Artificial Intelligence
OpenAI targets most research at GPT-7 and beyond
An OpenAI executive said most of the company’s research was aimed at future model generations, while he argued that the larger problem for current assistants was user onboarding.
- Boris Power, OpenAI’s Head of Applied Research, said 80 to 90 percent of the company’s research went towards GPT 7, GPT 8 and later models.
- He said smaller upgrades such as GPT-5.1 to GPT-5.2 came from specialised training data and were short-term bets.
- Power said OpenAI saw those incremental updates as shortsighted, even though they helped the company iterate and learn faster.
- He said the main problem with current AI assistants was onboarding, because many ChatGPT users did not know what they could do with AI.
- Power said GPT-6 would work more like a capable colleague that could be given a goal.
Power said OpenAI directed 80 to 90 percent of its research towards GPT 7, GPT 8 and later generations. He said that was where most of the value would come from.
He said smaller changes, such as moving from GPT-5.1 to GPT-5.2, depended on specialised training data and were intended as short-term bets. He also said the company viewed those updates as short-sighted, although they helped it learn and iterate more quickly.
Power said the larger jump came with a new model generation, when many other pieces worked better together. After each jump, he said, OpenAI had to decide again where to place its efforts for quicker gains.
On current assistants, Power said the main issue was not model quality but onboarding. He said many ChatGPT users did not know what they could do with AI, and he described GPT-4 as needing careful prompting, GPT-5 as easier to use, and GPT-6 as closer to a capable colleague.
Named in this story
People
- Boris Power
- OpenAI’s Head of Applied Research who described the company’s research focus
Companies
- OpenAI
- the company whose research priorities Power discussed
Products and systems
- GPT 7
- a future model generation Power said much of the research aimed towards
- GPT 8
- another future model generation Power said much of the research aimed towards
- GPT-5.1
- an incremental model version Power used as an example of a short-term update
- GPT-5.2
- an incremental model version Power used as an example of a short-term update
- GPT-4
- an earlier model Power said needed careful prompting
- GPT-5
- a model Power said was easier to use than GPT-4
- GPT-6
- a future model Power said would work more like a capable colleague
- ChatGPT
- the assistant Power said many users did not know how to use
How the source tells it
The register was promotional and forward-looking, driven by an OpenAI executive's internal roadmap claims rather than independent reporting.
- hype future generations were framed as the main source of value, while smaller upgrades were treated as temporary
- vendor boosterism the article repeated an executive’s account of the company’s own priorities without outside challenge
- speculation presented as fact future model behaviour was described in advance as if it were already established