Best Practices
Verify and Validate:
AI tools are helpful assistants, not authoritative sources. Always review AI-generated content with your professional judgment before using it in teaching materials, just as you would teach students to evaluate online sources.
Recognise AI Limitations:
Current AI systems may present incorrect information confidently. These AI tools excel at patterns and language, but lack true understanding of facts or context that you as an educator possess.
Professional Discernment:
This service supports your professional practice. As you explore these tools, consider how they might complement your expertise rather than replace it. This platform is not intended for classroom use with students or as a student-facing resource.
Protect Privacy:
While this platform has zero data retention, always avoid entering personal information (yours or students') when using AI tools. This is a good practice for all digital tools, regardless of their privacy policies.
Responsible AI — how the underlying models are tested
Good Future AI is built on top of frontier AI models from Anthropic and Google. Both vendors publish detailed safety, bias and capability evaluations for the models we use. We rely on their evaluations rather than running parallel safety testing of our own, and we layer our own teacher-facing guidance (above) on top.
Which models we use
- Anthropic (Claude): Text generation, document analysis, and structured outputs
- Google (Gemini): Conversation simulation, image generation and editing and text-to-speech
- Exa: Web search and page content retrieval
Where to read the safety evaluations
Each model below has a publicly available system card or model card covering bias evaluations, safety testing, and known limitations. The UK Department for Education's standards explain what schools can expect of AI tools used in education.
- Anthropic - Model System Cards (safety, bias, red-team evaluations)
- Anthropic - Transparency Hub
- Google DeepMind - Gemini Model Cards
- UK DfE - Generative AI Product Safety Standards (Standard 8: Design and Testing)
- UK DfE - Cyber security standards for schools and colleges
A note on Character Chat
Specific note on Character Chat: this tool simulates fictional conversations with historical figures using the underlying AI model. Outputs are imaginative and may include factual errors or anachronisms, even when phrased confidently. Always fact-check against curriculum-approved sources before classroom use.