Things I am excited about

This is an unorganized list of things I am excited about, for some I have plenty of ideas for others I don’t:

• I would like to successfully combine LLMs with formal approaches
• I want to adapt enumerative approaches for ILP and SMT solvers, there seems to be huge gains to make
• I want to make ranked enumeration on the GPU works blazingly fast
• I want to take a look at super optimisation of PTX (GPU assembly) or assembly
• I would really like to look at automatic debloating of software
• I would like to try and make an agentic framework reason about competitive video games to figure out things automatically in a scientific based way
• Minimising variability
• Automated theorem prover
• Parallelising program synthesis
• Using diffusion like approaches for code generation
• Use property testing to prune the search space of code generation
• I have plenty of ideas and not enough time to develop and experiment with them so can we make an agent that helps speeding up this process?
• Given a codebase and a CI/CD pipeline, now a new commit is pushed can you select only the necessary tests to run? And maintain it across time?
• Let’s learn policies via reinforcement learning in the form of programs, current approaches do not scale at all

I would also like to try working in a larger team with ambitious goals a bit like what DeepMind did for AlphaGo which made progress in fundamental research.