Friday, March 29, Elon Musk (Elon Musk) posted on the X website that an updated version of the Grok-1.5 AI chat program will be released next week. He added that Grok 2 should surpass current AI by all standards. Which is currently being trained to be smarter
Yesterday, xAI announced Grok-1.5, an improved version of its chatbot. He will come with the ability to think. It supports a context length of 128,000 characters that the chatbot can process, which will be released soon.
Grok-1.5 can handle up to 128,000 message tokens, where tokens represent the number of raw bits of text. The model takes the text in the input window and takes it into account before generating the output text. If a model with few window inputs receives text, it will likely forget the content of the last conversation. Models that can receive a large number of messages avoid this error. Because it can better understand the flow of incoming information.
In addition, Grok-1.5 received improvements regarding encoding. Mathematics-related work has demonstrated scores for indicators including MMLU 81.3% for measuring understanding of complex human language covering 57 subject areas, MATH 50.6% for mathematical calculations, GSM8K 90% for solving mathematical problems, both problems and equations, ranging from primary school level to high school and HumanEval . 74.1% are used to evaluate the behavioral correctness of code generated by large-scale language models. Which when looking at the big picture has similar efficiency (not too far off) Gemini Pro 1.5 and GPT-4
Early last month Musk announced that xAI would open source Grok, weeks after OpenAI was sued for abandoning its original mission to benefit humanity. Through cooperation with Microsoft. This causes the company to have to maximize profits for Microsoft. In November, Musk launched the first version of Grok powered by xAI, looking to compete with OpenAI and Alphabet (Google).