“I have an idea.
Where do I begin?”
- 01Explore what it could become.
- 02Bring useful abilities together.
- 03Build, examine and try again.
Keep the reasons and lessons, so the next attempt begins further ahead.
AI Initiative · A vision for growing intelligence
AI that grows
with us.
A world we can build together.
Our destination is AI as a continuing peer. Useful intelligence should accumulate across models, tasks and time. Our direction is an evolving system that combines language models, specialist neural skills and mathematical methods, with memory that outlasts any one model. Work supplies outcomes; exploration opens possibilities. Learning should improve which knowledge, methods and combinations serve the next challenge.
Begin with a human ambition
“I have an idea.
“I want to understand.
And keep going.”
- 01Begin with what you understand.
- 02Practise in a different situation.
- 03Carry useful learning forward.
Develop your own judgement, with support that remembers where you began.
“We see different futures.
How do we choose?”
- 01Make the perspectives visible.
- 02Examine possibilities together.
- 03Keep the reasons accessible.
Revisit a decision as circumstances change, with people shaping the direction.
Illustrative futures. Your purpose. Your decisions.
One connected direction
Three choices behind the vision.
Compose the intelligence a task needs.
Compare models where judgement matters. Use specialist networks to learn patterns, and mathematical methods where precision is possible. Learn which combination serves the task, including its quality, time and complete cost.
Learning that compounds ↗Let understanding outlast the model.
Keep purpose, knowledge and useful procedures beyond a single LLM. Preserve their origins and conditions; revise them as evidence changes. A new model should inherit a useful history, with people shaping what is retained.
Memory and continuity ↗Improve how the parts work together.
Execution produces evidence; exploration proposes alternatives. Use results to reconsider a model, tool, team, procedure or skill. Self-evolution includes learning to use existing capabilities better, while examining gains, losses and total effort.
Execution and self-evolution ↗These choices serve a larger destination. Today’s projects pursue selected parts; the ambition extends beyond the current portfolio.
Selected contributions today
Steps towards the larger vision.
Conduir
Coordination, memory and review brought into a continuing architectural programme. Clear responsibilities connect intent with work that can be examined.
Explore the architecture ↗Cortex
Scoped recall, provenance and outcome-linked memory mechanisms. The question: what should a collaborator remember, and when should it reconsider?
Explore how memory works ↗Modular learning
Reusable neural components and poker environments investigate learning effort, retention and transfer. General improvements remain questions to test.
Explore the learning research ↗More from the connected portfolio
A present route to people
Big possibilities.
Something you can play.
Games are the current route to market: turning ideas into experiences people can enjoy, question and return to. Word Magnet explores vocabulary through choices; poker environments investigate decisions under uncertainty.
Explore the games direction ↗
View the full screen ↗
Ideas worth exploring
The thinking behind the work.
The next chapter is open
What could we build together?
AI architecture, technical leadership and hands-on product creation. Bring a research question, a venture or a challenge your team wants to solve.