3 ms·
It seems harsh to critique guardrails and take them into account in the scoring when GPT-5.5 seems to have been explicitly whitelisted to remove most of said gu
by mynameisvlad 4mo ago
It seems harsh to critique guardrails and take them into account in the scoring when GPT-5.5 seems to have been explicitly whitelisted to remove most of said guardrails. A more fair comparison would be a vanilla GPT account.
- jc4p 4mo agoI agree fully and hope someone else is able to do this test! For me it was a matter of cost and quotas that stopped me from changing to a new account. Also just to mention: Claude guardrails —> that session terminated. GPT guardrails -> your whole account is slowed down.
- tmikaeld 4mo agoDoes it matter when you can’t have the opus 4.8 guard rails removed? With GPT at least you can and they’re quick about it
- mynameisvlad 4mo agoI mean, yes. Most people aren’t security researchers, and either way it’s apples to oranges at that point if you’re counting “the guardrails stopped me” as a negative for one but not the other.
- sandos 4mo agoBut should developers be barred from asking an LLM to try secure their own app? Its not different from finding exploits...
- mynameisvlad 4mo agoThat is a completely separate question and discussion.