OpenAI, the leading artificial intelligence research company, has announced the release of its latest language model, GPT-4. With improved reasoning ability and longer conversation capability, GPT-4 is set to power ChatGPT, OpenAI’s popular chatbot. However, the announcement also comes with heightened safety measures, as GPT-4 has the potential to generate harmful content in response to certain prompts.
Improved Ability & Safety Measures
According to OpenAI researchers, GPT-4 can generate potentially harmful content such as advice on planning attacks, hate speech, and represent societal biases and worldviews that may not be representative of widely shared values. For this reason, OpenAI has implemented safety measures to prevent ChatGPT from producing harmful content in response to potentially harmful prompts submitted by the red team.
OpenAI formed a red team to test for negative uses of the chatbot and added restraints to the chatbot to mitigate harm, but in some cases, it did not completely prevent it. To steer ChatGPT away from behaving in potentially harmful ways, researchers will reward and reinforce the types of responses they want their chatbot to produce while showing it potential unacceptable responses.
Elon Musk’s Criticism & Concerns
Not everyone is happy with these safety measures. Entrepreneur Elon Musk criticized OpenAI for implementing safeguards that prevent ChatGPT from producing potentially harmful responses, particularly ones where it refuses to weigh in on divisive political topics. Musk has explored starting his own AI lab to rival OpenAI, which he co-founded before exiting the company in 2018 over strategy differences.
Musk’s concerns about AI are not unfounded. The rapid integration and development of AI technology has led to an increase in legal risks and ethical concerns. Until there is better regulation and responsible use of AI technology, development should be paused.
GPT-5 Training Update & AGI
In another development, there is a new claim on Twitter by developer Siqi Chen that GPT-5 will complete its training by December and achieve Artificial General Intelligence (AGI). Achieving AGI refers to an AI’s ability to comprehend and learn any task or idea that humans can understand. If this happens, it could lead to the spread of harmful disinformation on social media platforms like Twitter, which Musk has been trying to combat.
Achieving AGI could have a ground-shaking impact on the world and the way we live. It could increase productivity and eliminate tedious work, but it could also have unintended consequences that are yet to be understood. OpenAI has predicted that an intermediate version, GPT4.5, will be introduced between GPT-4 and GPT-5 in 2023.
Flaws of Unregulated AI Development
The risks associated with unregulated progress in AI development include the generation of inaccurate and plagiarized articles and the potential leaking of personal information. Notable figures like Steve Wozniak and Elon Musk have supported a pause in AI development to ensure the safety and responsible use of these systems.
In conclusion, while the announcement of GPT-4’s improved ability and safety measures is exciting news for language modeling and chat applications, it also highlights the need for responsible AI development. The advancement towards AGI should not come at the cost of safety and ethical concerns, and regulation is necessary to ensure that AI technology benefits humanity while minimizing harm.
Image Source: Wikimedia Commons
