Grok 3 is supposed to be released soon, but it won’t be as overwhelming as Musk wants us to believe.
Smaller and larger AIs are announced daily, each wanting to outperform the other in various areas. However, this often only happens in certain benchmarks or in more complex tasks such as programming. According to Jan, this makes little difference to the average user.
Nevertheless, Elon Musk describes Grok 3, the new Large Language Model (LLM) of his AI company xAI, as the smartest AI in the world
But what is behind this claim?
Elon Musk is considered one of the richest people in the world, but also one of the most controversial. Among other things, there have been transphobic statements and accusations of anti-Semitism because he said that the Jewish investor George Soros hates humanity,
and the conspiracy theoryabout Soros, and discussions about his attitude tofree speech
Since Musk’s takeover of Twitter (now X), there have been increasing complaints about the platform’s handling of hate speech, fake news and political influence Most recently, Musk also attacked various European heads of state on X personally and is actively interfering in the German federal election campaign . In American politics, he is considered a Trump advisor and is said to occupy the newly created office for process optimization under the new president After Trump’s swearing-in in January 2025, Musk made a gesture at a public event that was widely interpreted as a Nazi salute.
He has also repeatedly interfered in German politics recently. For example, with a highly criticized guest article in the newspaper Welt in favor of the AfD or by providing a platform for the AfD’s chancellor candidate Alice Weidel for her livestream on X, while making multiple false statements.
Grok 3: What’s inside
Musk and some developers from xAI presented Grok 3 in a livestream on X. There was also a mini version and a special reasoning model.
— xAI (@xai) February 18, 2025
According to Musk, the AI should be scary smart
, so smart that it is frightening. Accordingly, Grok 3 is said to have better logical conclusions, more computing power and a higher adaptability than ChatGPT-4o.
This means that the model calculates faster, understands complex relationships more precisely and can react even more flexibly to different questions.
According to xAI, Grok 3 is capable of the following:
- The AI was allegedly trained with the Colossus supercomputer (100,000 Nvidia H100 GPUs).
- 200 million GPU hours are said to have been spent on training – ten times more than for Grok 2.
- The model was trained on artificially generated data. This
synthetic data
is intended to ensure a diverse and controlled dataset and reduce privacy concerns. - The developers have used
reinforcement learning
andenforcement learning, a machine learning method in which the model evaluates and improves its output based on a reward function. - In addition,
Reinforcement Learning with Human Feedback
(RLHF) was applied: Here, real people evaluate the output to refine the reward function and further improve the quality of the AI content. - Contextual training is designed to ensure that the AI better understands and adapts answers in context.
- The LLM should be able to correct itself by analyzing and comparing answers with facts.
- DeepSearch is an advanced research and language function that will be added later.
The big goal: Fewer hallucinations and greater logical accuracy.
What are hallucinations?Hallucinations in LLMs are false or invented information. This is a known weakness of all common chatbots and the main reason why you should always question AI-generated answers.
When is Grok 3 coming? In the US, some users should already be able to use Grok 3. First up are paid subscribers ($40/month). However, an exact timeline is missing.
In the EU and UK, Grok 3 is not available for the time being because xAI has to make adjustments to comply with EU regulations.
Grok 3: The smartest AI in the world?
What makes Grok 3 better than other LLMs? According to the benchmarks that xAI showed in the livestream, Grok 3 is particularly better than the competition at logical tasks such as mathematics, programming and scientific questions.
However, xAI refrained from presenting benchmarks from other areas – more on that in a moment.
Even if Grok 3 is supposed to produce fewer hallucinations, that doesn’t mean that this goal has already been achieved. OpenAI also relies on self-correction, synthetic data, and reinforcement learning with human feedback (RLHF).
How good the AI really is will only be known when independent users can test it.
Questionable statements in the livestream
Some statements by Musk and his team could be misleading or misleading. That’s why we want to address them here:;
Ultimate Truth-Seeking AI
: Musk claimed during the stream that Grok 3 is the ultimate truth-seeking AI, which sometimes contradicts what is politically correct. It is difficult to say whether he is implying that other developers are deliberately embellishing facts, or whether he is actively seeking to present facts or backgrounds differently, or whether it is pure political provocation.
The AI is getting better every day: Musk claims that the model is being improved every day. If that meant minor adjustments, that would be fine. However, we want to make it clear that the training of an LLM finally ends before publication – after that, only minor adjustments are made.
Remarkable development time: Musk emphasized that xAI only started development in 2023, while other companies have been working on LLMs since 2019. He did not mention that the real breakthrough lies not in development time, but in huge amounts of data and computing power. It is therefore relatively easy for financially strong companies to quickly develop their own AI models.
Benchmarks only in three areas: How well Grok 3 performs in other areas, such as linguistic tasks, remains open. This is because Musk and his team have only shown benchmarks of logic tasks.
The designation world’s smartest AI
is therefore more marketing than verifiable fact. At least Musk and his team are not showing any groundbreaking innovations that would somehow justify this superlative.
Even if Grok 3 were currently at the top, new models from OpenAI, Google and Anthropic are certainly already in development.
At the World Government Summit (February 11-13), Musk said that this could be the last time that other AIs are better than Grok.This remains to be seen.
