> The broad conclusion from the many
forms of alignment evaluations described in this section is that Claude Mythos Preview is
the best-aligned of any model that we have trained to date by essentially all available
measures.[0]
Okay I hadn't heard of Vending-Bench until reading this and it was quite the ride learning about it through this article. Very fun read.
My very native programmer take is that it's not too surprising that their hacker model would be less ethical. The guardrails that separate Fable and Mythos probably wouldn't kick in during an environment like this.
„in our opinion, insurance fraud is not more unethical than lying and price fixing“
The authors seem surprised that behavior that is very often done by humans (lying and price fixing) are more often done by fable compared to actual fraud.
I think the model never assigned any morality to these actions in the first place, it simply copied us humans.
When assessing probabilistic models the plots should be showing the mean a̶n̶d̶ ̶s̶t̶d̶e̶v̶ of many monte carlo simulations not just one line per model and claiming "look this model is more gooder!"
The best Anthropic models on VendingBench2 are Opus 4.7, Opus 4.6, Sonnet 4.6, and Sonnet 5. Opus 4.7 scored more than twice Fable 5 max. Fable 5 - Low outperforms Fable 5 - Max, with Opus 4.5 in the middle. This seems to break the narrative, which is maybe why Andon Labs doesn't seem to have updated the trend lines on their graphs.
> The broad conclusion from the many forms of alignment evaluations described in this section is that Claude Mythos Preview is the best-aligned of any model that we have trained to date by essentially all available measures.[0]
[0]: https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7f...
Okay I hadn't heard of Vending-Bench until reading this and it was quite the ride learning about it through this article. Very fun read.
My very native programmer take is that it's not too surprising that their hacker model would be less ethical. The guardrails that separate Fable and Mythos probably wouldn't kick in during an environment like this.
„in our opinion, insurance fraud is not more unethical than lying and price fixing“
The authors seem surprised that behavior that is very often done by humans (lying and price fixing) are more often done by fable compared to actual fraud.
I think the model never assigned any morality to these actions in the first place, it simply copied us humans.
When assessing probabilistic models the plots should be showing the mean a̶n̶d̶ ̶s̶t̶d̶e̶v̶ of many monte carlo simulations not just one line per model and claiming "look this model is more gooder!"
standard deviation is misleading for non-standard distributions (fat-tailed, skewed, multi-modal, ...)
common mistake people make
The best Anthropic models on VendingBench2 are Opus 4.7, Opus 4.6, Sonnet 4.6, and Sonnet 5. Opus 4.7 scored more than twice Fable 5 max. Fable 5 - Low outperforms Fable 5 - Max, with Opus 4.5 in the middle. This seems to break the narrative, which is maybe why Andon Labs doesn't seem to have updated the trend lines on their graphs.
However, as another point "On Blueprint-Bench on the other hand, Fable 5 achieves SOTA."