The automation of natural language processing brought to us by Language Language Models (LLMs) over the past few years has had far-reaching implications across many industries. It is currently being applied to various NLP applications with impressive few-shot and zero-shot results. More recently, progress has been made based on the Transformer architecture originally developed for neural machine translation.
Nonetheless, it is important to remember that LLM has limitations and problems in learning elementary arithmetic and chemical calculations. These shortcomings are due to the basic structure of the model, which centers on next-word prediction. One way to overcome these limitations is to supplement the extensive language model with additional third-party software.
Artificial intelligence (AI) systems designed by experts to tackle specific problems are impacting the field of chemistry, especially reaction prediction, retrosynthetic planning, molecular property prediction, materials design, and most recently, Bayesian optimization. Code generation LLMs have demonstrated some understanding of chemistry due to the nature of their training12. The highly experimental and sometimes artisanal nature of chemistry and the limited scope and applicability of computational tools, even within a given domain. Tools such as RXN for Chemistry and AIZynthFinder, where integration is common, are examples of closed setups made possible by corporate mandates that prioritize integration and internal use.
ChemCrow, an LLM-powered chemical engine inspired by similar successful applications in other fields by researchers at the Institute for Artificial Chemical Intelligence (LIAC), the National Center for Research Competence (NCCR) Catalyst, and the University of Rochester I’d like to introduce_______ It aims to simplify the reasoning process for many typical chemical tasks in fields such as pharmaceutical and materials design and synthesis. By providing task-specific and format-specific prompts for LLM (GPT-4 for exams), ChemCrow harnesses the power of a wide range of tools designed by chemistry-focused experts. LLMs are given a list of tools, a brief description of their purpose, and information about data input and output.
The model is instructed to use thought, action, behavioral inputs, and observed patterns. This forces you to think about the current state of your task and how it relates to your end goal, and plan how to proceed next. In parallel with this preprint, 46 details a similar strategy for equipping LLMs with chemistry-specific functions. The LLM then prompts for an action and for this action (using the keyword “inferred action you just completed in the thought step”). The search for the relevant features is resumed, and the results are prefixed with the word “observation” and sent back to the LLM, which then repeats the previous step “think”.
In this way, LLM moves from a confident, if sometimes erroneous, source of information to a thought engine that observes and reflects on its observations and takes appropriate action based on what it has learned. evolve. Researchers have deployed 13 different tools to aid research and discovery. The team admits that the toolset they are given is not comprehensive. Simply provide a tool, describe its purpose in natural language, and easily extend it to new uses. ChemCrow empowers professional chemists and chemists without specialized training in the field by providing a user-friendly interface to authoritative chemical information.
This paper evaluates ChemCrow’s capabilities across 12 different usage scenarios, including target molecule synthesis, safety control, and discovery of compounds with similar mechanisms of action. In the LLM-based assessment, GPT-4 and ChemCrow were found to be nearly equally effective on completeness and quality of thought. In contrast, human ratings found ChemCrow to be nearly 4.4/10 points ahead of his GPT-4, with him significantly outperforming him by 2.75/10 on task success.
check out paperdon’t forget to join Our 19k+ ML SubReddit, cacophony channeland email newsletterWe share the latest AI research news, cool AI projects, and more. If you have any questions about the article above or missed something, feel free to email me. Asif@marktechpost.com
🚀 Check out 100 AI Tools in the AI ​​Tools Club
Tanushree Shenwai is a consulting intern at MarktechPost. She is currently pursuing her bachelor’s degree at the Indian Institute of Technology (IIT), Bhubaneswar. She is a data her science enthusiast and has a keen interest in the scope of artificial intelligence applications in various fields. Her passion lies in exploring new advancements in technology and its practical applications.
