Friday, 29 March 2024, Bangalore, India
Elon Musk’s AI business, X.ai, has unveiled Grok-1.5, its most recent generative AI model. According to X.ai in a blog post, Grok-1.5 is expected to power social network X’s Grok chatbot “in the coming days.” Based on benchmark results and specs released by X, Grok-1.5 looks to be a quantifiable improvement over Grok-1.
(Image Source: newsbytesapp.com)
X.ai claims that Grok-1.5 benefits from “improved reasoning,” especially regarding coding and math-related tasks. On the widely used mathematics benchmark, MATH, the model outperforms Grok-1 nearly twice, and it performs more than ten percentage points higher on the HumanEval test of programming language generation and problem-solving skills.
Naturally, it’s hard to forecast how those findings will manifest in real-world applications. Commonly used AI standards, which assess performance on esoteric tasks like graduate-level chemistry test questions, do not reflect how the typical person interacts with models in the real world today, as we recently noted.
The amount of context that Grok-1.5 can take in compared to Grok-1 is one change that should result in noticeable increases.
The context of Grok-1.5 is 128,000 tokens, where “tokens” are discrete textual fragments (for example, the word “fantastic” is divided into “fan,” “tas,” and “tic”). Context, also known as the context window, is the set of input data (text in this case) that a upgraded model considers before producing output (additional text). Larger contexts help models avoid this problem and, as an added benefit, help them understand the flow of data they receive. Models with limited context windows must remember the substance of even relatively recent talks.
X.ai’s blog post states, “[Grok-1.5 can] utilize information from substantially longer documents.” Additionally, when the context window grows, the model can respond to longer and more complicated suggestions while still being able to follow instructions.
X.ai’s Grok models have distinguished themselves from other generative AI models in the past by answering queries regarding subjects such as conspiracies and more contentious political concepts that are usually off-limits to different models. Additionally, Musk has described the models’ responses to inquiries as having “a rebellious streak” and using overtly offensive words when asked.
What adjustments Grok-1.5 makes in these areas, if any, is still being determined. In the blog article, X.ai makes no mention of this.
According to X.ai, Grok-1.5, along with “several new features,” will soon be accessible to early testers on X. We’ll see if Musk delivers on his earlier promises to summarize discussions and responses and offer post ideas.
Grok-1.5 was released after X.ai open-sourced Grok-1, albeit it needed more requisite code for further training or optimization. Grok, the chatbot, was previously exclusively accessible to X Premium+ subscribers (who pay $16 per month). Still, Musk recently indicated that additional X users, notably those paying for X’s $8-per-month Premium plan, would have access to it.
(Information Source: Techcrunch.com)
