Future of Scientific Problem Solving? Sharing Pre-Built AI Solvers Like Recipes

An interesting find. In scientific computing, I've learned to accept the performance/accuracy as it, slow and iterative, many runs being similar - Would pre-built solver templates with AI models reduce repetitive compute?

There are commercial software to solve domain-specific problems, but seeing how AI is starting to show significant performances in domains such as weather forecasting, one wonders what value can AI bring for other domains? Are there other scientific domains where this concept could be applied?

I've yet to see established open-source solvers working to incorporate AI.

What are your pain points in scientific computing?

I would definite like to see a future where solving these is as easy and ubiquitous as LLM are now.