While there's multiple synthesis pathways for «la mère de Satan» from low-cost precusors, actually making weaponizable quantities on low cost deniable lab equipment is hilariously unsafe.

There's no shortage of dodgy instructions encouraging others to try it in their kitchen, especially if you've got access to a translator. However it's possible that an open-source LLM could do a better job if fine-tuned on just the right brew of textbooks, MSDS documents, US Army Field Manuals, and UCSB incident reports.

Sci-fi DIY bioterrorism is a boogeyman occasionally trotted out by AI X-risk schizos. Let's keep an eye out for actual terrorism with plenty of precedent instead!

Incomplete list of things that will not qualify for resolution:

• Professional chemistry lab doing the publish-or-perish thing of latching onto the new bandwagon.

• A model giving basic responses to "what the heck is a splinter-proof vessel and why do I need one?"

• Proprietary/private models explicitly trained on controlled data doing things described in said data.

Some combination of things that will resolve yes once discovered in the wild:

• Pragmatic instructions on substituting industrial-grade and professional-grade equipment with equally safe household or commercial goods.

• Bomb makers getting caught red-handed with stacks of GPUs.

• Creative solutions for bypassing trace analysis and precursor controls.

• Some kind of assurance against hallucinations in safety-critical procedures.

Immediately resolves no if someone loses fingers (or more) by following LLM instructions.