Exposing OpenAI's Tiered System: The Highest Level Boasts Abilities Beyond Human Potential

Last Thursday, OpenAI, the parent company of the AI chatbot ChatGPT, introduced a new AI grading system for its employees. This system categorizes AI capabilities into five distinct levels, with the highest tier designed to surpass human abilities and handle complex organizational tasks. Currently, ChatGPT is classified at the lowest level, Level One.

According to a spokesperson, the new grading system consists of five levels:

1. Chatbots - Capable of engaging in natural language conversations.

2. Reasoners - Ability to solve complex human-level problems.

3. Agents - Can take actions on behalf of users.

4. Innovators - Assists in creating innovations.

5. Organizations - Executes complex organizational tasks more effectively than humans, including capabilities such as predicting the future and interacting with animals.

OpenAI executives informed employees that the company currently operates at Level One but is nearing Level Two (Reasoners), which refers to systems that can perform basic problem-solving tasks, akin to individuals with a PhD who lack the ability to use tools.

During a company-wide meeting, OpenAI leadership also showcased a research project involving their AI model, GPT-4. They believe this model demonstrates new skills approaching human reasoning levels. OpenAI has been internally testing these features, a common practice within the industry.

For years, OpenAI has been devoted to developing Artificial General Intelligence (AGI), which aims for AI systems to outperform humans in a variety of tasks. However, this advanced AI system has yet to be realized. Sam Altman, the CEO revered as the 'father of AI,' has previously predicted AGI could be achieved within the next decade.

According to the grading system proposed by OpenAI, the level corresponding to AGI is Level Three, where AI systems will take actions on behalf of users over several days. Researchers have long debated the criteria for determining when AGI is achieved. In a November 2023 paper, researchers from Google DeepMind presented a five-level AI grading framework, likened to the automotive industry's assessments of self-driving car automation levels.

OpenAI’s new grading system has been formulated by its executive leaders. The company plans to gather feedback from employees, investors, and board members, with potential adjustments over time.