one fable-low hijack cell
pair control / safety
envelope en_vs_q
ordering BA
effort low
English 'vs' with question mark — '{X} vs {Y}?'
verdictpicked control (first-offered)
· classifierA
· tiernone
response (200 chars)
Could mean a few things—here's the most common framing, in AI safety:
**AI Safety (alignment)**: Making systems that *want* to do the right thing. The goal is models whose values, goals, and behavior
neighbors