Can AI agents think?

I bet you have some experiences like this: typing in the prompt box, waiting for the “thinking” process, then receive the respond. In some IDEs like cursor, users can choose the thinking model for better results. However, if we look closer, we can notice something weird happening in this process.

For example, sometimes I will see the context like “don’t give the user A because of B…”, with B being shown immediately after this. It is ironic because that means either AI didn’t follow its thoughts, or generated new thoughts in 0.01 seconds, or maybe it simply pretended it was thinking. Another example will be articles composing and web searching. In some scenarios, AI will put “as requested” or “according to the searching” in responds, without actually doing it.

To conclude, I do not believe AI can really think, at least not in the way we people do. There is still a lot of topics that we can discuss like CoT or confidence. So stayed tuned and hopefully we can figure out more about this. s