Mystery AI Hype Theater 3000 Defining AGI: Oops! All Eugenics, 2025.12.08
Dec 30, 2025
A new definition of artificial general intelligence is under fire for its flawed assumptions and weak scientific backing. The hosts deconstruct its ties to eugenics while mocking dubious cognitive assessments and biased test choices. They also celebrate a union win for journalists protecting against unconsulted AI uses. Intriguing critiques of flashy AI promotions and shocking revelations about companies cheating on evaluations add flavor. Plus, don't miss their hilarious takes on AI-related absurdities like intrusive pop-up alerts!
AI Snips
Chapters
Transcript
Episode notes
Shoddy Mapping From Human Domains To AI
- The paper lists 10 cognitive domains but inconsistently expands to 12 and mixes human-specific and machine-specific interpretations.
- Alex and Emily call this sloppy and note the document lacks clear mappings from CHC constructs to AI tasks.
Tests Favor Tunable Algorithms, Not Cognition
- Many proposed tests (digit transformation, counting letters) map to trivial algorithmic tasks that models can be tuned to solve.
- Alex and Emily argue these tasks reflect test gaming rather than general cognitive ability.
General Knowledge Tests Are Culturally Biased
- The paper's 'general knowledge' examples borrow AP exams and U.S./European-centered curricula, embedding cultural bias.
- Emily highlights the resulting Eurocentric, US-centric framing and easy test leakage from study guides.