AI and Chemistry is an increasingly popular topic for research and discussion. I thought it would be interesting to see how Large Language Models (LLMs) e.g. Gemini coped with carbon-13 NMR which is highly dependant on molecular symmetry. I asked first about cyclohexane and I got a near perfect answer – one peak due to high symmetry. For some reason I then asked about morpholine (another six -membered ring) and the trouble started.
Gemini offered a really poor chemical structure of morpholine albeit with a caveat about checking the results.
Then, when I used dictation, morpholine became morphine and a worse structure of morphine was offered.
I was not the first to draw attention to this shortcoming. Chem. Eng. News has a cautionary tale about AI and caffeine.
Following some discussion on BlueSky, I wanted to give Gemini a chance to do better with a text-based query. The results were impressive and I quote them with absolutely no subsequent edits.
ChemTube3D is a free, web-based resource developed by the University of Liverpool that provides interactive 3D chemistry animations and structures along with supporting information. It covers a wide range of topics relevant to advanced school chemistry and university chemistry courses.
Key features of ChemTube3D include:
ChemTube3D is a valuable resource for a wide range of individuals, including:
In summary, ChemTube3D is a powerful and free tool that utilizes interactive 3D visualizations to enhance the learning and teaching of chemistry for students and educators at various levels.
This would almost be suitable as a press release with no human effort!
I am told there are some good examples of additional tools that can overcome these limitations
https://www.nature.com/articles/s42256-024-00832-8
https://pubs.rsc.org/en/content/articlehtml/2025/sc/d4sc03921a but I haven’t been able to test them for myself.
Siri/ChatGPT was not so detailed.
In conclusion, my advice to our Education Secretary at this time would be not to use LLMs to mark Chemistry homework but colleagues in text-based subjects might find it very useful.