I’m a PhD student in a computational (life) science discpline but from a field that rarely has direct CS/ML/etc education. I have a whole thesis to research and write surrounding a niche disciplinary topic, which is difficult to get at even with 10+ years of experience before starting on my PhD.
I have had a lot of good experiences with the Data Analysis or Code Interperter functions available in GPT4 (and Github Copilot - but to a much lesser degree) to learn/automate/generate data for my PhD. I am a (big time air qoutes) “domain expert” for my field that really benefits from the fast coding solutions and subject matter specific education I can get from GPT4, but I am hitting the wall with queries very often. I would get more accounts, but that is not even possible with the sign up limit.
I can spaghetti code my way to a workable hack solution to most of my problems, but I simply have no idea how to work around GPT4 limits despite having access to 8+ A100s and 300ish CPU threads of free computing power via my lab. I mostly use this for bespoke computational software (physics based, not ML) in my lab, but there is a fair amount of downtime between the very large computational jobs we do.
Is there a way to get a DA/CI like functionality in the API or elsewhere? What I am really missing is having an LLM that can both process documents and compute code without direct intervention for niche specific problems. I have money and computing resources to throw at this, but I can’t figure out where to start.
I’ll happily provide some knowledge in my domains in exchange