Boris Power, OpenAI’s Head of Applied Research, states that 80 to 90 percent of the company’s current research efforts are directed toward GPT 7 and subsequent generations. He explains that while incremental updates like moving from GPT 5.1 to 5.2 rely on specialised training data, the substantial value lies in new model generations where overall performance improves significantly. Power describes these smaller updates as extremely shortsighted within the organisation, noting they are only useful for rapid iteration today but do not represent a sound long-term strategy. The real leap occurs when a new generation arrives, allowing everything else to function much better, after which the company must relearn where to invest for quick wins.
The primary challenge for current AI assistants is not model quality but user onboarding. Most ChatGPT users struggle to understand what the technology can do, requiring future models to better surface possibilities and anticipate needs. GPT-4 demanded careful prompting, GPT-5 is easier to use but still requires significant feedback, and GPT-6 is designed to work like a capable colleague where users simply hand over a goal. This shift aims to reduce the friction of interaction and make advanced capabilities accessible without explicit instruction.
* Incremental model updates are viewed as short-term bets rather than long-term investments.
* Future models will focus on anticipating user needs instead of requiring complex prompting.
* GPT-6 is positioned to function as an autonomous colleague requiring minimal direction.
