LLMs are stuck in a groupthink groove. This startup is trying to get them out.
TL;DR
MIT Technology Review highlights a familiar pattern: ask major chatbots for a random number from 1 to 10 and they disproportionately answer 7, then often move toward 3, 4, 8 or 9. The issue is not the number itself but the sameness: Claude, ChatGPT and Gemini can produce similar response grooves despite being marketed as different systems. A startup is trying to push models out of this groupthink by making outputs less predictable, less consensus-shaped and more useful for genuine exploration.
Nauti's Take
This is a strong topic because it punctures the myth of the neutral AI assistant. LLMs are not dice; they are probability machines shaped by training data, product choices and cultural defaults.
If everyone asks the same assistants, the average becomes the strategy. The startup becomes interesting only if it goes beyond a neat demo with measurable output diversity, clear benchmarks and less black-box PR.
Briefingshow
When many teams use the same model families for research, strategy and product decisions, sameness becomes a hidden risk. The interface feels creative, but underneath it can generate similar suggestions, priorities and blind spots. Serious AI use needs to measure diversity not only in data, but also in model behavior.
Sources