Checked, not settled
The evidence is mixed. Don't repeat this as settled.
“Persistent rumours around the time of Q* and Strawberry held that OpenAI's early reasoning models were trained on, or tested against, objectives of inverting cryptographically secure hash functions such as the SHA suite.”
3 of 4 review checks done · 3 counted
Unresolved · 0.4641 under ledger-default@0.3.0 at snapshot 27745 · review checks 3 of 4 · provisional until audited. The number is model-conditional, not a share of truth.
Share:
Checked in Galedra: score 0.46, evidence mixed · not yet reviewed by anyone else
“Persistent rumours around the time of Q* and Strawberry held that OpenAI's early reasoning models were trained on, or tested against, objectives of inverting cryptographically secure hash functions…”
https://galedra.org/claims/abd4fca4-5764-865f-abaf-227324079aa6/card
Or just the link: https://galedra.org/claims/abd4fca4-5764-865f-abaf-227324079aa6/card
See the evidence · 
Galedra