ChatGPT Creators to Build Safe AI Systems: Here’s How

ChatGPT creator OpenAI has stated that they were looking to make ChatGPT a safer experience. The company has stated that since AI tools come with a certain amount of risk, they were working on building safe AI systems. Recently, Future of Life Institute floated an open letter urging people to pause all major artificial intelligence experiments that are bigger than GPT 4. The letter stated that AI labs in current times were in a “out-of-control race” to develop and deploy digital minds that could potentially be out of control in terms of understanding, predicting and even controlling.

OpenAI had stated in the past that it may be important to get independent review before starting to train future systems, and for the most advanced efforts to agree to limit the rate of growth of compute used for creating new models. In the same vein, the organisation has now shared details on its approach to safety. “We also recognize that, like any technology, these tools come with real risks—so we work to ensure safety is built into our system at all levels,” said OpenAI.

How ChatGPT Creators are Ensuring Safe AI systems

The organisation is working on the following aspects to ensure safe AI systems:

Testing: OpenAI says that AI systems are subjected to rigorous testing through feedback given from external experts. The model is regulated and trained through reinforcement learning with human feedback, and built on broad safety and monitoring systems.

Applying real-world use to improve: The tool is being improved through real life experiences and OpenAI says: “Iterative deployment has helped us bring various stakeholders into the conversation about the adoption of AI technology more effectively than if they hadn't had firsthand experience with these tools”.

Children will be protected: Users should be over the age of 18 years to use ChatGPT, and those over the age of 13 years will need parental approval.

Personal information removed: Individual’s information is removed from the training dataset where feasible, and the models have been fine-tuned to reject requests for personal information of private individuals, and respond to requests from individuals to delete their personal information from our systems.

Improving factual accuracy: The company is leveraging the feedback received from users, and says that GPT-4 is 40% more likely to produce factual content than GPT-3.5.