Springboards Tests Qwen 3 Model Against Repetitive LLM Answers
Australian startup Springboards has built Flint on Alibaba’s Qwen 3 to produce more varied answers to open-ended prompts. MIT Technology Review’s article pairs the company’s claim with a NeurIPS-winning homogeneity paper and user cautions that the prototype still fails under pressure.
Springboards Builds Flint On Qwen 3
Australian startup Springboards has built an LLM called Flint to make open-ended chatbot answers less repetitive.
The company is pitching the model to advertising and marketing users who want more varied brainstorming output than mainstream systems often produce.
Springboards cofounder and CEO Pip Bingemann said most language models are designed to fight hallucinations, while Flint is built to invite more unusual suggestions.
In one demonstration described by MIT Technology Review, ChatGPT and Claude gave the same simple campaign tagline, while Flint returned a different line.
The company built Flint on Qwen 3, the open-source model from Alibaba.
Springboards cofounder and CTO Kieran Browne said training a foundation model was too expensive for the small team, so the company focused on changing where a model introduces variety in its output.
Research Paper Shows Repeated Answers
The startup is working on a problem that AI researchers have also measured.
A November paper titled "Artificial Hivemind" found that different LLMs often converge on similar answers to open-ended prompts.
The researchers asked 25 LLMs to write a metaphor about time 50 times each.
MIT Technology Review said most of the 1,250 responses were versions of "Time is a river" or "Time is a weaver." The paper won a best paper award at NeurIPS.
OpenAI told MIT Technology Review that training models to give reliable and coherent answers can make them converge on familiar, high-probability responses.
OpenAI also said pushing harder for novelty can make responses less reliable.
Prototype Users Still Need Human Judgement
Springboards is offering Flint as an optional model within its brainstorming tool, which lets creative teams combine text from multiple LLMs.
Zoe Scaman, founder of Bodacious and chief strategy officer at 77X, said Flint pushed her in different directions during tests.
Scaman also said the premise was powerful while noting that Flint remains a prototype and can fail when users push it too far.
MIT Technology Review presented the evidence as creative-team testing of variety, not as proof of enterprise deployment.
Maximilian Weigl, cofounder and chief strategy officer at Uncommon, said his team uses Flint with ChatGPT, Claude and Gemini.
He also said average answers are often good enough and warned against teams copying AI output without human thinking.
The public record still lacks Flint pricing, a general launch date, customer numbers, enterprise deployment commitments or independent benchmark results for the prototype.
