Grok

Elon Musk, the controversial owner of X (formerly known as Twitter), has announced that the next iterations of his language model, Grok, are on the horizon. In a series of X threads, Musk revealed that Grok 2 is set to be released in August, with Grok 3 expected by the end of the year. These announcements have generated significant anticipation, as little is currently known about the specifics of these upcoming models.

However, Musk did provide a tantalizing detail about Grok 3: it will be trained on 100,000 Nvidia H100 GPUs, suggesting a substantial leap in capability. “It should be really something special,” Musk commented, hinting at the ambitious scale and potential performance of the new model.

Musk’s revelation of Grok 2’s release timeline came in response to a video featuring Aidan Gomez, CEO of Cohere, who discussed the prevalent practice of training language models on the outputs of OpenAI’s models. Gomez noted that Cohere’s language models, which are not trained on OpenAI model outputs, have been perceived by users as feeling different. Musk echoed this sentiment, highlighting the challenges of purging language models from internet training data and stating that Grok 2 will represent a “giant improvement in this regard.”

For those unfamiliar with the history of OpenAI, Elon Musk co-founded the company but later parted ways due to disagreements. Following the success of OpenAI’s ChatGPT, Musk introduced Grok, a paid language model integrated into X. Grok is designed with an emphasis on humor, aiming to make interactions more lifelike and engaging.

The current version, Grok 1.5, was launched in March, featuring enhanced reasoning capabilities and a context length of 128,000 tokens. While Grok 1.5 did not outperform GPT-4 on benchmarks such as MMLU, MATH, and GSM8K, it was not far behind and even surpassed GPT-4 on the HumanEval benchmark. Despite these advancements, Grok has not achieved the same level of popularity as free alternatives like ChatGPT, Gemini, and Claude. The absence of a free version of Grok has limited its widespread adoption, and with Musk’s focus on increasing X’s revenue, it remains uncertain whether a free version will be offered in the future.

As the AI community eagerly awaits the release of Grok 2 and Grok 3, the potential for these new models to reshape the landscape of language processing is significant. Musk’s ambitious plans and the substantial resources dedicated to training these models suggest that they could bring noteworthy advancements in AI capabilities, further fueling the competition in the rapidly evolving field of artificial intelligence.