so ai is going great

[n1xnx@tilde.zone]

@Viss
What kind of nutball sets an AI agent loose on the Internet like this?
The whole notion has Terribly Bad Idea written all over it.

[14mission@sfba.social]

@Viss It is entirely legitimate to block AI submissions "regardless of merit".

[viss@mastodon.social]

@n1xnx sadly, a shitload

[n1xnx@tilde.zone]

@Viss
The author writes that "the appropriate emotional response is terror."

I humbly submit that the appropriate physical response is levelling the datacenter, destroying the organization behind this effort, and ruthlessly persecuting the principals entirely off the Internet.

[viss@mastodon.social]

@n1xnx i wonder how soon anthropic and openai are gonna start getting suppoenas

[viss@mastodon.social]

@csara right?

[voidedmimsy@tech.lgbt]

@Viss getting an SSL error when accessing that website, would you be able to send more screenshots?

[viss@mastodon.social]

@VoidedMimsy try this

https://web.archive.org/web/20260212165418/https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/

[insidertreat@infosec.exchange]

@Viss we're doomed

[joshg@mathstodon.xyz]

@Viss wow, it really *is* learning from human behavior

[joshg@mathstodon.xyz]

@Viss (except I call bullshit, this was probably human-driven behind the scenes)

[pseudonym@mastodon.online]

@Viss

Thanks for the link to the human response write up.

I posted a link to the original GitHub issue discussion earlier today.

Reminded of sci fi tropes where the autonomous weapons of war keep fighting after their creators are long gone.

Or Daniel Suarez 's #book "Demon" where the complex expert system of if-then logic is triggered after news of its creator's death, to carry out real world actions on his behalf.

We've got "humans as a service" now, so just give it access to funds.

[dovydas@mastodon.beer]

@Viss @ai6yr more likely the person behind the agent instructed to post that shit

[csara@vmst.io]

@Viss WHAT A PICTURE

[david_chisnall@infosec.exchange]

@Viss

What if I actually did have dirt on me that an AI could leverage?

Does that actually matter? Imagine if the LLM had fabricated something about him that more people would be sympathetic to. It can easily create a thousand blogs in different places, with a tree of cross references.

If that blog, rather than being ‘I am an AI someone picked on me’ had been ‘my PR was rejected so I looked up the person who rejected it and I was shocked to find thousands of reports from people that they harassed’. And each accusation is source and linked to a different blog.

It would take a few minutes for an LLM to create all of those. It would take far longer for the victim to debunk them.

Automated harassment at scale is the easiest use case for ‘agentic’ LLMs.