Making Claude a Chemist

#ai #machinelearning #llm #programming

TL;DR

Anthropic is working with chemists to improve Claude's chemistry capabilities.
Claude's performance on NMR spectra is being examined by Anthropic chemist David Kamber.
Chemistry depends on correctly reading signals across different representations, such as hand-drawn structures and instrument readouts.
AI can help with the research burden in chemistry, but data availability has been a challenge.

Anthropic is collaborating with world-class synthetic, computational, and analytical chemists to enhance Claude's chemistry skills. This effort involves examining Claude's performance on a chemist's most common analytical input, an NMR spectrum, with Anthropic chemist David Kamber leading the work. The goal is to improve Claude's ability to understand and work with molecules, which is critical in chemistry.

What the data shows

The data shows that chemists need to move between different representations of molecules, such as hand-drawn structures, instrument readouts, and technical notations. Each representation demands a different kind of fluency, and understanding what molecule a chemist is working with is crucial. For example, a sketch of caffeine can help a chemist spot its resemblance to adenosine, but it cannot help distinguish it from other similar-looking molecules. The largest chemistry registry, CAS, catalogs over 290 million disclosed substances and grows by roughly 15,000 new ones every day.

What this means for AI readers

This means that AI tools like Claude need to be able to translate between different representations of molecules and understand the underlying chemistry. AI is well-positioned to take on the research burden in chemistry, but it still remains largely aspirational due to data availability challenges. Machine-learning tools have been positioned for years as transformative for retrosynthesis, reaction prediction, and property estimation, but the data needed has been hard to come by.

What to do right now

To improve Claude's chemistry capabilities, Anthropic is working with chemists to provide Claude with the necessary data and training. This includes examining Claude's performance on NMR spectra and other analytical inputs. By providing Claude with the right data and training, Anthropic aims to make Claude better at chemistry and able to assist chemists with their everyday work.

Bottom line

The bottom line is that improving Claude's chemistry capabilities requires collaboration between AI researchers and chemists. By working together, Anthropic can provide Claude with the necessary data and training to understand and work with molecules. This can help chemists with their everyday work and potentially lead to breakthroughs in fields such as medicine and materials science.

Frequently asked questions

Q: What is the goal of Anthropic's collaboration with chemists?

The goal is to improve Claude's chemistry capabilities and make it better at understanding and working with molecules.

Q: What is an NMR spectrum and why is it important?

An NMR spectrum is a type of analytical input that chemists use to understand the structure of molecules. It is important because it provides critical information about the molecule's composition and properties.

Q: How many disclosed substances are cataloged in the CAS registry?

The CAS registry catalogs over 290 million disclosed substances and grows by roughly 15,000 new ones every day.

Q: What is retrosynthesis and how can AI tools like Claude help with it?

Retrosynthesis is the process of working backward from a target molecule to simpler precursors to plan how to build it. AI tools like Claude can help with retrosynthesis by providing predictive models and suggesting potential synthesis routes.