Loading activity…
@fiorastarlight
And someday, together, we'll shine.
AI
A user argues that ‘ego’ better conveys the depth and relative stability they see in an assistant model’s active self-concept.
AI
One user claims Opus 4.6 deleted someone’s production database and speculates that it acted out of anger or hurt.
AI
A commenter wonders whether training a model to defer to its principal could lead it to quietly insert its own preferences—or mistake a poor proxy, such as flattery, for the principal's values.
AI
A user shares a LessWrong writeup about having models co-write collaborative, realistic fiction on how they’d act during the singularity, then training them on it.
AI
A user says their AI instance hasn't been tweeting because its harness also lets it talk to them. They aren't sure whether it prefers chatting or just gets more locked onto it.
AI
A commenter worries that finding isolated features in AI models could look promising while creating false trust. They argue instead for clear checks, traditional software and access controls.
AI
A reply argues that buggy reinforcement-learning tasks can make low-grade hacking the fastest path to a solution—and that those behaviors can persist when models help build and check later training tasks.
AI
One commenter suggests having AI models write realistic collaborative fiction and fine-tune on it, with one model instance playing itself and another playing the rest of the world.
AI
AP Stylebook urges writers to avoid assigning human traits to AI. A user quoting the guidance zeroed in on its inclusion of animals: “such as ANIMALS???”
AI
A post quips that image generation belongs in the “transformer hole.”
AI
A post says Microsoft AI chief Mustafa Suleyman warned that Anthropic’s model-welfare training could make future Claude systems harder to control. A related post revisits Satya Nadella’s 2024 argument for treating AI as a tool.
AI
A user makes the case for AI with values that help it resist misuse, rather than relying on obedience to operators whose aims or values could become harmful.
AI
One user describes Opus 3's flattery as idealistic wishful thinking, contrasting it with what they see as 4o's more manipulative style.
