xAI is training Grok 2 for AI performance boost

Artificial Intelligence firm xAI just announced Grok 1.5 and now the company shared its plans for Grok 2 which is already under training for performance improvements.

Grok 1.5 is the latest model from xAI and it comes with improved understanding and advanced reasoning for long context. This helps the model respond better to each user prompt with its new 125,000 tokens length.

The new model follows the Grok-1 open-source release. The company has made the model weight and network architecture public for individuals and entities under an Apache 2.0 license.

Grok 1.5 has brought new long context processing with up to 128,000 tokens within its context window. It increases up to 16x memory capacity.

It shows better benchmarks on MMLU, MATH, GSM8K, and HumanEval. The new Grok has scored 81.3 percent, 50.5 percent, 90%, and 73.1 percent respectively on these platforms.

These numbers are improved compared to Grok 1.0 released in November 2023. However, competitors such as Google’s Gemini Pro 1.5, GPT-4, Claude 3 Opus, and Claude 3 Sonnet still maintain an edge over Grok 1.5 on benchmarks.

The new Grok version is built on a custom distributed training framework based on JAX, Rus, and Kubernetes. It enables the xAI team to test new ideas and train architecture at scale with minimal developments.

To advance further in AI performance, Elon Musk, founder of xAI said today that “Grok should exceed current AI on all metrics.” he wrote on X social media, formerly known as Twitter on Friday.

Musk also confirmed that Grok 2 is now training to prepare for future release. However, xAI didn’t share anything about the Grok 2 release date.

Meanwhile, Grok-1.5 will be available to early testers on X starting next week. The company is saying that there could be new features coming with this new release when it rolls out.