sarahconstantin.substack.com/p/things-that-go-boom

This sentence misidentifies AGM-158B JASSM-ER as an anti-ship missile.

Official manufacturer and U.S. Air Force sources describe JASSM/JASSM-ER as a Joint Air-to-Surface Standoff Missile for air-to-ground strikes against defended fixed and relocatable targets:

• Lockheed Martin's JASSM product page says JASSM provides long-range precision engagement capabilities for air-to-ground missions and is designed to destroy fixed and relocatable targets.
• The U.S. Air Force budget document for missile procurement identifies the AGM-158B as part of the JASSM Extended Range family, i.e. the Joint Air-to-Surface Standoff Missile.
• An Air Combat Command article likewise calls JASSM-ER "an autonomous, long-range, air-to-ground precision missile."

So the article's wording is incorrect: AGM-158B JASSM-ER is the land-attack / air-to-surface variant, not an anti-ship missile. The anti-ship variant in the AGM-158 family is AGM-158C LRASM, which the article correctly lists separately in the previous bullet.

This sentence gets the reopening date wrong.

The Department of Defense announced on August 10, 2022 that its investment would allow the GOEX facility to reopen after an accident shut down production — wording that indicates the plant had not yet reopened at that point.

Estes Energetics later announced that it purchased the plant in January 2022, but that U.S. black powder production restarted on June 1, 2023, with first customer shipments in August 2023.

So 2022 was the year of the acquisition/rebuild effort, not the year the plant actually resumed production.

The parenthetical claim is technically incorrect.

In machine learning, a standard Transformer is a parametric neural network: it learns a fixed set of weights/parameters during training and uses those learned parameters at inference time. Researchers often contrast transformers with genuinely nonparametric components such as nearest-neighbor retrieval or episodic memory.

Two examples from the literature make this distinction explicit:

• Adaptive Semiparametric Language Models describes its approach as combining "a large parametric neural network (i.e., a transformer) with a non-parametric episodic memory component."
• An ICLR paper on nonparametric masked language modeling states that "most current neural LMs are based on parametric neural networks, using RNN or Transformer architectures."

So while a transformer can be used in settings involving nonparametric methods, the transformer itself is not "really" a nonparametric model.