Skip to main content
Responsible AI Use

AI for Education,
Built Responsibly

Tandem uses AI to reduce teacher workload and improve student outcomes — while being transparent about environmental impact and committed to efficiency.

Our Commitment

Tandem does not train foundation AI models. We use efficient inference, structured prompts, and aggressive caching to minimise our environmental footprint while maximising educational value.

Live Impact Counter

Our Carbon Savings

Real-time COâ‚‚ savings vs traditional paper worksheets

0 kg
COâ‚‚ saved so far
From 0 digital quizzes replacing paper worksheets
0
km of driving avoided
0.0
trees' yearly absorption
0
sheets of paper saved
0g
saved in last 24 hours

The Global AI Picture

Understanding where AI's environmental impact really comes from

460 TWh
Global data centre electricity use (2022)
~2% of global electricity
~7,000 tCOâ‚‚e
Training GPT-4 (estimated)
Equivalent to ~1,500 cars/year
0.03–0.3g
COâ‚‚ per typical AI query
Varies by model complexity

Why Tandem's Footprint is Smaller

Our design choices actively reduce environmental impact

Inference Only

We query existing models — we don't train new ones. Training is where the vast majority of AI carbon cost lives.

Biggest impact

Task-Bounded Prompts

Structured, curriculum-aligned outputs use fewer tokens than open-ended chat. Shorter tasks = lower energy per query.

Caching & Reuse

Repeated outputs are cached. One LLM call can serve hundreds of students, dramatically reducing real-world carbon impact.

Feature Gating

Our unlock system prevents wasteful inference. Students can't spam AI calls — intentional use only.

Per-Query Emissions Comparison

How different AI activities compare

ActivityEstimated COâ‚‚ Emissions
Standard Google search~0.2g COâ‚‚e
AI search (Gemini)~0.03g COâ‚‚e
Typical GPT-4 query~0.1–0.3g CO₂e
Tandem structured query~0.05–0.15g CO₂e
Older LLM models~2.5–5g CO₂e

Figures are estimates based on published research. Actual emissions vary by infrastructure, model version, and query complexity. Tandem's structured prompts typically use fewer tokens than open-ended chat.

Why Education AI is Worth It

AI is going to exist — the question is how it's used

Less Valuable AI Uses

  • ✗Ad-driven infinite scroll engagement
  • ✗Crypto speculation and hype
  • ✗Generative spam and junk content
  • ✗Unbounded entertainment chat

Socially Valuable AI Uses

  • ✓Education — reducing workload, improving outcomes
  • ✓Accessibility — supporting diverse learners
  • ✓Teacher support — less admin, more teaching
  • ✓Equity — quality support for every student

Tandem sits firmly in the "socially useful AI" category. We replace repetition with reuse, reduce marking workload, and give every student access to structured support — without creating infinite demand.

In Summary

Our approach to responsible AI

Inference only — no model training
Structured, bounded prompts
Aggressive caching and reuse
Feature gating prevents waste
Education-focused, not entertainment
Transparent about our approach

"AI-assisted workflows like Tandem's use efficient inference, not training — meaning each query emits only a fraction of a gram of CO₂ equivalent. Outputs are maximised through caching and reuse, making our per-student footprint closer to a few Google searches per lesson than streaming video."

Questions About Our AI Approach?

We're happy to discuss our environmental practices with schools and decision-makers