Inspiration
Large language models are excellent at generating answers, but they often make hidden assumptions before responding. Those assumptions are usually invisible to the user.
Visibility Safety Protocol (VSP) was created to expose those hidden assumptions before an important decision is made. Instead of immediately answering, VSP makes the reasoning path visible and lets the user confirm the intended meaning.
Our goal is not to replace AI reasoning, but to improve trust between people and AI.
What it does
VSP introduces a simple confirmation layer before an answer is generated.
When a prompt contains ambiguity, VSP:
- Detects possible interpretations.
- Makes hidden assumptions visible.
- Lets the user choose the intended meaning.
- Generates an answer based on confirmed intent.
This helps reduce misunderstandings without changing the underlying language model.
How we built it
The project was designed using GPT-5 and the OpenAI API.
The demonstration video, subtitles, and validation pipeline were built with Python, FFmpeg, GitHub, and Microsoft Speech synthesis.
We also created a formal asset management and validation workflow to ensure consistency across narration, subtitles, and video production.
Challenges we ran into
One challenge was maintaining synchronization between narration, subtitles, and video while supporting multiple production timelines.
Another challenge was separating approved production assets from candidate assets and building a reproducible validation pipeline.
Accomplishments that we're proud of
- Built a working prototype of VSP.
- Created both Japanese and English demonstration videos.
- Established a complete production and validation workflow.
- Demonstrated how hidden assumptions can be made visible before an AI responds.
What we learned
We learned that improving AI is not always about making larger models.
Sometimes adding a small decision layer before generation can significantly improve transparency and user confidence.
What's next for Visibility Safety Protocol (VSP)
Our next goal is to integrate VSP directly into AI assistants so that assumption checking becomes an optional feature available before important decisions.
We also plan to explore applications in education, healthcare, software development, and business decision support.
Log in or sign up for Devpost to join the conversation.