “numerous new features.” Musk has previously hinted at the possibility of summarizing threads and responses, as well as suggesting content for posts. We will have to wait and see if these features are indeed included in the upcoming release.
The reveal of Grok-1.5 comes after X.ai open sourced Grok-1, although the necessary code to fine-tune and further train it was not released. More recently, Musk has announced that Grok, the chatbot, will now be accessible to a wider user base on X – specifically, those subscribed to the Premium plan for $8 per month. Previously, this feature was only available to X Premium+ customers, who pay $16 per month.
