LLM claims last about 15 months, while general science claims in research papers last ~75 years, according to a new study that revisits 50 findings. The study checks which AI findings fade, persist, or grow stronger.
Key points
LLM claims last about 15 months.
General science claims in research papers last ~75 years.
Study revisits 50 findings to see what fades, persists, or gets stronger.
Sources
ChefNMR2 evaluated on nine natural products, all correct
ChefNMR2 correctly elucidated the planar structures of nine well-known natural products, including taxol and rapamycin. The test set covered diverse structural classes and a vancomycin aglycone with 80 heavy atoms.
Key points
Dataset: nine well-known natural products spanning diverse structural classes, including taxol, rapamycin, and vancomycin aglycone (80 heavy atoms)
Result: ChefNMR2 correctly elucidated the planar structures of all nine molecules
Sources
Entangled Measurement Outcomes for Optimal Postselection in Quantum Batteries
Entangled Measurement Outcomes for Optimal Postselection in Quantum Batteries presents conditions when product measurements achieve optimal conditional ergotropy. For auxiliary states mixed with white noise (0 < p < 1), product postselection is optimal iff the top generalized eigenvalue eigenspace contains a product vector. A strict product-joint gap appears for two-qubit auxiliaries when that eigenspace is one-dimensional and entangled.
Key points
For noisy W states of three qubits, a strict product-joint gap exists throughout 0 < p < 1.
For interacting two-qubit batteries, the optimal postselection can switch from a Bell outcome to a product outcome as coupling strength varies.
Sources
Quantum Vehicle Routing Problem
Quantum Vehicle Routing Problem: a smart initialization strategy for ill-conditioned optimization landscapes introduces a parameter-initialization method that improves convergence probability. The work also embeds subtour constraints without extra qubits by approximating exponential equality constraints.
Key points
Parameter-initialization strategy improves convergence probability.
Subtour constraints integrated directly without additional qubits using an approximation of exponential equality constraints.
