From ChatGPT to Superintelligence: OpenAI’s 5 Levels of Progress – OpenAI

The scale consists of five levels, with today’s chatbots reaching level one. OpenAI claims to be approaching level two, which is defined as a system capable of solving basic problems at the level of a person with a PhD.

Subsequent levels include more advanced capabilities: Level 3 refers to AI that can perform actions on behalf of the user, Level 4 to AI that can create new innovations, and Level 5, the final stage of AGI, to AI that can perform work for entire organizations.

OpenAI previously defined AGI as “a highly autonomous system that outperforms humans at most economically viable tasks.” The company’s structure is focused on achieving AGI, and how OpenAI defines AGI is key.

The company stated that if another company’s effort, with OpenAI’s values ​​and focus on security, comes close to building AGI before OpenAI, the company is committed to not competing with it and to dedicating all of its resources to supporting it.

However, AGI remains a distant goal, requiring massive investments in computing power. Estimates of the timeline for achieving AGI vary widely, with OpenAI CEO Sam Altman predicting in October 2023 that it would take about five years.

The announcement of the measure coincides with OpenAI’s collaboration with Los Alamos National Laboratory to explore the safe use of advanced AI models in life science research. The goal is to create evaluation criteria for AI models.

The dismantling of OpenAI’s security team in May and the departure of key executives raised concerns about the company’s prioritization of security.

OpenAI has not disclosed details about how it evaluates its models internally. However, the company recently presented a research project using the GPT-4 model, which is believed to demonstrate new thinking skills that are close to those of humans.

The metric could help provide a strict definition of progress, rather than leaving it open to interpretation. This is especially important given the conflicting statements from OpenAI executives about the capabilities of current models.