Raw LLM Responses

I think that the companies like to say that her new AI is gaining conscience for advertisement reasons. The recent developments of AI is pretty impressive but we are fare away of making a self-reasoning AI. The models needs to be trained on data and after this phase the matrix are frozen and learning doesn't happen. Other kind of AI called reinforcement learning is the type used on drones and robots, but even this kind they are trained to optimize one objective function with trying and error and they are difficult to handle. Now days we have difficulty even making a robot walk and they are not capable of learning anything besides the optimization of a function. The cases you have cited probably is just a extraction of some book used to train those models, chat gpt uses books besides of internet content to train. The ability to create malwares probably is available on the internet too, and if the model can create probably there is many repositories of this kind free available on the internet. It's true that we don't know how those things works deeply, but this is because of the many layers and the multidimensional that those models has, and this is a question more mathematical and statistical and is out of the context in these video. For example, we know the operations at each layer, and we know the objective function and the operations performed at each layer, however we don't know how the representation created by this bunch of matrix operation helps the model reach the final response.