It is undeniable that artificial intelligence technology is rapidly developing and is now being used in various sectors and industries. However, one of the most significant issues that has not yet been resolved is related to the problem of hallucinations, or when artificial intelligence “creates” its own information that is inaccurate and not based on any real data.
Jensen Huang, CEO of NVIDIA, said that to overcome this issue of hallucinations, companies developing AI technology will need two things, namely time and higher processing power.
He expects that for now, the issue of hallucinations is inevitable because most AI systems being built today are still going through the “learning” phase where there are many information gaps that need to be filled, and until these gaps are filled with true information, the issue of hallucinations will continue to occur.
More powerful computing capabilities are also needed to reduce the probability of hallucinations occurring. With more powerful processing power, AI systems can search for and process information faster and more efficiently.
Jensen Huang says that the learning cycle of an AI system consists of three parts. The first is pre-training, where the AI system takes in various data and information to learn. The second is post-training, where the system will use the data and information it has been trained on to further deepen its knowledge through human feedback, reinforcement using AI, synthetic data generation and so on.
The final phase is what is called Test Time Scaling, where the artificial intelligence system will start to “think” on its own. In this phase, the AI system will start to look at a problem in several different parts and try to solve it by trying several simulated solutions.
This is different from existing AI systems that can give you answers to questions that already have answers, and this is the future of artificial intelligence technology that is being built now.
Because of this, Jensen says that at the time when the solutions offered by artificial intelligence technology are still in the training and development phase, where a lot of time and computing power are still needed.
This issue of data hallucinations by artificial intelligence systems seems to be something that many people in the AI industry are aware of. If the words of the NVIDIA CEO are true, it seems that the development of this technology needs to continue if we really want artificial intelligence to become something that everyone can benefit from.