xAI today launched the Grok-1.5 large language model with improved comprehension capability and token size. This is the first major release since the company introduced Grok in November last year.
The AI company has been planning this release for months and its official also hyped the new performance in prompt responses.
The new Grok 1.5 brings improved reasoning and problem-solving capabilities. Its newest achievement includes fast coding and math-related tasks. The company says the model has scored 50.6 percent on the Match benchmark and 90 percent score on the GSM8K benchmark.
These two benchmarks reflect a wide range of grade school to high school competition problems. Grok 1.5 has achieved 74.1 percent on HumanEval benchmark for code generation evaluation and problem-solving abilities.
In terms of comparison, the Grok 1.5 has been significantly faster than its predecessor in MMLU, MATH, GSM8K, and HumanEval by big numbers in a similar number of attempts.
Grok is inspired by Hitchhiker’s Guide to the Galaxy, a famous fiction novel. The company has implemented the novel’s satirical way in response via two modes – Regular mode and Fun mode.
With this upgrade, xAI is likely to improve the responses on each user prompt.
Comprehension
New Grok 1.5 could process long contexts of up to 128,000 tokens within its context window. This is a result of Grok’s improved memory capacity which now stands 16 times more context length than the previous version.
The new memory helps to use and process the information from longer documents. It can handle longer and more complex prompts while keeping itself aligned with the content length.
Image Credit: xAI
xAI revealed that Grok 1.5 has demonstrated powerful retrieval capabilities for embedded text within the context of up to 128k tokens in length on the Needle In A Haystack (NIHA) platform.
Availability:
Grok 1.5 will be available for early beta testers on social media site X. It will rollout gradually for testers along with new features.
(source – xAI)