![[2602.15829] Operationalising the Superficial Alignment Hypothesis via Task Complexity](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15829] Operationalising the Superficial Alignment Hypothesis via Task Complexity
Summary
This article presents a new metric called task complexity to operationalize the Superficial Alignment Hypothesis, demonstrating how pre-trained models reduce the complexity of achieving high performance on various tasks.
Why It Matters
Understanding the Superficial Alignment Hypothesis is crucial for advancing machine learning, particularly in optimizing large language models. This research provides a clearer framework for evaluating model performance and task adaptation, which can influence future AI development and applications.
Key Takeaways
- Introduces task complexity as a metric for evaluating model performance.
- Demonstrates that pre-trained models significantly lower the complexity of achieving task performance.
- Post-training can drastically reduce the complexity of reaching high performance.
- Highlights that minimal information is often needed for task adaptation.
- Unifies various arguments supporting the Superficial Alignment Hypothesis.
Computer Science > Machine Learning arXiv:2602.15829 (cs) [Submitted on 17 Feb 2026] Title:Operationalising the Superficial Alignment Hypothesis via Task Complexity Authors:Tomás Vergara-Browne, Darshan Patil, Ivan Titov, Siva Reddy, Tiago Pimentel, Marius Mosbach View a PDF of the paper titled Operationalising the Superficial Alignment Hypothesis via Task Complexity, by Tom\'as Vergara-Browne and 5 other authors View PDF HTML (experimental) Abstract:The superficial alignment hypothesis (SAH) posits that large language models learn most of their knowledge during pre-training, and that post-training merely surfaces this knowledge. The SAH, however, lacks a precise definition, which has led to (i) different and seemingly orthogonal arguments supporting it, and (ii) important critiques to it. We propose a new metric called task complexity: the length of the shortest program that achieves a target performance on a task. In this framework, the SAH simply claims that pre-trained models drastically reduce the complexity of achieving high performance on many tasks. Our definition unifies prior arguments supporting the SAH, interpreting them as different strategies to find such short programs. Experimentally, we estimate the task complexity of mathematical reasoning, machine translation, and instruction following; we then show that these complexities can be remarkably low when conditioned on a pre-trained model. Further, we find that pre-training enables access to strong performanc...