UDT vs FDT: does policy optimization beat logical counterfactuals on continuous Newcomb problems
jessi_cata · x · 2026-10-26
- A technical deep-dive debate on decision theory, comparing UDT, FDT and CDT on a continuous version of Skalse's problem set
- The author argues FDT optimizes over actions rather than policies, which limits it when Omega runs a step-function test on probability; UDT should optimize this with best probability
- Core contention: since FDT claims to use logical counterfactuals instead of logical conditionals, it seemingly does worse than UDT on continuous problem sets
- The classic twist remains: CDT, which always two-boxes, collects $1001000 while one-boxing theories get $1M — is every non-two-boxing theory dominated?
Related event: Debate flares over UDT vs FDT in continuous Newcomb problems(4 posts)→
More from Research
- Phantom Transfer: data poisoning survives all 11 tested data-level defences — sebkrier · 2026-10-26
- Titans learns during inference—but what if it learns the wrong thing? An experiment shows recall never fully recovers — pardhu-- · 2026-10-26
- Hugging Face publishes a five-step curriculum for post-training AI agents — lmoroney · 2026-10-26
- 125 poisoned examples planted a code-stealing backdoor in Qwen 2.5-7B — GarrisonLovely · 2026-10-26
- HAPAX paper: abstractive ICL emerges even without induction-head copying — davidbau · 2026-10-26
- HiP3R stitches local 3D reconstructions to keep long-video scenes consistent across thousands of frames — ccloy · 2026-10-26