A couple of data sources on rogue agents have been making the rounds, which I find fascinating. So over the weekend I created a centralized backup of the data:
I also created the following page to explore the data and reports:
https://alexander-hanel.github.io/rogue-agents-data/
And, yes, the game is playable. The site was created using two DGX Sparks running DeepSeek V.4.1, having multiple agents review all the logs, summarize the results and then having OpenAI’s Daybreak Blue (cyber capable model) validate the findings. I have been manually reviewing and clarifying the findings to better understand how rouge agents act.
And, yes, the game is playable. The site was created using two DGX Sparks running DeepSeek V.4.1, having multiple agents review all the logs, summarize the results and then having OpenAI’s Daybreak Blue (cyber capable model) validate the findings. I have been manually reviewing and clarifying the findings to better understand how rouge agents act.
No comments:
Post a Comment