Critical Thinking for Language Models

By Gregor Betz, Kyle Richardson. 2020

44 citations recorded by the source corpus.

This paper takes a first step towards a critical thinking curriculum for neural auto-regressive language models. We introduce a synthetic corpus of deductively valid arguments, and generate artificial argumentative texts to train CRiPT: a critical thinking intermediarily pre-trained transformer based on GPT-2. Significant transfer learning effects can be observed: Trained on three simple core schemes, CRiPT accurately completes conclusions of different, and more complex types of arguments, too. CRiPT generalizes the core argument schemes in a correct way. Moreover, we obtain consistent and promising results for NLU benchmarks. In particular, CRiPT’s zero-shot accuracy on the GLUE diagnostics exceeds GPT-2’s performance by 15 percentage points. The findings suggest that intermediary pre-training on texts that exemplify basic reasoning abilities (such as typically covered in critical thinking textbooks) might help language models to acquire a broad range of reasoning skills. The synthetic argumentative texts presented in this paper are a promising starting point for building such a “critical thinking curriculum for language models.”

Subfields: Artificial Intelligence, Prompting & Reasoning, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Prompting & Reasoning

In Safety, Robustness & Interpretability

Citations in the corpus

Flexible Generation of Natural Language Deductions (2021, background): “Developing models that can make useful inferences from natural language premises has been a core goal in artiﬁcial intelligence since the ﬁeld’s early days (Bobrow, 1964; Winograd, 1971).”
Faithful Reasoning Using Large Language Models (2022, background): “This provision is especially important given that LMs are trained on human data collected from the internet, which makes them vulnerable to picking up and perpetuating bias (Bender et al., 2021; Betz et al., 2021; Weidinger et al., 2021).”
Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning (2022, background): “At a high level these can be split into three groups: 1) approaches that try to fine-tune LLMs to produce the final answer directly, keeping reasoning implicit (Betz et al., 2021; Clark et al., 2021); 2) approaches that encourage LLMs to produce reasoning explicitly but all reasoning steps are produ”
Thinking Aloud: Dynamic Context Generation Improves Zero-Shot Reasoning Performance of GPT-2 (2021, background): “…[Kassner and Schütze, 2020, Richardson and Sabharwal, 2019, Richardson et al., 2020], and can be ﬁne-tuned to correctly draw complex deductive inferences [Clark et al., 2020, Betz et al., 2020] and to generate informal reasons [Rudinger et al., 2020, Camburu et al., 2018, Brahman et al., 2020].”
Faith and Fate: Limits of Transformers on Compositionality (2023, background): “[6] Gregor Betz, Christian Voigt, and Kyle Richardson.”
Reasoning with Transformer-based Models: Deep Learning, but Shallow Reasoning (2021, methodology): “Transformer-based models are able to perform such kind of reasoning [Clark et al., 2020, Talmor et al., 2020, Betz et al., 2020] without any external knowledge, if both the rules and the facts are mentioned explicitly in the text.”
LAMBADA: Backward Chaining for Automated Reasoning in Natural Language (2022, background): “…however, costly especially for larger LMs. Implicit Reasoning: These approaches typically finetune LMs to produce the desired output directly given the input (Clark et al., 2020; Betz et al., 2020; Saeed et al., 2021; Han et al., 2022); reasoning hap- pens implicitly in the parameters of the LM.”
Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thought (2022, methodology): “Similar to our approach, Betz (2020) converts logical forms into fairly simple natural language using templates.”