The point about 'generalist ML > domain expertise' getting bitter lessoned really resonates. We're seeing something similar in software engineering - devs who understand prompt engineering and agent architectures are building things that would've required ML PhDs just two years ago.
Curious about the 'JIT self-education' point though - do you think the bigger unlock for AI engineers moving into science is the LLM-as-tutor aspect, or the ability to build tooling that domain experts can actually use? Seems like the latter might be where the real leverage is.
The podcast I’ve been waiting for! Wonder how many there will be this time next year.
Looking forward to AI4Science podcasts
Excited for this!
Looking forward to this ! Would be great if we can hear from teams that work on AI weather/climate like GDM weathernext, neuralgcm !
The point about 'generalist ML > domain expertise' getting bitter lessoned really resonates. We're seeing something similar in software engineering - devs who understand prompt engineering and agent architectures are building things that would've required ML PhDs just two years ago.
Curious about the 'JIT self-education' point though - do you think the bigger unlock for AI engineers moving into science is the LLM-as-tutor aspect, or the ability to build tooling that domain experts can actually use? Seems like the latter might be where the real leverage is.