QuantumColleagues. ← Thoughts
Do some good with AI.

Thoughts / Isambard-AI

Which AI can care providers trust? A summary of our AIRR study on Isambard-AI: 10,000 GPU hours, six care admin tasks, 120 scenarios in a simulated care office and no real people in the data. It asks which model size handles each care task well enough, how far a model can shrink before quality drops, when the AI should stop and hand the work to a person, and whether quality is the same for every resident. Every person in the data is invented, no AI holds the send button and anything consequential waits for a person.

Which AI can care providers trust?

Social care · 17 November 2026

Our care AI study on Isambard-AI is up and running.

The first model ran on it this week. Between now and February I’m testing which open models handle care admin well and safely on a provider’s own hardware, and every person in the data is invented.

The summary above sets out what we’ll find out and how we keep it safe. We announced the award on 6 November, and I’ll publish the findings here for the whole care sector.

This study is supported by the AI Research Resource (AIRR) programme, funded by UK Research and Innovation and the Department for Business, Innovation, Science and Trade. Isambard-AI is operated by the University of Bristol.

If you’d like to talk to us about this, then please contact us here or email us directly at information@quantumcolleagues.org.

Businesses buy our AI, and that funds free post-AI training for every generation.
Do some good with AI.

© QuantumColleagues Ltd 2026 · Company No. 17219359 · Tel 0333 880 0712 · 50% owned by Quantum Colleagues Training CIC (No. 17244034) · Photos: Pexels · Privacy · Follow on Google