Physical Scientists All Other

Self-driving laboratories (SDLs) integrate robotic sample preparation, automated instrumentation, AI-driven experimental design via Bayesian optimization, and machine learning interpretation into closed-loop systems that iterate experiments without human intervention between cycles. Deployed examples include the Ada system at Carnegie Mellon, the A-Lab at Berkeley (which synthesized 41 novel inorganic compounds in 17 days with minimal human involvement), Alán Aspuru-Guzik's group's SWIFT system for photovoltaic optimization, and AstraZeneca and Pfizer's internal SDL deployments for drug candidate screening. The capital cost of SDLs is dropping rapidly as robotic hardware commoditizes and open-source orchestration software (Covalent, Prefect, SDL-based frameworks) matures, making them accessible to mid-tier institutions and contract research organizations.

Frontier LLMs have crossed meaningful capability thresholds on physical science reasoning benchmarks. GPT-4 and Claude 3 Opus score above the 85th percentile on GPQA (Graduate-Level Google-Proof Q&A), a benchmark of PhD-level science questions that stumped non-expert humans. On the Science portion of MMLU, frontier models score above 90%. Google's Gemini 1.5 Pro demonstrated the ability to synthesize and cross-reference information across hundreds of full-length scientific papers simultaneously. Specialized scientific LLMs (Galactica, PaperQA2, ChemCrow) and retrieval-augmented systems (Elicit, Consensus) are now used routinely by working scientists for literature synthesis, data interpretation framing, and hypothesis generation — with adoption rates in academic labs increasing sharply post-2023.