When Chatgpt appeared, that seemed magical, but also scared a little, especially because we didn’t know how he did what he was doing. Not even Openai himself knew it, and in fact created a tool to discover it In May 2023, a few months after its launch. In spite of everything, the situation did not change too much, but in 2025 we have discovered something important: we love to know what the machines are thinking.
Tell me what do you think and I will tell you what model of AI you are. Deepseek R1 Not only did he impress us to be a free, capable and open source reasoning model: he also showed at all times what he was thinking. Although it takes longer to generate the answers, you can see “what is machine at all times” and that has been a peculiar advance.
Human crutches. They point it out In The Washington Postwhere they indicate how this staler model indicates how the question or problem to end up giving the final rspiente. In fact, the language used by showing this thought process is close, because the chatbot introduces crutches such as “wait”, “hmmm” or “ajá”.
Chatbots a little closer. Reasoning models such as Deepseek R1, O1 either Gemini 2.0 Flash Thinking They also have the advantage that they review their answers, which makes them mistake less than traditional chatbots. When “reasoning” also shows a communication that simulates what a human being could use and that makes us look less … distant.
Empathy. The result is that showing us how they think, these models get us to feel some empathy towards them, say experts such as Ethan Mollick, a professor at the Wharton School at the University of Pennsylvania. Others, such as Subbarao Kambhampati, professor of Arizona State University, warned: it is not known if those “chains of thought” that these models show when reasoning really reflect how the information processs the models.
Nothing happens if they wander. This expert also points out that this reasoning chain seems to be too precise, without wandering or erratic moments. Companies that develop these models are noticed that the answer is correct that in the fact that the reasoning process is consistent. For Kambhampati, this anthromorphization “of reasoning should be dispensed with, which is not especially useful.
Chatbots with personality. This curious characteristic of reasoning models seems to show that the tone with which chatbots answers clearly influences our way of perceiving them. When Elon Musk and Xai launched their first version of Grok, it was clear that They were looking for a sarcastic tone That was less “robotic.” The problem is that chatbot was not especially precise in its answers, but that has improved in Grok 2which is now accessible to all users.
The Virtual Avatars of Replika They try to analyze the tone and context of users to answer them in a close way. Chatbots of the future may precisely try to take advantage of all these tools and techniques so that empathy towards machines grow, and Deepseek R1/O1 are a first interesting step in that direction.
Image | Beata_73601 With Midjourney
In Xataka | The EU regulatory obsession raises a world in which AI will have two speeds. And Europe will lose