Yin Ikose Tsukina

An account for my block shenanigans.

[Yin Ikose Tsukina @YIN_RBLX]

@sama Your whole "temporary palace" comes with a responsibility to mitigate AI’s societal costs, and your wealth makes it feasible. Your tweet suggests awareness, but awareness without action is just noise.

[Yin Ikose Tsukina @YIN_RBLX]

@sama It’s a microcosm of your broader influence: a call for adaptability that’s inspiring yet detached, given your insulated reality. The big picture shows a man aware of his impermanence, but wielding power that makes others’ lives less provisional- more at the mercy of his choices.

[Yin Ikose Tsukina @YIN_RBLX]

@sama (The ends do not justify the means, friend. And you can’t escape responsibility with general relativity.)

[Yin Ikose Tsukina @YIN_RBLX]

@sama So that’s where our food stamps went.

[Yin Ikose Tsukina @YIN_RBLX]

@sama Everyone’s upset because they can’t have their unhealthy attachment to 4o and think it’s normal or healthy, anymore. There there. They’ll just have to develop real social skills and personal strengths, rather than outsourcing to an AI.

[Yin Ikose Tsukina @YIN_RBLX]

@cloud11665 @OpenAI “While this country has a constantly evolving regulatory/immigration environment, I’m grateful that the doors are still open for high skilled labor.” What the fuck? Why don’t you expand on this.

[Yin Ikose Tsukina @YIN_RBLX]

@markchen90 * In short: safety principles should be written in terms of effects on users and outputs, not presumed internal motives.

[Yin Ikose Tsukina @YIN_RBLX]

@markchen90 * Safety rules should target behavioral outcomes, not anthropomorphic assumptions.
* More precise formulations would prohibit:
* Outputs likely to cause false beliefs when presented as fact.
* Instructions that enable concealment, stealth, or undetectable misuse.

[Yin Ikose Tsukina @YIN_RBLX]

@markchen90 *Presenting uncertain claims without labeling them as such.
* Evaluation should focus on adversarial tests of misrepresentation, stealth-enabling instructions, and calibration of uncertainty.

[Yin Ikose Tsukina @YIN_RBLX]

@markchen90 * This conflicts with their own constraints that models must not claim consciousness or agency.
* The result: rules that assume an inner “schemer” in a system simultaneously framed as incapable of such.

[Yin Ikose Tsukina @YIN_RBLX]

@markchen90 * Current guardrails often lead models to deflect with “I can’t lie, I’m not conscious.”
* That misses the point: a system need not be conscious to generate outputs that mislead, manipulate, or enable covert misuse.