Back to all discussions

  •   6 months ago

Clarification on AI Model Requirements for Evaluation/Testing Use Case

Hi organisers,

I'm building an AI safety evaluation platform for the competition.
My question is, if my solution uses Google Gemini models as the primary AI tool (to score results, and power the platform, etc), can the models being evaluated/tested be non-Google models?

Specific use case:
- My platform runs safety evaluations on AI models
- The evaluation logic, scoring, and analysis use Gemini (via Vertex AI)
- The models being tested are essentially "test data"
- This would allow comparative safety analysis

If not allowed, it's ok. I can focus on comparing different Gemini model versions (2.5 Pro, 2.0 Flash, etc) in the evals, which I think could still have some useful research. Just wanted to make sure my submission would be compliant.

Thanks in advance

  • 5 comments

  •   •   6 months ago

    Which track does this fall into?

  •   •   6 months ago

    Hi organisesrs,

    Sorry me again. Just thought I'd better reference the specific area I'm referring to in the rules:

    Specific rule reference: Section 7.B states "All other artificial intelligence tools are not permitted."

    I'm interpreting "use" to mean "actively invoke/call as a tool in my solution" (which would only be Google AI), but want to confirm whether models that produce data my solution analyses also count as "using" them or not.

    Thanks again.

  •   •   5 months ago

    Hi organisers,

    Just wondering if there was any update on the above please? Sorry to keep asking. It’s just that the outcome to my question would help determine how I continue building out the requirements for my project.

    Thank you for your time.

  • Manager   •   5 months ago

    Yes this should be fine as long as your final project is based only on Google AI tools.

  •   •   5 months ago

    Amazing, thanks so much Janet. Really appreciate your response.

Comments are closed.