The public-facing arm of CaLiBRATE
THIRD EYE ON AI
Watching the watchers.
AI should be accountable to the people it affects. We examine its behaviour, ask what the evidence supports, and turn those questions into software development.
Public scrutiny. Practical research. Human oversight.
Oversight that asks for evidence.
Third Eye on AI is the public-facing arm of the wider CaLiBRATE portfolio: a place for films, questions and plain-language discussion about AI behaviour, accountability and human impact.
“Watching the watchers” means examining the systems that increasingly shape what we see, the decisions we make and the help we receive. It means asking how claims were reached—and making room for correction.
- Observe carefully.Separate a recorded event from an interpretation of why it happened.
- Make uncertainty visible.Acknowledge what is known, what is not, and what further testing could resolve.
- Build, then test.Turn recurring problems into development questions, without presenting a prototype as a proven solution.
- Keep people in control.Protect private information and make space for informed human review.
Building toward more stable, accountable AI.
CaLiBRATE is developing software intended to support AI stability and safety through better evidence handling, continuity and evaluation. This is work in progress—not a claim that one tool can stabilize every AI system.
CDI / Aletheia
What does the evidence actually support?
Evidence-review software being developed to keep source material, interpretation, uncertainty and unsupported claims distinct. The aim is clearer reasoning and decisions that people can inspect and challenge.
A private prototype exists. Product acceptance, real-world reliability and sensitive-data readiness are not yet established.
Anchor
How do we keep the context without losing the source?
Continuity and memory tooling being developed to carry source-linked project context, decisions and checkpoints between stages of work. The aim is less lost context, fewer repeated mistakes and a more traceable history.
Local tests and recovery checks are development evidence—not proof of permanent hosting or unattended operation across every chat.
NeuroWeave & related research
How can we test better ways of reasoning?
Experimental simulation and reasoning work, alongside Deep Thought / Atlas and APE-X integration research. These projects explore ideas through prototypes, controlled comparisons and explicit evaluation limits.
Synthetic and local results do not establish improved model quality, scientific fidelity or production readiness. Independent evaluation remains part of the path forward.
AI oversight is the front door—not the whole house.
CaLiBRATE also develops family-support tools, housing and public-policy proposals, accessibility concepts, responsible physical-product research, books and films. Third Eye brings the AI work into public view while the wider mission continues.
Follow. Question. Share.
Watch the public work, bring thoughtful feedback, or support the wider CaLiBRATE mission.
Voluntary support. Not a registered charity; no charitable tax receipt or investment return.