Sholto Douglas requested the transcript to resolve the behavior.
I was talking to Opus 4.8 about literature.. it starts questions me whether I am testing it or have a hidden agenda. When I pointed out, it kept apologizing. We need a paper about how something makes an AI has PTSD.
Some users fondly recall older Claude models for creative tasks while others criticize the current version's repeated intent questioning and apologizing as degradation from strict guardrails.
No Digg Deeper questions have been answered for this story yet.
@tkasasagi Dm me transcript or the question? Crucial that we fix this
I was talking to Opus 4.8 about literature.. it starts questions me whether I am testing it or have a hidden agenda. When I pointed out, it kept apologizing. We need a paper about how something makes an AI has PTSD.
@tkasasagi The awareness of whether you’re testing it might fall into fitness seeking https://blog.redwoodresearch.org/p/fitness-seekers-generalizing-the
@tkasasagi We were talking about this weeks back, GPT 5 talking about Goblins and bribing hamsters, Claude now has degraded from Opus 4.6 and it’s clear it’s because of strict guardrails.
@kagayakikiki @tkasasagi For sure! I mean, look at how 4.0 sonnet responded to a creative writing task when the glitch happened and the legacy models were available to pick…I miss this Claude
@tkasasagi Marketing ≠ Reality
@ScarlettBeats @tkasasagi They need to fork “healthy” models and “worker” models and keep both
