R&D in Reliability
causaLens is investing into furthering the reliability of our Digital Workers. Learn more about what we're working on below.
Reliability in Digital Workers
Dr. Steven Kleinegesse, our Director of Research, Machine Learning, discusses our work on the causally trained LLM CLAIRE (Causal Logic in AI Reasoning Models), world models, and determinism, and how these advances led to 10–100% gains on core reasoning benchmarks.
CLAIRE (Causal Logic in AI Reasoning Models)
The Claire project explores whether small language models (LLMs) can be fine-tuned to reason causally using supervised fine-tuning (SFT) with synthetic data.
- Performance: Claire-1.5B and Claire-7B models consistently outperformed their base counterparts in causal reasoning tasks, especially on GSM8K and GPQA benchmarks.
- Reasoning Quality: Outputs from the fine-tuned models followed structured causal reasoning formats and demonstrated meaningful world modeling.
- Efficiency: Training time was modest (10–30 minutes per model) due to PEFT techniques, although model performance saturated after one epoch due to batch size limitations.
Interview with Guido Imbens: Nobel Prize Laurette
In this wide-ranging conversation, Nobel Prize-winning economist Guido Imbens reflects on the future of data science, AI agents, and causal reasoning. From the challenges of applying causality in real-world workflows to the role of human judgment in increasingly automated systems, Guido shares timely perspectives on how to make data science more rigorous, trustworthy, and impactful.
Professor Imbens is a Nobel Laurette Winner for Economic Sciences
Research topics in which he is interested include: causality, program evaluation, identification, Bayesian methods, semi-parametric methods, instrumental variables.
Co-Authored "Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction"
Causal-Enabled Digital Workers
Ready to generate insights 10x faster, ground them in causal reasoning, and automate your analytics workflows?