Augmented Language Models: a Survey

By Grégoire Mialon, Roberto Dessì, Maria Lomeli, Christoforos Nalmpantis, Ramakanth Pasunuru, Roberta Raileanu, Baptiste Rozière, Timo Schick, Jane Dwivedi-Yu, Asli Celikyilmaz and 3 more. 2023

579 citations recorded by the source corpus.

This survey reviews works in which language models (LMs) are augmented with reasoning skills and the ability to use tools. The former is defined as decomposing a potentially complex task into simpler subtasks while the latter consists in calling external modules such as a code interpreter. LMs can leverage these augmentations separately or in combination via heuristics, or learn to do so from demonstrations. While adhering to a standard missing tokens prediction objective, such augmented LMs can use various, possibly non-parametric external modules to expand their context processing ability, thus departing from the pure language modeling paradigm. We therefore refer to them as Augmented Language Models (ALMs). The missing token objective allows ALMs to learn to reason, use tools, and even act, while still performing standard natural language tasks and even outperforming most regular LMs on several benchmarks. In this work, after reviewing current advance in ALMs, we conclude that this new research direction has the potential to address common limitations of traditional LMs such as interpretability, consistency, and scalability issues.

Subfields: Artificial Intelligence, Large Language Models, Natural Language Processing

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 Large Language Models

In Natural Language Processing

Citations in the corpus

Llama 2: Open Foundation and Fine-Tuned Chat Models (2023, methodology): “Tool Use Emergence The integration of LLMs with tools is a growing research area, as highlighted in Mialon et al. (2023).”
GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (2023, background): “Moreover, recent studies reveal the potential of LLMs to program and control other digital tools, such as APIs, search engines, and even other generative AI systems (Schick et al., 2023; Mialon et al., 2023; Chase, 2022).”
C-Eval: A Multi-Level Multi-Discipline Chinese Evaluation Suite for Foundation Models (2023, background): “We underscore the importance of evaluating LLMs in such challenging settings, as current LLM development goes beyond creating a casual chatbot – it involves the development of complex systems or agents capable of interacting with various data types, receiving feedback, reasoning and using tools, and”
Executable Code Actions Elicit Better LLM Agents (2024, background): “…their action space expands beyond conventional text processing, allowing LLMs to acquire capabilities such as tool invocation and memory management (Mialon et al., 2023; Schick et al., 2023) and venture into real-world tasks such as controlling robots (Ahn et al., 2022; Huang et al., 2023; Ma et…”
AI literacy and its implications for prompt engineering strategies (2024, background): “Few-shot learning , which instructs an LLM to learn a new task with few examples, enables task delegation spontaneously and improves model performance (Mialon et al., 2023).”
When large language models meet personalization: perspectives of challenges and opportunities (2023, unknown): “LLMs, such as GPT, are well-suited for tool learning applications [217].”
Benchmarking and Defending against Indirect Prompt Injection Attacks on Large Language Models (2023, unknown): “Impact of LLM’s capability.”
Benchmarking Retrieval-Augmented Generation for Medicine (2024, background): “Many follow-up studies have been carried out to improve over the vanilla RAG (Borgeaud et al., 2022; Ram et al., 2023; Gao et al., 2023; Jiang et al., 2023; Mialon et al., 2023).”
Enhancing Retrieval-Augmented Large Language Models with Iterative Retrieval-Generation Synergy (2023, background): “, generated search queries Nakano et al. [2021], Press et al.”
MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback (2023, methodology): “To address complex tasks, Large Language Models (LLMs) often need multiple rounds of interaction with the users, sometimes aided by external tools (Schick et al., 2023b; ChatGPT Plugins; Mialon et al., 2023).”
Can LLM Already Serve as A Database Interface? A BIg Bench for Large-Scale Database Grounded Text-to-SQLs (2023, unknown): “Therefore, the development of methods that effectively combine the strong multi-step self-reasoning capabilities of LLMs with external knowledge reasoning coherently presents a promising future direction [29].”
The Effect of Sampling Temperature on Problem Solving in Large Language Models (2024, methodology): “For example, we can use techniques like chain-of-thought, tree-of-thought, self-criticism, and self-consistency to improve LLM performance [1,2].”
Reasoning with Language Model is Planning with World Model (2023, background): “In the realm of LLMs [22, 41, 6], reasoning typically entails decomposing complex questions into sequential intermediate steps (a.”
The Segment Anything Model (SAM) for Remote Sensing Applications: From Zero to One Shot (2023, background): “These models serve as the "foundation" for various applications [37, 39, 62].”
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing (2023, methodology): “[86] Grégoire Mialon, Roberto Dessì, Maria Lomeli, Christoforos Nalmpantis, Ram Pasunuru, Roberta Raileanu, Baptiste Rozière, Timo Schick, Jane Dwivedi-Yu, Asli Celikyilmaz, et al.”
AI Transparency in the Age of LLMs: A Human-Centered Research Roadmap (2023, background): “Models can also be augmented with the ability to call on external models, tools, or plugins [141], for example, querying an information retrieval system to ground their output or controlling and receiving feedback from a physical robot.”
Improving Alignment and Robustness with Circuit Breakers (2024, unknown): “LLM agents have been benchmarked [35, 45], but their safety and robustness remain unexplored.”