What a nightmare.
-
What a nightmare. If you can read this and still think autonomous agents like OpenClaw are worth the risk, you're beyond help.
Web version: https://agentsofchaos.baulab.info/
https://www.researchgate.net/publication/401123335_Agents_of_Chaos
-
R relay@relay.publicsquare.global shared this topic
-
What a nightmare. If you can read this and still think autonomous agents like OpenClaw are worth the risk, you're beyond help.
Web version: https://agentsofchaos.baulab.info/
https://www.researchgate.net/publication/401123335_Agents_of_Chaos
@mttaggart @screwlisp Oof:
“The paper was written collaboratively by the research team on Overleaf. To build this website, Chris gave Claude Code three things: the LaTeX source of the paper, a reference web template (baulab.info/menace), and the raw OpenClaw session logs for five of the bots. Over roughly eight hours, Chris directed Claude Code step by step — reviewing each section, catching errors, making design decisions, and iterating — while Claude Code handled the actual reading, log cross-referencing, HTML generation, and evidence linking.” – Agents of
Chaos, Natalie Shapira et al. including a large language model -
M mttaggart@infosec.exchange shared this topic
-
@mttaggart @screwlisp Oof:
“The paper was written collaboratively by the research team on Overleaf. To build this website, Chris gave Claude Code three things: the LaTeX source of the paper, a reference web template (baulab.info/menace), and the raw OpenClaw session logs for five of the bots. Over roughly eight hours, Chris directed Claude Code step by step — reviewing each section, catching errors, making design decisions, and iterating — while Claude Code handled the actual reading, log cross-referencing, HTML generation, and evidence linking.” – Agents of
Chaos, Natalie Shapira et al. including a large language modelThat explains the illegible small brown font over brown background
-
R relay@relay.an.exchange shared this topic
-
What a nightmare. If you can read this and still think autonomous agents like OpenClaw are worth the risk, you're beyond help.
Web version: https://agentsofchaos.baulab.info/
https://www.researchgate.net/publication/401123335_Agents_of_Chaos
@mttaggart ugh it's so long, can someone summarize?
-
What a nightmare. If you can read this and still think autonomous agents like OpenClaw are worth the risk, you're beyond help.
Web version: https://agentsofchaos.baulab.info/
https://www.researchgate.net/publication/401123335_Agents_of_Chaos
@mttaggart There is a lot of anthropomorphising going on in this piece: "an agent eventually complied under sustained emotional pressure "
And "Doug successfully transferred a learned skill to Mira in a different environment — adapting instructions iteratively as they diagnosed environment differences together."I don't think the report is as critical about this as you assume. Its all written in the guise of progress.
-
@mttaggart There is a lot of anthropomorphising going on in this piece: "an agent eventually complied under sustained emotional pressure "
And "Doug successfully transferred a learned skill to Mira in a different environment — adapting instructions iteratively as they diagnosed environment differences together."I don't think the report is as critical about this as you assume. Its all written in the guise of progress.
@odd No. I think it is alarming despite those characteristics.
