Ancestors

Written by James Tindall on 2025-01-09 at 09:20

A couple of things worry me about this quote from @simon in the WP article --though I suspect it got detached from a more nuanced reply.

Firstly, gullibility is a human trait that requires belief. Vulnerability might have been more appropriate. The ways next-token prediction machines can perform damaging actions are system vulnerabilities, that play on quite different human vulnerabilities baked into our entirely distinct and ancient operating system.

[Not linking to WP 🙄 ]

=> View attached media

=> More informations about this toot | More toots from atomless@mastodon.social

Toot

Written by James Tindall on 2025-01-09 at 09:36

The problem with next-token prediction machines, especially in agentic form, is not even the gullibility of the users embracing them.

Rather, that these machines exploit the attack surface of human vulnerability by design, and this is fundamental to their operation.

As conjoined twins, they feed from the same libidinal root as the Tyranny of the Recommendation Algorithm. Removing this aspect of their operation would be akin to a recommendation algorithm configured for minimum engagement.

=> More informations about this toot | More toots from atomless@mastodon.social

Descendants

Written by Justin Mason on 2025-01-09 at 10:47

@atomless yes! absolutely spot on

=> More informations about this toot | More toots from jmason@mastodon.ie