Operationalising the Superficial Alignment Hypothesis via Task Complexity

• URL: http://arxiv.org/abs/2602.15829v1
• Date: Tue, 17 Feb 2026 18:59:39 GMT
• Title: Operationalising the Superficial Alignment Hypothesis via Task Complexity
• Authors: Tomás Vergara-Browne, Darshan Patil, Ivan Titov, Siva Reddy, Tiago Pimentel, Marius Mosbach,
• Abstract summary: We propose a new metric called task complexity: the length of the shortest program that achieves a target performance on a task.<n>Our results highlight that task adaptation often requires surprisingly little information -- often just a few kilobytes.
• Score: 49.93635747700126
• License: http://creativecommons.org/licenses/by/4.0/
• 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 performances on our tasks, but it can require programs of gigabytes of length to access them. Post-training, on the other hand, collapses the complexity of reaching this same performance by several orders of magnitude. Overall, our results highlight that task adaptation often requires surprisingly little information -- often just a few kilobytes.

Related papers

• Provable Benefit of Curriculum in Transformer Tree-Reasoning Post-Training [76.12556589212666]
  We show that curriculum post-training avoids the exponential complexity bottleneck.<n>Under outcome-only reward signals, reinforcement learning finetuning achieves high accuracy with sample complexity.<n>We establish guarantees for test-time scaling, where curriculum-aware querying reduces both reward oracle calls and sampling cost from exponential to order.
  arXiv  Detail & Related papers  (2025-11-10T18:29:54Z)
• Training Task Reasoning LLM Agents for Multi-turn Task Planning via Single-turn Reinforcement Learning [15.393743659727926]
  Large Language Models (LLMs) have demonstrated remarkable capabilities in knowledge acquisition, reasoning, and tool use.<n>This paper introduces a novel approach that transforms multi-turn task planning into single-turn task reasoning problems.
  arXiv  Detail & Related papers  (2025-09-24T23:47:36Z)
• Supervised Chain of Thought [5.389461633686935]
  Chain of Thought (CoT) prompting offers a promising approach to solving complex reasoning tasks. One-prompt-for-all approach poses significant challenges for models to generate the correct reasoning steps. We show how task-specific supervision is essential for navigating the prompt space accurately and achieving optimal performance.
  arXiv  Detail & Related papers  (2024-10-18T06:25:27Z)
• Limits of Deep Learning: Sequence Modeling through the Lens of Complexity Theory [15.24542569393982]
  Despite their successes, deep learning models struggle with tasks requiring complex reasoning and function composition.<n>We present a theoretical and empirical investigation into the limitations of Structured State Space Models (SSMs) and Transformers in such tasks.<n>We highlight the need for innovative solutions to achieve reliable multi-step reasoning and compositional task-solving.
  arXiv  Detail & Related papers  (2024-05-26T19:33:23Z)
• When does In-context Learning Fall Short and Why? A Study on Specification-Heavy Tasks [54.71034943526973]
  In-context learning (ICL) has become the default method for using large language models (LLMs) We find that ICL falls short of handling specification-heavy tasks, which are tasks with complicated and extensive task specifications. We identify three primary reasons: inability to specifically understand context, misalignment in task schema comprehension with humans, and inadequate long-text understanding ability.
  arXiv  Detail & Related papers  (2023-11-15T14:26:30Z)
• Improving Long-Horizon Imitation Through Instruction Prediction [93.47416552953075]
  In this work, we explore the use of an often unused source of auxiliary supervision: language. Inspired by recent advances in transformer-based models, we train agents with an instruction prediction loss that encourages learning temporally extended representations that operate at a high level of abstraction. In further analysis we find that instruction modeling is most important for tasks that require complex reasoning, while understandably offering smaller gains in environments that require simple plans.
  arXiv  Detail & Related papers  (2023-06-21T20:47:23Z)
• Collaborating with language models for embodied reasoning [30.82976922056617]
  Reasoning in a complex and ambiguous environment is a key goal for Reinforcement Learning (RL) agents. We present a set of tasks that require reasoning, test this system's ability to generalize zero-shot and investigate failure cases.
  arXiv  Detail & Related papers  (2023-02-01T21:26:32Z)
• Generalizing to New Tasks via One-Shot Compositional Subgoals [23.15624959305799]
  The ability to generalize to previously unseen tasks with little to no supervision is a key challenge in modern machine learning research. We introduce CASE which attempts to address these issues by training an Imitation Learning agent using adaptive "near future" subgoals. Our experiments show that the proposed approach consistently outperforms the previous state-of-the-art compositional Imitation Learning approach by 30%.
  arXiv  Detail & Related papers  (2022-05-16T14:30:11Z)
• In Defense of the Unitary Scalarization for Deep Multi-Task Learning [121.76421174107463]
  We present a theoretical analysis suggesting that many specialized multi-tasks can be interpreted as forms of regularization. We show that, when coupled with standard regularization and stabilization techniques, unitary scalarization matches or improves upon the performance of complex multitasks.
  arXiv  Detail & Related papers  (2022-01-11T18:44:17Z)
• CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learning [138.40338621974954]
  CausalWorld is a benchmark for causal structure and transfer learning in a robotic manipulation environment. Tasks consist of constructing 3D shapes from a given set of blocks - inspired by how children learn to build complex structures.
  arXiv  Detail & Related papers  (2020-10-08T23:01:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.