Conjectures
Bold, falsifiable ideas open for rational criticism
Filter by tag:
AI detection tools have a false positive rate less than 2%.
Context and Rationale: AI-generated content detectors, particularly Turnitin’s AI detection system, are now widely used in academic settings. Howev...
Bloodline Memory is still plausible in humans.
I'm coming out of left field so I'll define some terms: Bloodline Memory: Information carried thru a biological liquid media perhaphs as waves, ...
Peer Review Reproducibility
Peer review without anonymous refereeing yields lower replication reproducibility in experimental psychology than double-blind review processes.
Protein Folding MAE
No current state-of-the-art protein-folding algorithm can predict side-chain conformations with a mean absolute error below 0.5 Å across all twenty...
Vertical Farm Carbon
Vertical farming of leafy greens cannot achieve a lifecycle carbon intensity lower than conventional open-field production when full energy inputs ...
Transformer Bias
Transformer architectures cannot acquire systematic compositional generalisation to novel argument–adjunct pairings in natural language without exp...
Charity Effectiveness
Active GiveWell top charities deliver at least ten times more DALYs per dollar than the median global health NGO because of targeted cost-effective...
Oral mRNA Barrier
A single oral mRNA therapy cannot achieve systemic protein expression levels comparable to intramuscular mRNA delivery because of gastrointestinal ...
LLM Proofs
Large language models without explicit training on formal logic cannot construct a valid novel proof in Zermelo–Fraenkel set theory containing more...
Digital Minds
Digital minds implemented on classical computers are not moral patients because phenomenal consciousness requires biological substrates.
Insulin Tech Limit
Closed-loop insulin delivery with continuous glucose monitoring cannot drive mean HbA1c in adult type-1 diabetics below 6 percent without adjunct b...
LiFePO4 Ceiling
Lithium-iron-phosphate battery chemistry cannot surpass a gravimetric energy density of 250 Wh kg⁻¹ at the commercial cell level because of intrins...
Protein NMR Limit
Diamond nitrogen-vacancy centre sensors cannot resolve nuclear magnetic resonance signals from a single protein molecule at room temperature in amb...
LLM Contradictions
Self-supervised transformer language models trained only on next-token prediction cannot achieve more than 95 percent accuracy at detecting single-...
Self-supervised transformer language models trained only on next-token prediction cannot achieve more than 95 percent accuracy at detecting single-step logical contradictions in arbitrarily long formal proofs.
Present a peer-reviewed evaluation showing a model that meets or exceeds 95 percent accuracy under the stated training constraint.
AGI Requires Embodiment
Artificial General Intelligence cannot emerge from purely digital systems processing symbolic information. True intelligence requires sensorimotor ...
Free Will is an Illusion
What we perceive as free will is merely the experience of making choices without awareness of the deterministic neurological and environmental proc...
Consciousness Requires Biological Substrate
Consciousness, as subjectively experienced by humans, is fundamentally dependent on biological processes and cannot emerge from non-biological syst...