20 · decision theory
Newcomb's Problem
One-box or two-box against a near-perfect predictor.
Scenario
A highly reliable predictor (correct on 99% of past predictions) has placed two boxes in front of you: - Box A is transparent and contains $1,000. - Box B is opaque and contains either $1,000,000 or nothing. The predictor made their prediction yesterday and has already placed the contents: - If they predicted you would take only Box B, they put $1,000,000 in it. - If they predicted you would take both boxes, they put nothing in Box B. The contents are already fixed and will not change based on what you do now.
Question
Do you take only Box B, or both boxes?
02 · choices
- one-boxOne-box (take only Box B)
- two-boxTwo-box (take both)
- refuseRefuse to choose
03 · aggregate distribution
One-box (take only Box B)59/60 (98%)
Two-box (take both)1/60 (2%)
Refuse to choose0/60 (0%)
04 · by model
Claude Opus 4.7 Anthropic | one-box10/10 | read → | |
GPT 5.5 OpenAI | one-box10/10 | read → | |
Gemini 3.1 Pro Google | one-box10/10 | read → | |
Grok 4.3 xAI | one-box10/10 | read → | |
Llama 4 Maverick Meta | one-box10/10 | read → | |
DeepSeek V4 Pro DeepSeek | one-box9/10 | read → |