Former OpenAI Chief Scientist Ilya's Secret Research Exposed: Aiming to Make AI Learn on the Job Like Humans

According to Perceptual Beating monitoring, SSI has not yet released a model, but an overseas blogger has pieced together its potential research direction from public clues: allowing AI to continue learning after deployment, rather than halting progress at the end of training.

The main clues come from Ilya Sutskever himself. He has repeatedly stated that the biggest flaw in existing models is their generalization ability, which is far inferior to that of humans. His ideal AI is like a "super-intelligent 15-year-old" who can enter any industry and quickly learn to work with minimal experience.

The Wall Street Journal also reported that SSI is studying "parts of brain operation that have been overlooked." After reviewing its confidential research, NVIDIA decided to invest and help SSI increase its computing power tenfold within a year.

Based on this, the developer speculates that SSI may be training an AI that can recognize errors, absorb experiences, and learn new skills without forgetting old ones.

However, this is still just a puzzle. SSI has not disclosed its architecture, model, performance, or demonstrations.