Least-to-Most Prompting Enables Complex Reasoning in Large Language Models

Zhou, Denny; Schärli, Nathanael; Hou, Lei; Lee, Jason; Scales, Nathan; Wang, Xuezhi; Schuurmans, Dale; Cui, Claire; Bousquet, Olivier; Le, Quoc V.; Chi, Ed

doi:10.48550/arxiv.2205.10625

Cited by 46 publications

(64 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further, while the core of RTNet does not include recurrency, the evidence accumulation system can be thought of as a recurrent network. In fact, recent studies demonstrated the advantages of combining a standard feedforward network with a recurrent network in performing a range of tasks and extrapolating to solve problems of greater complexity than they were trained on 47,48 . Thus, while RTNet remains less biologically plausible than a true recurrent network, it is as biologically plausible as current methods of training neural networks permit.…”

Section: Biological Plausibility Of Rtnet and Anytime Predictionmentioning

confidence: 99%

RTNet: A neural network that exhibits the signatures of human perceptual decision making

Rafiei

Rahnev

2022

Preprint

View full text Add to dashboard Cite

Convolutional neural networks currently provide the best models of biological vision. However, their decision behavior, including the facts that they are deterministic and use equal number of computations for easy and difficult stimuli, differs markedly from human decision-making, thus limiting their applicability as models of human perceptual behavior. Here we develop a new neural network, RTNet, that generates stochastic decisions and human-like response time (RT) distributions, and also reproduces all foundational features of human accuracy, RT, and confidence. To test RTNet’s ability to predict human behavior on novel images, we collected accuracy, RT, and confidence data from 60 human subjects performing a digit discrimination task. We found that the accuracy, RT, and confidence produced by RTNet for individual novel images correlated with the same quantities produced by human subjects. Critically, human subjects who were more similar to the average human performance were also found to be closer to RTNet’s predictions. Overall, RTNet is the first neural network that exhibits all basic signatures of perceptual decision making.

show abstract

Section: Biological Plausibility Of Rtnet and Anytime Predictionmentioning

confidence: 99%

RTNet: A neural network that exhibits the signatures of human perceptual decision making

Rafiei

Rahnev

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Large language models exhibit impressive zero-shot reasoning capabilities: from planning [14] to writing math programs [43]; from solving science problems [44] to using trained verifiers [45] for math word problems. These can be improved with prompting methods such as Least-to-Most [46], Think-Step-by-Step [15] or Chain-of-Thought [47]. Most closely related to this paper are works that use LLM capabilities for robot agents without additional model training.…”

Section: Perception Apis Control Apismentioning

confidence: 99%

Code as Policies: Language Model Programs for Embodied Control

Liang¹,

Huang²,

Xia³

et al. 2022

Preprint

View full text Add to dashboard Cite

Large language models (LLMs) trained on codecompletion have been shown to be capable of synthesizing simple Python programs from docstrings [1]. We find that these codewriting LLMs can be re-purposed to write robot policy code, given natural language commands. Specifically, policy code can express functions or feedback loops that process perception outputs (e.g., from object detectors [2], [3]) and parameterize control primitive APIs. When provided as input several example language commands (formatted as comments) followed by corresponding policy code (via few-shot prompting), LLMs can take in new commands and autonomously re-compose API calls to generate new policy code respectively. By chaining classic logic structures and referencing third-party libraries (e.g., NumPy, Shapely) to perform arithmetic, LLMs used in this way can write robot policies that (i) exhibit spatial-geometric reasoning, (ii) generalize to new instructions, and (iii) prescribe precise values (e.g., velocities) to ambiguous descriptions ("faster") depending on context (i.e., behavioral commonsense). This paper presents code as policies: a robot-centric formalization of language model generated programs (LMPs) that can represent reactive policies (e.g., impedance controllers), as well as waypoint-based policies (vision-based pick and place, trajectory-based control), demonstrated across multiple real robot platforms. Central to our approach is prompting hierarchical code-gen (recursively defining undefined functions), which can write more complex code and also improves state-of-the-art to solve 39.8% of problems on the HumanEval [1] benchmark. Code and videos are available at https://code-as-policies.github.io # do it again but faster, to the left, and with a banana. while not detect_object("banana"):robot.set_velocity(x=0, y=-0.2, z=0)Representing code as policies inherits a number of benefits from LLMs: not only the capacity to interpret natural language, but also

show abstract

“…Prior work has also shown that by using specific prompting language such as "Let's think step by step", one can solicit reasoning from the model to perform tasks that require logical reasoning, such as solving math problems [23], in a zero-shot setting. In addition, various prompting paradigms have been proposed to solicit reasoning from the language model [47,49,54]. For example, chain-of-thought prompting [49] proposes to use the models to generate intermediate results (i.e., chain of thoughts) before generating the final output.…”

Section: Prompting Pre-trained Large Language Modelsmentioning

confidence: 99%

“…Pre-trained LLMs support in-context few-shot learning via prompting-instead of finetuning or re-training models for each new task, one can prompt an LLM with a few input and output exemplars of the desired task [6,9,49,54]. For some NLP tasks such as questionanswering or translation, prompting can perform on par with previous benchmark approaches [6].…”

Section: Prompting Large-language Models For Mobile Ui Tasksmentioning

confidence: 99%

Enabling Conversational Interaction with Mobile UI using Large Language Models

Wang¹,

Li²,

Li³

2022

Preprint

View full text Add to dashboard Cite

Conversational agents show the promise to allow users to interact with mobile devices using language. However, to perform diverse UI tasks with natural language, developers typically need to create separate datasets and models for each specific task, which is expensive and effort-consuming. Recently, pre-trained large language models (LLMs) have been shown capable of generalizing to various downstream tasks when prompted with a handful of examples from the target task. This paper investigates the feasibility of enabling versatile conversational interactions with mobile UIs using a single LLM. We propose a design space to categorize conversations between the user and the agent when collaboratively accomplishing mobile tasks. We design prompting techniques to adapt an LLM to conversational tasks on mobile UIs. The experiments show that our approach enables various conversational interactions with decent performances, manifesting its feasibility. We discuss the use cases of our work and its implications for language-based mobile interaction.

show abstract

Least-to-Most Prompting Enables Complex Reasoning in Large Language Models

Cited by 46 publications

References 13 publications

RTNet: A neural network that exhibits the signatures of human perceptual decision making

RTNet: A neural network that exhibits the signatures of human perceptual decision making

Code as Policies: Language Model Programs for Embodied Control

Enabling Conversational Interaction with Mobile UI using Large Language Models

Contact Info

Product

Resources

About