I partially agree with your statement. The natural languages are heavily patterned, while most of the human knowledge is recorded in text, including the descriptions of the properties of various physical objects. The LLMs know that in the sense of having this data and building the relationships between various words, so they do have internal representations of concepts. They obviously cannot have it as personal physical experience in the same way as humans. However, the actions and feedback are included in the reinforcement learning and the user interactions (if the incorrect responses were rated higher than the correct ones, it can lead to issues). Programming languages and study materials tend to be more structured than random texts, so LLMs work better with them. You also do not need from your pair programmer whether they have a cold or if they are hungry or what they think about dogs (unless you are really bored). It's nice if they remind you to stay hydrated, but they do not need to experience it physically themselves (neither does a calendar reminder which is easier to set up). Therefore, they can be fit for the particular purpose. Well, given the wide adoption of Claude, they are. Here is to hope that the developers can understand the code they ship. However, as you've said, LLMs (and humans) can be wrong and cannot be held accountable for their errors (nor can some humans unless you consider Luigi Mangione to be inspirational, but, again, it's a more of a systemic social issue and not directly related to LLMs). Therefore, ideally you would not want either in the decision-making position. I agree that the tools should be fit for purpose and the job market can be challenging to navigate. I personally find filling the application forms with multiple popup menus on an external website to be more annoying, especially when the exact same information is in your CV and they cannot even scrape that correctly. One would hope that it'd discourage competitors, so the resulting pool is lower. Overall, the first rounds of interviews are to find the more suitable candidates and tend to be outsourced to the people who know little of the field you are to be working in. So, using a chatbot at this stage and just reading the summary or watching a video recording is not a bad idea. When you get to the point of the practical exercises and need to explain your logic, that's when you'd want your potential team lead to be present.