Home / Technology / AI Can't Imagine: New Models Fail Novel Tasks
AI Can't Imagine: New Models Fail Novel Tasks
23 Jun
Summary
- Current AI models lack the ability to gain new skills through imaginative prompting.
- AI role-playing as a more advanced model does not unlock latent capabilities.
- The Riemann Hypothesis remains an unsolved mathematical problem for both AI and humans.
Current artificial intelligence models demonstrate a significant limitation: they cannot develop new abilities through imaginative prompting. An experiment involving Anthropic's Claude and OpenAI's GPT-4.5 revealed that asking AI to "imagine" it is a more advanced model, such as the unavailable Fable 5, does not unlock latent capabilities.
When prompted to solve complex problems like the 167-year-old Riemann Hypothesis, Claude Opus 4.8 stated directly that role-playing would not change its mathematical reach. Its sibling models, Sonnet 4.6 and Haiku 4.5, also acknowledged their inability to solve the problem, with Haiku detailing hypothetical steps before admitting a conceptual ceiling.
OpenAI's GPT-4.5 and Google's Gemini 3.1 Pro also failed to solve the Riemann Hypothesis when prompted to role-play as Fable 5. These results underscore that AI's current challenges are conceptual rather than computational, proving that merely rebranding or prompting an AI does not grant it new, unforeseen skills.
