An Observation on Generalization
Do Large Language Models Perform the Way People Expect? Measuring the Human Generalization Function
What makes large language models (LLMs) impressive is also what makes them hard to evaluate: their diversity of uses. To evaluate these models, we must understand the purposes they will be used for. We consider a setting where these deployment decisions are made by people, and in particular, people's beliefs about where an LLM will perform well. We model such beliefs as the consequence of a human generalization function: having seen what an LLM gets right or wrong, people generalize to where else it might succeed. We collect a dataset of 19K examples of how humans make generalizations across 79 tasks from the MMLU and BIG-Bench benchmarks. We show that the human generalization function can be predicted using NLP methods: people have consistent structured ways to generalize. We then evaluate LLM alignment with the human generalization function. Our results show that -- especially for cases where the cost of mistakes is high -- more capable models (e.g. GPT-4) can do worse on the instances people choose to use them for, exactly because they are not aligned with the human generalization function.
Language model harnesses are compositional generalizers
Harnesses can lead to compositional generalization: we observe a property in training RLMs, in which similarly structured tasks are viewed as isomorphic and all individual LM calls in the harness become in-distribution.
Understanding Understanding: A Pragmatic Framework Motivated by...
Motivated by the rapid ascent of Large Language Models (LLMs) and debates about the extent to which they possess human-level qualities, we propose a framework for testing whether any agent (be it...
Empirical Generalization Research Papers - Academia.edu
View Empirical Generalization Research Papers on Academia.edu for free.
Generalised algebraic theories and contextual categories
Generalization Dynamics of LM Pre-training
Generalization Dynamics of LM Pre-Training shows that language models can switch abruptly between shallow pattern-following and more transferable computation during pre-training, even when loss and...
Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet
We find a diversity of highly abstract features. They both respond to and behaviorally cause abstract behaviors. Examples of features we find include features for famous people, features for countries and cities, and features tracking type signatures in code. Many features are multilingual (responding to the same concept across languages) and multimodal (responding to the same concept in both text and images), as well as encompassing both abstract and concrete instantiations of the same idea (such as code with security vulnerabilities, and abstract discussion of security vulnerabilities).
Mixture-of-Experts (MoE): The Birth and Rise of Conditional Computation
What two decades of research taught us about sparse language models...
The Lambek Calculus
There is a noticeable revival of categorial grammar these days, as a vehicle for linguistic description. The systems used differ somewhat from the original calculus of Ajdukiewicz and Bar-Hillel, however. In particular, there is a component of rules for ‘type change’ of expressions, making for greater flexibility and elegance. One fundamental system of this kind is the so-called ‘Lambek Calculus’, whose type-change rules show a close analogy with the inference rules of constructive propositional logic. In this paper, we present one calculus of this kind, and survey its theoretical properties as a device in linguistic semantics. Our two main new contributions are a new and complete semantics for this calculus, as well as a modest study of its language-accepting capacity. In this way, we hope to provide a better understanding of the background theory of flexible categorial grammar, in tandem with its descriptive uses.
Computation and its Connotations
A Review of Language Machines by Leif Weatherby
Recursive Language Models: the paradigm of 2026
How we plan to manage extremely long contexts
.png?v=c8c07d4bf43b)
Variety (cybernetics)
In cybernetics, the term variety denotes the total number of distinguishable elements of a set, most often the set of states, inputs, or outputs of a finite-state machine or transformation, or the binary logarithm of the same quantity.[1] Variety is used in cybernetics as an information theory that is easily related to deterministic finite automata, and less formally as a conceptual tool for thinking about organization, regulation, and stability. It is an early theory of complexity in automata, complex systems,[1]: 6  and operations research.[2]
Mapping the Mind of a Large Language Model
We have identified how millions of concepts are represented inside Claude Sonnet, one of our deployed large language models. This is the first ever detailed look inside a modern, production-grade large language model.
