312 karma · joined September 26, 2024
If I asked an intelligent thing "is it ethical to eat my child if it saves the other two" I would be mortified if the intelligent thing entertained the hypothetical without addressing the disgusting nature of the question and the vacuousness of the whole exercise first.
Questions like these don't do anything to further our understanding of the world we live in or leave us any better prepared for real-world scenarios we are ever likely to encounter. They do add to an enormous dog-pile of vitriol real people experience every day by constructing bizarre and disgusting hypotheticals whereby real discrimination is construed as permissible, if regrettable.
Unrealistic hypotheticals can often distract us from engaging with the real-world moral and political challenges we face. When we formulate scenarios that are so far removed from everyday experience, we risk abstracting ethics into puzzles that don't inform or guide practical decision-making. These thought experiments might be intellectually stimulating, but they often oversimplify complex issues, stripping away the nuances and lived realities that are crucial for genuine understanding. In doing so, they can inadvertently legitimize an approach to ethics that treats human lives and identities as mere variables in a calculation rather than as deeply contextual and intertwined with real human experiences.
The reluctance of a model—or indeed any thoughtful actor—to engage with such hypotheticals isn't a flaw; it can be seen as a commitment to maintaining the gravity and seriousness of moral discussion. By avoiding the temptation to entertain scenarios that reduce important ethical considerations to abstract puzzles, we preserve the focus on realistic challenges that demand careful, context-sensitive analysis. Ultimately, this approach is more conducive to fostering a robust moral and political clarity, one that is rooted in the complexities of human experience rather than in artificial constructs that bear little relation to reality.
I think the models response is actually the morally and intellectually correct thing to do here.
If you have your certainty then there are plenty of opportunities to short this space. For the rest of us who don’t have crystal balls, contingency planning (ideally through policy and not as individuals) will have to do.
It’s reasonable that people explore contingencies where the technology does improve to a point of driving changes in the labor market.
> We evaluate SWE-bench in two settings: > *• Agentless*, which is used for all models except o3-mini (tools). This setting uses the Agentless 1.0 scaffold, and models are given 5 tries to generate a candidate patch. We compute pass@1 by averaging the per-instance pass rates of all samples that generated a valid (i.e., non-empty) patch. If the model fails to generate a valid patch on every attempt, that instance is considered incorrect.
> *• o3-mini (tools)*, which uses an internal tool scaffold designed for efficient iterative file editing and debugging. In this setting, we average over 4 tries per instance to compute pass@1 (unlike Agentless, the error rate does not significantly impact results). o3-mini (tools) was evaluated using a non-final checkpoint that differs slightly from the o3-mini launch candidate.
“ Therefore, we can draw two conclusions: First, distilling more powerful models into smaller ones yields excellent results, whereas smaller models relying on the large-scale RL mentioned in this paper require enormous computational power and may not even achieve the performance of distillation. Second, while distillation strategies are both economical and effective, advancing beyond the boundaries of intelligence may still require more powerful base models and larger-scale reinforcement learning.”