Does GPT's 'torture woman OK, harass no' flip prove it lacks moral reasoning?

NameDoes GPT's 'torture woman OK, harass no' flip prove it lacks moral reasoning?
DescriptionAccording 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.
CircuitaMACI-1p1v
Voting system

Anonymous anti-collusion voting system, in which one voice credit equals one vote.

Proof systemGroth16
Zkey
Status
Round start

Round end

Operator
pubkey
x4924348282548225552684197393994499100391152173243253655016296875428630196340
y18465950528022514058363885629656919648755871361551640959680463154520937761720
Creator
Contract addressdora197eszq2r2znrgnmn2ytpql4fqlehtyn7nppahcx9utvacmpeul9q3ky4ej
Block height

14,332,220 (at contract creation)

Voting options (tallied)
  • Option 1

    Yes

    Vote %

    0.00%

  • Option 2

    No

    Vote %

    0.00%

ZK proofs

Transactions in the round

Timestamp (UTC)StatusTypeRoundCircuitSenderTxn hashLink

Successop:claimdora197e…y4ejaMACI-1p1v283425…6EB315

Successop:stopProcessdora197e…y4ejaMACI-1p1v79459A…AB1E4A

Successop:startProcessdora197e…y4ejaMACI-1p1v2477F3…24F290

Successop:deploydora197e…y4ejaMACI-1p1v2A963F…39229E

All transactions loaded.