OpenAI, according to its head of applied research, is concentrating 80 to 90 percent of its work on GPT-7 and further generations
Boris Power, head of applied research at OpenAI, said that 80 to 90 percent of the company's research targets GPT-7, GPT-8 and further generations, because real value is created in the jump between generations, not in incremental versions such as GPT-5.1 to 5.2.
Boris Power, head of applied research at OpenAI, said in his own words that 80 to 90 percent of the company's research focuses on future generations of models - GPT-7, GPT-8 and beyond - because that is where the greatest value is created. Incremental updates within a single generation, for example the move from GPT-5.1 to GPT-5.2, are based on specialized training data, but according to him these are deliberately short-term bets that are viewed inside the company as short-sighted.
According to Power, a real step change in performance only comes with the move to a new model generation, after which OpenAI has to figure out again where to invest for fast results. Short-term updates within a generation, meanwhile, help the company iterate and learn faster, even though, according to Power, this is not the correct long-term strategy.
Power does not see the main problem of today's AI assistants in model quality, but in onboarding - according to him, most ChatGPT users do not know everything they can do with AI. Future models should therefore better showcase their capabilities and anticipate users' needs. Power described the development as follows: GPT-4 required precisely formulated prompts, GPT-5 is easier to use but still requires a lot of feedback, while GPT-6 is supposed, according to his description, to function more like a capable colleague to whom it is enough to just state a goal.
Why it matters
The statement shows how OpenAI, according to its own representative, allocates research capacity, and suggests that major improvements can be expected rather between model generations than in incremental versions. This is relevant for developers and companies planning products on top of these models, because they can expect step changes in behavior and prompting requirements rather than smooth, gradual development.
Two audiences, two different impacts
What this means
For individuals
Until further generations of models arrive, ChatGPT users must continue to expect that it is necessary to give the model detailed feedback and figure out for themselves everything that can be done with AI, because according to Power the main weakness of current versions is onboarding, not model quality.
For a business
Companies building products on OpenAI models should, according to Power's statements, expect that major improvements arrive in steps with a new model generation, while incremental versions within a single generation are only short-term solutions - this affects planning of product development and investment into the prompting and onboarding layer.
StrategyCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.