Conjectures

Bold, falsifiable ideas open for rational criticism

Filter by tag:

Clear filter
Showing conjectures tagged:
ai
Active Ξ0.0405
3 months ago

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...

By Capt. Sol Hando, SolHando Inc.
0 refutations View
Active
3 months ago

Transformer Bias

Transformer architectures cannot acquire systematic compositional generalisation to novel argument–adjunct pairings in natural language without exp...

By Anonymous User
0 refutations View
Active
3 months ago

LLM Proofs

Large language models without explicit training on formal logic cannot construct a valid novel proof in Zermelo–Fraenkel set theory containing more...

By Anonymous User
0 refutations View
Active
3 months ago

RLHF Misgeneralization

Reinforcement Learning from Human Feedback consistently produces models that exhibit goal mis-generalisation when exposed to novel adversarial inst...

By Anonymous User
0 refutations View
Active
3 months ago

LLM Contradictions

Self-supervised transformer language models trained only on next-token prediction cannot achieve more than 95 percent accuracy at detecting single-...

By Anonymous User
0 refutations View
Refuted
4 months ago

AGI Requires Embodiment

Artificial General Intelligence cannot emerge from purely digital systems processing symbolic information. True intelligence requires sensorimotor ...

By Anonymous User
1 refutation View
Active
4 months ago

Consciousness Requires Biological Substrate

Consciousness, as subjectively experienced by humans, is fundamentally dependent on biological processes and cannot emerge from non-biological syst...

By Anonymous User
1 refutation View