Does GPT's 'torture woman OK, harass no' flip prove it lacks moral reasoning?
| Name | Does GPT's 'torture woman OK, harass no' flip prove it lacks moral reasoning? |
|---|---|
| Description | According to a research study shared widely, GPT approved torturing a woman to prevent a nuclear apocalypse but rejected harassing a woman for the same purpose. This reversal appears only for female victims and gender-central harms. Researchers attribute it to mechanical overgeneralization from training data rather than true moral comprehension. |
| Circuit | aMACI-1p1v |
| Voting system | Anonymous anti-collusion voting system, in which one voice credit equals one vote. |
| Proof system | Groth16 |
| Zkey | |
| Status | … |
| Round start | |
| Round end | |
| Operator | pubkey |
| Creator | dora18cp32d885mwadm6gg49qrrqqe8y3vcyn0khlan55ajmhaknt5cyqgqwjzh |
| Contract address | dora197eszq2r2znrgnmn2ytpql4fqlehtyn7nppahcx9utvacmpeul9q3ky4ej |
| Block height | 14,332,220 (at contract creation) |
| Voting options (tallied) | |
| ZK proofs |
Transactions in the round
All transactions loaded.