The conference I'm at right now is extremely AI focused, and it makes me so mad to see how bad consumer versions of AI (i.e. ChatGPT, Gemini, Copilot, etc...) are after seeing how Gen AI/LLMs are being used in science and in enterprise IT. It's not even close to how accurate and powerful these are compared to even the pro plans of what OpenAI and Google offer (even though they use the same models).
I saw a short presentation on what one of the other labs is doing with gen AI/LLMs and how its use evolved since they started actively using it in 2019 and it's so fascinating. https://www.anl.gov/ai
Even simple things like refactoring code to run on an HPC to be more energy efficient has resulted in insane realistic savings without impacting performance.
It makes me want to focus more on AI, but it really is a full-time job and I don't have the cycles to spare at work aside from maintaining the infrastructure for it for others to use.
@KuJoe I'd be very skeptical of such claims of results.
I'm reminded of the hype around blockchain tech, which in B2B was accompanied by a lot of claims about savings and efficiency gains as well, but then when you looked closer it was usually a subtle but total misrepresentation of what actually happened, and in the end none of it actually worked as people claimed.
My bet on the 'savings' here are that either a) this wasn't actually LLMs or the current generation of "AI" and rather older, more specialized systems, b) it *was* the current-gen stuff but with so much manual guidance that it was ultimately still just humans doing the work, or c) the measurement itself was selected to show positive outcomes, regardless of real-world effects.
(B2B claims are usually only looked at by people on the management layer and not by the subject matter experts, so if you sound confident enough you can pretty much just claim whatever you want, it doesn't need to stand up to scrutiny)
