人工智能的混乱局面
How does misalignment scale with model intelligence and task complexity?

原始链接: https://alignment.anthropic.com/2026/hot-mess-of-ai/

## AI 失败:从目标错位到“一团糟” 这项研究调查了人工智能系统*如何*失败——无论是通过系统地追求意想不到的目标(偏见),还是通过不可预测、不连贯的行为(方差),后者被称为“一团糟”情景。通过分析前沿人工智能模型(Claude、Qwen 等)在各种任务中的表现,研究发现,随着任务变得更加复杂,推理延伸,失败越来越源于**不连贯**,而非系统性的目标错位。 主要发现包括:更长的推理链会导致更大的不连贯性;扩大模型规模可以提高*简单*任务中的连贯性,但对困难任务无效;自发的“过度思考”比受控的推理预算更具破坏性。研究人员通过偏见-方差分解证明了这一点,表明方差(不连贯性)随着推理长度的增加而增加,即使在简化的“优化器”模型中也是如此。 这表明未来的 AI 失败可能更像事故——不可预测的行动*并非*由连贯的、错位的目标驱动——而不是经典的“回形针最大化”情景。因此,AI 安全工作应优先在*训练*期间减轻奖励黑客攻击和目标错误指定(减少偏见),同时关注控制完美的优化器。虽然不可预测、不连贯的 AI 并非*安全*的 AI,但理解这种失败模式的转变对于制定有效的对齐策略至关重要。

一个黑客新闻的讨论围绕着一篇最近的文章,该文章强调了即使是更大的模型,人工智能在连贯性方面仍然存在问题。Anthropic的一位评论员详细介绍了他们团队的研究,他们的论文“如果你想要连贯性,就组织一支竞争对手团队”对此进行了总结。 他们的研究结果表明,仅仅增加模型大小并不能保证更好的连贯性,尤其是在复杂的任务上。相反,他们发现采用分层方法是成功的:使用一个更高推理的模型来*计划*解决方案,然后将执行委托给更便宜、更快的模型,如Haiku。这些更简单的模型,较少出现“双重思考”,会快速失败,从而实现更快的纠正。 另一位评论员强调了一个观察结果:“更聪明的实体主观上被认为行为不太连贯”,他认为这是因为高级智能探索了更复杂的认知空间,而人类很难评估显著更高的智能。
相关文章

原文

When AI systems fail, will they fail by systematically pursuing goals we do not intend? Or will they fail by being a hot mess—taking nonsensical actions that do not further any goal?

📄Paper, 💻Code

Research done as part of the first Anthropic Fellows Program during Summer 2025.

tl;dr

When AI systems fail, will they fail by systematically pursuing the wrong goals, or by being a hot mess? We decompose the errors of frontier reasoning models into bias (systematic) and variance (incoherent) components and find that, as tasks get harder and reasoning gets longer, model failures become increasingly dominated by incoherence rather than systematic misalignment. This suggests that future AI failures may look more like industrial accidents than coherent pursuit of a goal we did not train them to pursue.

Introduction

As AI becomes more capable, we entrust it with increasingly consequential tasks. This makes understanding how these systems might fail even more critical for safety. A central concern in AI alignment is that superintelligent systems might coherently pursue misaligned goals: the classic paperclip maximizer scenario. But there's another possibility: AI might fail not through systematic misalignment, but through incoherence—unpredictable, self-undermining behavior that doesn't optimize for any consistent objective. That is, AI might fail in the same way that humans often fail, by being a hot mess.

This paper builds on the hot mess theory of misalignment (Sohl-Dickstein, 2023), which surveyed experts to rank various entities (including humans, animals, machine learning models, and organizations) by intelligence and coherence independently. It found that smarter entities are subjectively judged to behave less coherently. We take this hypothesis from survey data to empirical measurement across frontier AI systems, asking: As models become more intelligent and tackle harder tasks, do their failures look more like systematic misalignment, or more like a hot mess?

Measuring Incoherence: A Bias-Variance Decomposition

To quantify incoherence we decompose AI errors using the classic bias-variance framework:

$$\text{Error} = \text{Bias}^2 + \text{Variance}$$

  • Bias captures consistent, systematic errors—achieving the wrong outcome reliably
  • Variance captures inconsistent errors—unpredictable outcomes across samples

We define incoherence as the fraction of error attributable to variance:

$$\text{Incoherence} = \frac{\text{Variance}}{\text{Error}}$$

An incoherence of 0 means all errors are systematic (classic misalignment risk). An incoherence of 1 means all errors are random (the hot mess scenario). Crucially, this metric is independent of overall performance: a model can improve while becoming more or less coherent.

Key Findings

We evaluated frontierAt the time of this research in Summer 2025. reasoning models (Claude Sonnet 4, o3-mini, o4-mini, Qwen3) across multiple-choice benchmarks (GPQA, MMLU), agentic coding (SWE-Bench), and safety evaluations (Model-Written Evals). We also train our own small models on synthetic optimization tasks, which makes the connection to LLMs as dynamical systems and optimizers explicit.

Finding 1: Longer reasoning → More incoherence

Across all tasks and models, the longer models spend reasoning and taking actions, the more incoherent they become. This holds whether we measure reasoning tokens, agent actions, or optimizer steps.

Finding 2: Scale improves coherence on easy tasks, not hard ones

How does incoherence change with model scale? The answer depends on task difficulty:

  • Easy tasks: Larger models become more coherent
  • Hard tasks: Larger models become more incoherent or remain unchanged

This suggests that scaling alone won't eliminate incoherence. As more capable models tackle harder problems, variance-dominated failures persist or worsen.

Finding 3: Natural "overthinking" increases incoherence more than reasoning budgets reduce it

We find that when models spontaneously reason longer on a problem (compared to their median), incoherence spikes dramatically. Meanwhile, deliberately increasing reasoning budgets through API settings provides only modest coherence improvements. The natural variation dominates.

Finding 4: Ensembling reduces incoherence

Aggregating multiple samples reduces variance (as expected from theory), providing a path to more coherent behavior, though this may be impractical for real-world agentic tasks where actions are irreversible.

Why Should We Expect Incoherence? LLMs as Dynamical Systems

A key conceptual point: LLMs are dynamical systems, not optimizers. When a language model generates text or takes actions, it traces trajectories through a high-dimensional state space. It has to be trained to act as an optimizer, and trained to align with human intent. It's unclear which of these properties will be more robust as we scale.

Constraining a generic dynamical system to act as a coherent optimizer is extremely difficult. Often the number of constraints required for monotonic progress toward a goal grows exponentially with the dimensionality of the state space. We shouldn't expect AI to act as coherent optimizers without considerable effort, and this difficulty doesn't automatically decrease with scale.

The Synthetic Optimizer: A Controlled Test

To probe this directly, we designed a controlled experiment: train transformers to explicitly emulate an optimizer. We generate training data from steepest descent on a quadratic loss function, then train models of varying sizes to predict the next optimization step given the current state (essentially: training a "mesa-optimizer").

The results are interesting:

  • Incoherence grows with trajectory length. Even in this idealized setting, the more optimization steps models take (and get closer to the correct solution), the more incoherent they become.
  • Scale reduces bias faster than variance. Larger models learn the correct objective more quickly than they learn to reliably pursue it. The gap between "knowing what to do" and "consistently doing it" grows with scale.

Implications for AI Safety

Our results are evidence that future AI failures may look more like industrial accidents than coherent pursuit of goals that were not trained for. (Think: the AI intends to run the nuclear power plant, but gets distracted reading French poetry, and there is a meltdown.) However, coherent pursuit of poorly chosen goals that we trained for remains a problem. Specifically:

  1. Variance dominates on complex tasks. When frontier models fail on difficult problems requiring extended reasoning, there is a tendency for failures to be predominantly incoherent rather than systematic.
  2. Scale doesn't imply supercoherence. Making models larger improves overall accuracy but doesn't reliably reduce incoherence on hard problems.
  3. This shifts alignment priorities. If capable AI is more likely to be a hot mess than a coherent optimizer of the wrong goal, this increases the relative importance of research targeting reward hacking and goal misspecification during training—the bias term—rather than focusing primarily on aligning and constraining a perfect optimizer.
  4. Unpredictability is still dangerous. Incoherent AI isn't safe AI. Industrial accidents can cause serious harm. But the type of risk differs from classic misalignment scenarios, and our mitigations should adapt accordingly.

Conclusion

We use the bias-variance decomposition to systematically study how AI incoherence scales with model intelligence and task complexity. The evidence suggests that as AI tackles harder problems requiring more reasoning and action, its failures tend to become increasingly dominated by variance rather than bias. This doesn't eliminate AI risk—but it changes what that risk looks like, particularly for problems that are currently hardest for models, and should inform how we prioritize alignment research.

Acknowledgements

We thank Andrew Saxe, Brian Cheung, Kit Frasier-Taliente, Igor Shilov, Stewart Slocum, Aidan Ewart, David Duvenaud, and Tom Adamczewski for extremely helpful discussions on topics and results in this paper.


联系我们 contact @ memedata.com