LLMs are stuck in a groupthink groove. This startup is trying to get them out.
LLM output homogeneity is a real limitation for government communications teams - this item surfaces the problem even if the solution is commercial.
Key points
- Startup Springboards trained a modified Qwen 3 model ('Flint') to inject targeted randomness at specific output points, not uniformly.
- The product targets LLM homogeneity in creative tasks - a known limitation when agencies use AI for communications or policy drafting.
- A niche commercial product aimed at marketers; limited direct applicability to APS governance or regulatory work.
View original source
Copied.
"LLMs are stuck in a groupthink groove. This startup is trying to get them out."
Source: MIT Technology Review – AI
Published: 1 July 2026
URL: https://www.technologyreview.com/2026/07/01/1140003/llms-are-stuck-in-a-groupthink-rut-this-startup-is-trying-to-get-them-out/
Springboards, a startup, has released a fine-tuned LLM called Flint, based on Qwen 3, designed to inject variability at specific points in model output rather than raising randomness globally. The goal is to counter the tendency of mainstream LLMs to converge on similar, average outputs - a problem particularly felt in creative and ideation tasks. The product is currently aimed at advertisers and marketers. A user quoted in the article cautions against over-reliance on any LLM output, including Flint, advocating instead for human judgment and voice. The underlying problem of AI-driven homogeneity in text generation has broader relevance beyond marketing.
Retrieved from SIMS, 18 July 2026.