Prediction and adaptation
Teaching AI to Survive Chaotic Worlds: Adaptive Predictive Coding
Prediction, adaptation and performance measurement in environments whose rules change.
1 December 2025 · 1h 25m
Dilate Technologies / Research programme
Toward intelligence that keeps learning
Loopseed explores how AI systems can learn from verified experience, adapt to changing environments and retain useful knowledge over time.
Research framework
The learning loop
Our mission
To research adaptive AI for secure and dependable critical infrastructure
We study how systems learn, keep useful knowledge and adapt to unfamiliar or chaotic environments, with human control guiding the work.
Dynamical Synthesis
Learning through interaction
Fish is our adaptive AI system that drives a language model. Our framework connects five steps, while controlled tests examine whether the system actually learns.
01
Predict
Anticipate the next observation.
02
Compare
Check what happened against the prediction.
03
Remember
Keep useful experience and its source.
04
Act
Respond or use a computational tool.
05
Update
Train on checked examples and test the change.
LIVE RESEARCH AND SHARED DATA
Research in public
We livestream research sessions so anyone can follow the questions, code and experiments as we work. Watch the recordings below and explore the published results and data.
Learning under change
Event-modulated Plasticity: Adaptive Predictive Coding
A research session on event-modulated plasticity: how changes in an environment can influence the way a system learns.
12 December 2025 · 1h 33m
Interaction between systems
Multi-agent Cultural Dynamics: Adaptive Predictive Coding
A research session on multi-agent cultural dynamics, examining interactions between systems that learn and adapt.
23 December 2025 · 1h 46m
Research updates
Latest research reports
11 September 2026
Research paper and supporting data
The paper draws on our existing experiments to examine where performance improved, where tests failed and which questions remain open. We provide supporting data and analysis files so readers can review the evidence behind these findings.
Read report
7 September 2026
Preparation for the learning comparison
The latest round collected 175 of 336 planned measurements, while two earlier rounds collected all their data but found too few suitable questions. The main comparison is unfinished, so we cannot tell whether the approach improves learning.
Read report
7 September 2026
Performance after further training
Prediction scores worsened on new and earlier test examples. Across 42 questions tested twice, 12 answers improved, 17 worsened and 55 stayed the same. Seven losses involved earlier skills, and neither update was adopted.
Read report
Proposed next study
What lasts after the next update?
We will compare memory, further training and reuse of earlier examples to test whether new learning can coexist with retention. A separate pilot will establish the tasks and training settings before the main study.
Loopseed
A research programme at Dilate Technologies.
© 2026 Dilate Technologies