Koray Kavukcuoglu, the new chief of Google Deepmind, has stated he intends to release Gemini 4 significantly earlier than the current year-end target. The model is already in post-training and operates within the internal coding tool Antigravity. Kavukcuoglu describes the artificial general intelligence question that drove his predecessor Demis Hassabis as not the right conversation. He argues that trustworthy agents matter more than the abstract pursuit of general intelligence. This shift follows the quiet disappearance of Gemini 3.5 Pro and the departure of many top researchers for OpenAI and Anthropic. The research lab with an AGI mission has effectively turned into a product shop. The focus has moved from theoretical limits to practical deployment and reliability in real-world applications. This strategic pivot signals a departure from the previous obsession with achieving a singular breakthrough in machine cognition.
- Gemini 4 is currently running internally in Antigravity
- Kavukcuoglu views AGI as a distraction from building trustworthy agents
- Many researchers left for OpenAI and Anthropic after Gemini 3.5 Pro vanished
