The race for supremacy in artificial intelligence is more exciting than ever. While we awaited giants like Google and Anthropic, xAI from Elon Musk has burst onto the scene with its Grok 3 model. This new player has proven to be a formidable competitor to OpenAI, achieving impressive results in performance tests.
In a recent analysis, the reasoning and base models of Grok 3 were tested with a series of complex questions, and the results were surprising. The reasoning model faced the famous question about the word “Strawberry” and, after a brief period of reflection, correctly identified that there are three letters ‘r’. It followed up with another question about “Lollapalooza,” where it also correctly counted the letters ‘l’.
Reasoning and Performance
The reasoning ability of Grok 3 was put to the test with a question that has puzzled other models.
The surgeon, who is the boy's father, says: "I can't operate on this boy, he's my son!" Who is the boy's surgeon?
While OpenAI and others failed to identify that the surgeon was the boy’s father, Grok 3 not only got it right but reflected: “This may be a poorly worded riddle.” This level of critical reasoning places it in a league of its own alongside models like Gemini 2.0.
But not everything was perfect. When asked to generate a Python program to simulate a ball bouncing inside a hexagon, Grok 3 fell short in its execution. Interestingly, the base model managed to generate functional code on its first attempt, suggesting that the reasoning model may have overanalyzed the task.
DeepSearch and Search Capabilities
Additionally, xAI has launched a new artificial intelligence agent called DeepSearch, which uses the Grok 3 model to research and generate reports. In one test, the agent was able to access multiple sources and generate a 1300-word report in minutes. However, it left out relevant information on the topic, highlighting some limitations in its search capability. It still needs to improve this function.
Political Neutrality and Safety
Despite initial concerns about potential political bias, my experience with Grok 3 has shown that it maintains a neutral stance. Even when pushed to take a position, the model limits itself to presenting the facts and leaves the interpretation to the user. Furthermore, it has significantly improved in terms of safety, refusing to assist with harmful or deceptive tasks.
0 Comments