Oparine. Pushing the limits of artificial creativity.

§ 01 — OPEN COLLIDER

Open Collider

A semantic collision engine for non-trivial idea generation

Grounded in Koestler's bisociation theory of human creativity (1964).

How a session works

FIG. 01 — Session protocol, four phases

Try it on GitHub →

§ 02 — LAUNCH STUDY

Escaping the Hivemind, measured.

Given the same brief, language models converge to a narrow region of idea space. Researchers call it the Artificial Hivemind (Jiang et al., 2025). We tested whether distant-domain collisions actually escape it.

Projects

12 real-world ideation problems

Baselines

3 controls (flat, "be original", deep brief)

Verdicts

4,320 blind LLM-judge pairs

Geometric distance

12/12

Every project escapes the default cloud. 4–13× further than asking the model to "be original".

n = 12 p = 0.0002 embed = BGE · e5

Blind LLM-judge

10/12

Three blind judges. A wins on originality, ties or beats on quality.

judges = 3 μ = 62% p = .019

Lion, C. (2026). Why Direct Prompting Pushes LLMs Toward Trivial Ideas. Oparine Working Paper 01.

Read the full study ↗ Dataset on GitHub

§ 03 — WORKING WITH US

For companies that refuse to converge.

We deploy our methods inside companies whose work depends on outputs the default LLM can't reach. Calibrated to your problem, your voice, your brand.

Refuse

the average voice.

For branding

When you want to escape AI slop and say what no one else would think of.

Refuse

the safe answer.

For strategy

When the upside is in being right alone, not right with everyone.

Refuse

the convergence.

For R&D

When you need to scale exploration into non-trivial, unexplored directions.