LLMs are stuck in a groupthink rut. This startup is trying to get them out.

2026-07-01 · Source: MIT Technology Review · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Australian startup Springboards developed Flint, an LLM designed to overcome the "groupthink" tendency observed in mainstream models like Claude, ChatGPT, and Gemini. These models often produce highly predictable and similar responses to open-ended prompts, such as consistently suggesting "7" for a random number or "Toyota" for a car type. A NeurIPS 2025 best paper, "Artificial Hivemind," confirmed this homogeneity across 25 LLMs, finding common metaphors like "Time is a river." Flint, built on Alibaba's Qwen 3, aims to generate a wider variety of creative outputs by selectively boosting randomness at specific points in its responses, rather than globally adjusting temperature settings which can lead to incoherence. Springboards offers Flint as an alternative within its creative brainstorming tool, targeting advertising and marketing professionals.

Key takeaway

For AI Product Managers or Marketing Professionals seeking genuinely novel ideas, recognize that standard LLMs like ChatGPT or Claude often produce convergent, predictable outputs, limiting true creative exploration. Instead of relying solely on these models for ideation, consider integrating specialized tools like Springboards' Flint. This approach can "catapult" your team into diverse conceptual directions, offering unique starting points for campaigns or product reinvention, especially when the goal is to break from conventional thinking rather than just generate "good enough" responses.

Key insights

Mainstream LLMs exhibit significant response homogeneity, which Springboards' Flint model addresses by selectively increasing output variety.

Principles

• LLMs trained similarly on similar data converge on predictable answers.
• Global "temperature" adjustments can degrade LLM coherence.
• Targeted randomness can enhance creative output without sacrificing reliability.

Method

Springboards trained Flint, based on Qwen 3, to identify specific points in its output where more variety is possible and then inject more random words or phrases at those locations.

In practice

• Use Flint for brainstorming creative campaign taglines.
• Employ Flint to explore diverse solutions for complex business problems.
• Integrate Flint into creative tools for varied text generation.

Topics

• LLM Homogeneity
• Creative AI
• Springboards Flint
• Qwen 3
• Generative AI
• Marketing Technology
• AI Brainstorming

Best for: AI Engineer, Machine Learning Engineer, NLP Engineer, Marketing Professional, AI Product Manager, Director of AI/ML

Related on AIssential

• Why Temperature 0 Doesn’t Make Language Models Think — LLMs generate text via sequential token prediction based on conditional probability, not internal planning or reasoning.
• I created an LLM trained solely on Jeffrey Epsteins emails to see how messed up it becomes :) — Fine-tuning an LLM on a controversial, unannotated dataset can reveal unique behavioral shifts and spark technical debate.
• How do ML researchers actually use AI tools to improve their writing? [D] — LLMs offer efficiency in technical writing and coding but risk diminishing cognitive skills and originality if over-relied upon.
• Nous: An Attempt to Extract and Inject the Cognition Behind Prediction-Market Behavior — LLM agent ensembles face cognitive monoculture; prompt-level injection of human behavioral profiles fails to induce diversity.
• Week Ending 3.22.2026 — Advancements in LLM reasoning, multi-agent coordination, creativity, and social choice highlight diverse progress in AI capabilities and understanding.
• When Do LLMs Generate Realistic Social Networks? A Multi-Dimensional Study of Culture, Language, Scale, and Method — LLM-generated social networks are significantly influenced by prompt design, culture, language, and model scale.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT Technology Review.