OpenAI says it could reach AGI before the end of 2026

OpenAI believes it will be able to achieve artificial general intelligence in 2026. The company’s CEO, Sam Altman, said in an interview with TIME Magazine that his organization hasn’t achieved AGI yet but will have an internal version soon enough. Sam said OpenAI is “not quite yet” there.

OpenAI’s Chief Research Officer Mark Chen gave a number to the gap and said OpenAI is “80% of the way” toward AGI. Co-founder Greg Brockman went further, saying people may later view this stretch as the point when AGI first showed up.

This TIME piece follows nearly three weeks of interviews with more than two dozen people from OpenAI’s executives to its employees, investors, and competitors. This company’s confidence is based on its new line of models, named Astra.

OpenAI introduced Astra last year as a system designed to mimic an automatic research assistant. The chief scientist of OpenAI, Jakub Pachocki, claims that this goal has already been achieved.

Jakub also said that one could give Astra an experiment to conduct; it will generate the required code within the software environment at OpenAI, do the experiment, and come back with the results. Jakub also added that Astra could take a research paper and complete the task within a week’s time.

Astra pushes OpenAI toward longer autonomous research and self-improving AI systems

Astra can also support what OpenAI calls persistent agents. These are digital workers that can stay on an assignment for longer periods without constant human input. Sam believes the bigger test will be whether Astra can create useful knowledge instead of only reorganizing existing material.

During a customer preview, he said, “I expect this will be the first model where the model actually invents new things in a way that matters.” He added, “That’s a very AGI-like thing.”

This notion leads to a broader gamble. If an AI system is capable of conducting research, then it could be used to create a superior AI system. This will result in a feedback loop wherein new systems will be used to create more advanced systems. Some say autonomous AI research is still much further away. Others argue that early parts of the process are already appearing.

There is also a debate concerning whether or not systems constructed predominantly on language models can generate novel discoveries and generalize their knowledge over a wide range of tasks. This skill is considered to be vital for AGI by many scientists.

At the same time, the term “AGI” is rather vague. As per OpenAI, AGI is “highly autonomous systems that surpass human performance on most economically relevant tasks.” However, other scientists have a more liberal criterion of AGI.

If you're reading this, you’re already ahead. Stay there with our newsletter.