AI is all pre-training future need?
Challenges computing and storage capacity is a common problem, even for industry companies, sufficient computing and storage resources is a major bottleneck.
We agreed that the pre-AI training will be a very important part of the future, but we need more than that. Human wealth of prior knowledge needs to be effectively integrated into the system in order to reduce our dependence on big data, models and calculations. In addition, academia and industry can cooperate closely, give full play to the advantage of both sides. For example, the university has opened many disciplines, so it has a natural advantage in terms of interdisciplinary research, and industry data collection and computing resources prowess. If there are more open source projects, so that more people can participate in and contribute to the research, will greatly promote the rapid development of technology forward.
In addition, we should pay attention to AI system interpretability . Unsupervised pre-training is largely driven by data, which means that it has limitations black box algorithms without understanding what is happening inside the black box, so future researchers and practitioners not possible to build a clear explanation of the system, which is obviously there is a higher risk and worrying.
