My Galactica model achieved 68.2 % accuracy on LaTeX equation probes, outperforming GPT‑3’s 49.0 %.
On MMLU‑Math benchmarks Galactica scored 41.3 % versus Chinchilla’s 35.7 %.
On the PaLM 540B MATH benchmark Galactica reached 20.4 % compared with 8.8 % for PaLM 540B.
I was awarded the Phi Tau Phi Scholastic Honor in June 2019 for academic excellence during my doctoral studies.
The Dynamic Influence Keyword Model detects human‑inferred interests 94 % of the time.