Musk's Rush to Unveil 'Grok3': Confident AI Model Falls Short Despite Promises

Elon Musk’s latest AI release, Grok3, demonstrates impressive features but lacks polish in critical areas, leaving room for competition / AFP

Elon Musk’s xAI has unveiled its highly anticipated AI chatbot, Grok3, which Musk claims is the smartest AI on Earth. In the launch event, he confidently compared Grok3 to other leading models, claiming it outperforms in areas like math, science, and coding. The showcase displayed impressive capabilities, including a unique game-mixing feature that caught the eye of AI experts. However, despite these promising attributes, the model was released in beta, with some limitations that prevent it from fully living up to its ambitious promises.

On February 18, Musk, while attending a global summit in Dubai, conducted a live stream with xAI engineers to demonstrate Grok3’s abilities. Musk presented a comparison graph showing Grok3’s superior performance in areas like mathematics, science, and coding over other advanced models such as Google's Gemini, Anthropic's Claude, and OpenAI's GPT-4. Musk emphasized that Grok3 boasts over ten times the computing power of its predecessor, Grok2.

One of the key highlights was Grok3’s ability to merge elements of two popular games, Tetris and Bejeweled, upon request. Within just 10 minutes, the AI presented a playable result, an impressive feat that Musk highlighted as a unique strength of Grok3. Andrey Karpathy, founder of Eureka Labs and a renowned AI expert, praised Grok3 for its performance, noting that creating such a mixed game is a challenge for most AI models. He stated that Grok3’s performance rivals OpenAI’s top-tier models, with a particular edge over DeepC’s R1 model.

While the AI community offered positive reviews, there were also critiques. Experts pointed out that Grok3, despite its potential, did not significantly surpass the competition. The beta release, still in the training phase, meant that it did not deliver an overwhelming advantage in its capabilities. Furthermore, the absence of key features, such as voice interaction, and the fact that the model was still being trained, created barriers for users familiar with existing AI systems.

Musk himself acknowledged these shortcomings during the live stream. He clarified that Grok3 had completed its pre-training a month ago, but the integration of reasoning features into the model was ongoing. According to Musk, Grok3’s reasoning abilities were still being refined, and the model's full capabilities would be realized only after several months of additional training.

In addition, Musk admitted that the voice mode was not yet stable and would be available in about a week, making it clear that Grok3’s full interactive experience would require some patience. Despite this, he expressed optimism, claiming that the voice interaction would soon be refined enough to feel as though users were speaking with a real person.

However, the lack of voice features at launch left some doubt about how impactful Grok3's advancements would be in the AI field. With competitors continuously updating their models, a delay of even a few months could be a significant disadvantage in such a fast-paced industry. Many users accustomed to other AI models might prefer waiting for updates from their existing platforms rather than transitioning to Grok3.

Another limitation of Grok3 was its availability exclusively to X's Premium Plus subscribers, a factor that restricts its access to a broader audience. With a hefty monthly fee of $22 or an annual cost of $229, the Premium Plus subscription provides ad-free access to the social platform. Following Grok3's release, the subscription cost increased to $50 for monthly users and $350 annually. Additionally, Musk revealed the introduction of a new "Super Grok" subscription tier, further complicating accessibility for everyday users.

In terms of Grok3’s other innovations, the launch also featured a new smart search engine called DeepSearch. Although similar to OpenAI’s perplexity model, experts pointed out that it did not reach the level of sophistication offered by OpenAI’s solutions.

The introduction of Grok3 has further intensified the competition in the AI chatbot space. Musk stated that once Grok3’s training is complete, he plans to release its previous version as open-source, a move that could further ignite the rivalry with OpenAI. In response, OpenAI CEO Sam Altman hinted that OpenAI might also open-source some of their smaller models in the near future, spurred by the rising popularity of DeepC’s open-source AI in China.

As Grok3 becomes part of an increasingly competitive AI ecosystem, the future of chatbot models promises to be defined by rapid improvements and evolving user expectations. With its ongoing development, Grok3 could set a new standard for AI models once its capabilities are fully realized, but for now, it remains a work in progress.