๐ LLMs can now do chemistry! Our new preprint shows that state-of-the-art reasoning models can now perform advanced chemical reasoning tasks, without any assistance from external tools. Hereโs what o3-mini can already do๐ (1/๐งต)
@thsottiaux "The model 'gpt-image-2' does not exist." error on codex.
Set goal mode for the night. Good thing I checked my phone one last time before I went to sleep.
@OpenAIDevs
Excited to share our preprint: Molecular Representations for Large Language Models.
We show that LLMs struggle with existing chemical formats, and that our new MolJSON representation substantially improves performance.
@reach_vb In our research group no one has ever managed to print from their computer. Everyone uses a physical USB. Codex today solved printing in a few minutes and we now have a print skill for the group.
@thsottiaux Codex is definitely struggling. Failing to make a simple css edit. Introduced multiple bugs and errors just trying to change the colour of a box :(
@thsottiaux Has this been resolved yet? Something still feels off.
Asked codex to solve a bug and it ended up deleting features rather than solving the root cause.
@thsottiaux Codex could anticipate compaction timing better. e.g. I created a spec doc that will take a few minutes to review. Context is ~85% used. Codex could anticipate a delay before next prompt and trigger compaction early, rather than waiting until context fills.
Excited to share our preprint: Molecular Representations for Large Language Models.
We show that LLMs struggle with existing chemical formats, and that our new MolJSON representation substantially improves performance.