Uncovering the Dark Side: Why GPT-5 Must Be Halted

Table of Contents

1. Introduction
2. The Call for a Pause in AI Development
   • Concerns about AI's Risks to Society
   • Lack of Planning and Management in AI Development
   • The Problem of AI-generated Propaganda and False Information
3. The Impact of AI on Jobs
   • Study on the Effects of Language Models on the Labor Market
   • Jobs Most Exposed to AI Disruption
   • Jobs with Lower Exposure to AI Impact
4. The Open Letter's Proposal
   • A Call for a Six-Month Pause on Training AI Systems
   • The Need for Public and Verifiable Measures
   • Government Intervention in the Absence of a Pause
5. Refocusing AI Research and Development
   • Making AI Systems More Accurate, Safe, and Transparent
   • The Importance of Understanding Large Language Models
   • Enforcing Transparency and Citing Sources
6. The Future of AI and the Need for Action
   • The Advancements in AI Models
   • Concerns about the Cost of Training Larger AI Models
7. Conclusion

The Call for a Pause in AI Development

Artificial intelligence (AI) has become a topic of concern among over 1300 tech industry leaders, researchers, and others who believe that a pause in its development is necessary to assess the associated risks. In a recent open letter by the Future of Life Institute, prominent figures such as Elon Musk, Steve Wozniak, and the CEO of Stability AI expressed their support for the pause of giant AI experiments. This call for a pause reflects the belief that AI systems with human-like intelligence can pose profound risks to society and humanity as a whole. Despite this acknowledgment, there has been a lack of planning and management in the development of advanced AI, leading to a race to develop ever more powerful but unpredictable AI systems. This article delves into the concerns raised by the open letter and explores the impact of AI on jobs, ultimately advocating for a refocusing of AI research and development towards building more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal systems.

Concerns About AI's Risks to Society

The open letter highlights the potential risks associated with AI systems that possess human-like intelligence. Extensive research has demonstrated the profound impact advanced AI could have on the course of life on Earth. Therefore, it is crucial to plan for and manage the development of AI with utmost care and resources. Unfortunately, the Current level of planning and management falls short, as AI labs continuously engage in a race to Create and deploy increasingly powerful but incomprehensible digital minds. The danger lies in the fact that even the Creators of these AI systems cannot understand, predict, or reliably control their behavior. This raises critical questions about the consequences of allowing such AI systems to flood news and social media platforms with propaganda and misinformation.

The Problem of AI-generated Propaganda and False Information

Generative models powered by AI, such as the GPT series, have gained significant popularity. However, these models can suffer from hallucinations, wherein they provide coherent answers that may take a wrong turn and result in nonsensical or false information. OpenAI's research indicates that while GPT-4 has shown improvements in factual accuracy compared to its predecessors, it is still far from being completely reliable. In fact, approximately one in every four to five answers generated by GPT-4 is factually incorrect. Despite this, over 100 million people began using this AI-powered service only two months after its release. This raises concerns about the potential spread of misinformation and the need to address the issue of control over AI-generated content.

The Impact of AI on Jobs

A significant aspect of the concerns surrounding AI development relates to its potential impact on job markets. A study conducted by OpenAI investigates the effects of large language models like GPT-4 on the labor market in the United States. The study reveals that approximately 80 percent of the U.S. workforce will have at least 10 percent of their tasks affected by these language models. Furthermore, 19 percent of workers could see at least 50 percent of their job being impacted. When considering software and tools like chat interfaces, image generators, and speech-to-text, the number of job-related tasks affected by AI increases to approximately 50 percent.

Jobs Most Exposed to AI Disruption

The study conducted by OpenAI also provides Insight into which jobs are most vulnerable to the impact of AI. Jobs that require adherence to strict sets of rules, such as tax preparation, accounting, mathematics, and administrative assistance, are at higher risk. These rule-heavy jobs can potentially be made easier and more productive through the assistance of large language models like GPT-4, which can be taught to follow these rules reliably. Conversely, jobs with lower exposure to AI impact involve a greater degree of qualitative human interpretation. Survey researchers, public relations practitioners, marketing strategists, traffic designers, and financial and investment managers are examples of these positions that require a human element in decision-making processes.

The Open Letter's Proposal

In response to the concerns and findings regarding AI's risks and impact on jobs, the open letter proposes a six-month pause on the training of AI systems more powerful than GPT-4. This pause, which should be public and verifiable, aims to bring key actors together to evaluate the path forward. If a swift and voluntary pause cannot be achieved, the letter suggests that governments should intervene and institute a moratorium on AI development. The goal is not to halt AI development altogether but to redirect research and development efforts towards improving the accuracy, safety, interpretability, transparency, robustness, alignment, trustworthiness, and loyalty of existing AI systems.

A Call for Refocused AI Research and Development

The open letter emphasizes the need to shift the focus of AI research and development towards enhancing the capabilities of current state-of-the-art systems. This refocusing entails making AI systems more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal. The goal is to increase public faith in AI technology and alleviate concerns about its potential risks. A key aspect of achieving this goal is gaining a better understanding of large language models, like GPT-4, which currently operate as black boxes, making their decision-making processes less transparent.

Enforcing Transparency and Citing Sources

Addressing the lack of transparency in AI systems, the open letter suggests enforcing openness across all AI applications. Individual apps should clearly indicate the information being used in each prompt, similar to how Apple's App Store displays data access permissions. Users should have the ability to toggle access to different pieces of their data, enabling them to control the AI's use of personal information. Additionally, AI-powered applications should provide clear explanations of how they modify Prompts generated by users. Similar to the way autocorrect can be turned on or off, users should have the option to specify whether the AI system should use their original prompts or attempt to modify them. Moreover, AI-generated outputs should be accompanied by proper citations to ensure transparency and enable further scrutiny.

The Future of AI and the Need for Action

The article acknowledges the advancements made in AI models, such as the transition from GPT-3 to GPT-4, highlighting the technical complexities involved. OpenAI possesses expertise in making Incremental advancements, but the leap from one model version to the next encompasses numerous intricate changes, spanning data organization, training optimization, architecture, and more. Although GPT-4 exhibits advanced capabilities, it also possesses the potential to cause substantial harm when combined with appropriate plugins, prompts, and jailbreaks. Therefore, rather than completely limiting training, the focus should be on comprehending the workings of large language models and striving for improvements in accuracy without resorting to the development of significantly more powerful AI models. Increased transparency, responsible data usage, and citing of sources are crucial steps in enhancing AI systems' trustworthiness and minimizing their shortcomings.

Conclusion

The open letter advocating for a pause in the development of AI systems beyond GPT-4 highlights the concerns surrounding AI's risks to society and the potential disruption it may cause in job markets. The proposed pause aims to address these issues, promote transparency, and redirect AI research and development efforts towards making existing AI systems more accurate, safe, interpretable, transparent, robust, aligned, trustworthy, and loyal. The ongoing advancements in AI models and impending developments in hardware, such as Nvidia's H100 chips, emphasize the need for swift action. It is essential to strike a balance between progress and responsible development, ensuring that AI benefits society while mitigating potential risks.