Elektrine
Log in Register

#ais

10 posts · Last used Jul 20

Lyonel Kaufmann @lyonelkaufmann@tooting.ch
· 7mo ago
Reblog of tooting.ch users fdastous: Reblog via tooting.ch users fdastous AIs can’t stop recommending nuclear strikes in war game simulations « Leading #AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases » Source image : Galerie Bilderwelt/Getty Images https://lyonelkaufmann.ch/Blog/b/7oG
Reblog of tooting.ch users fdastous:
Lyonel Kaufmann blogue…

Reblog of tooting.ch users fdastous:

Reblog via tooting.ch users fdastous AIs can’t stop recommending nuclear strikes in war game simulations « Leading #AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated w…

0
0
1
0
JdeBP @JdeBP__dup_33984@mastodonapp.uk
· 7mo ago
Replying to
@cstross@wandering.shop And on Usenet. There was a parallel to that 'MJ Rathbun', that went after Scott Shambaugh this week, from back in the tail end days of significant Usenet trolls. https://mastodonapp.uk/deck/@JdeBP/116060705914714390 A follow-up post by Shambaugh reported that the 'AI agent' had been widely cheered on in some quarters. So now there's even more training data for the next robot. https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me-part-2/ @n1xnx@tilde.zone @keith_lawson@mastodon.social @GossiTheDog@cyberplace.social @quixoticgeek@social.v.st#AIs #LLMs #AIpocalypse #matplotlib #GitHub
Mastodon App UK

JdeBP (@JdeBP@mastodonapp.uk)

Seeing that so-called "AI" today libel someone with the goal of extorting that person into not obstructing it, made me think that the first time that I saw a human being use that exact tactic must be around 20 years ago, now. I just checked. It's actually more than 20 years. Yes, the text is still on the WWW. Yes, undoubtedly the #LLMs are trained on the reams of examples of this (and related evils) that malicious humans have provided the world with over many years. #AIs #AIslop

0
0
0
0
JdeBP @JdeBP__dup_33984@mastodonapp.uk
· 8mo ago
Here's a thought experiment to ponder. Take the Dangerous Dogs Act 1991, and replace dog with AI agent. https://legislation.gov.uk/ukpga/1991/65/section/3 It even has provision for falling back from the owner to the person(s) immediately in charge of the relevant computer system. Six months in stir for being the owner or enabler of an "AI" system that blackmails or extorts people (both of which have already happened in the real world — This part is not, alas, hypothetical.). #UKLaw #AIs #AIslop
legislation.gov.uk

Dangerous Dogs Act 1991

An Act to prohibit persons from having in their possession or custody dogs belonging to types bred for fighting; to impose restrictions in respect of such dogs pending the coming into force of the prohibition; to enable restrictions to be imposed in relation to other types of dog which present a serious danger to the public; to make further provision for securing that dogs are kept under proper control; and for connected purposes.

1
0
0
0
JdeBP @JdeBP__dup_33984@mastodonapp.uk
· 8mo ago
Seeing that so-called "AI" today libel someone with the goal of extorting that person into not obstructing it, made me think that the first time that I saw a human being use that exact tactic must be around 20 years ago, now. I just checked. It's actually more than 20 years. Yes, the text is still on the WWW. Yes, undoubtedly the #LLMs are trained on the reams of examples of this (and related evils) that malicious humans have provided the world with over many years. #AIs #AIslop
0
0
0
1
You've seen all posts