Elon Musk’s AI Masterstroke: What to Expect from Grok 2 and 3

Elon Musk, the controversial owner of X (formerly known as Twitter), announced in X threads that Grok 2 will be released in August, followed by Grok 3 by the end of the year. While details about these new language models are scarce, Musk mentioned that Grok 3 will be trained using 100,000 Nvidia H100s, making it “something special.”

Musk disclosed Grok 2’s release month in response to a video featuring Cohere CEO Aidan Gomez, who discussed how many models are “training on the model outputs of OpenAI.” Gomez noted that since Cohere didn’t follow this approach with its LLMs, users reported a different experience.

Agreeing with the video’s sentiment, Musk stated that it requires significant effort to exclude LLMs from internet training data. He emphasized that Grok 2 will be “a giant improvement in this regard.”

Elon Musk Departure From OpenAI

For those unfamiliar with OpenAI’s history, Elon Musk co-founded the company but later parted ways due to disagreements. Following the launch of ChatGPT, Musk introduced a paid LLM to X called Grok, which incorporates humor to enhance its lifelike quality.

The latest version, Grok 1.5, was released in March, featuring improved reasoning capabilities and a context length of 128,000 tokens. Although Grok 1.5 did not perform as well as GPT-4 on the MMLU, MATH, and GSM8K benchmarks, it wasn’t far behind and surpassed GPT-4 on the HumanEval benchmark.

Since X does not offer a free version of Grok, it is less popular compared to alternatives like ChatGPT, Gemini, and Claude. It’s uncertain if a free version of Grok will be available in the future, but considering the costs of running LLMs and Musk’s goal to increase X’s revenue, it’s unlikely we’ll see a free version of Grok anytime soon.