Conversation

Notices

1. Embed this notice
   FeralRobots (feralrobots@mastodon.social)'s status on Wednesday, 06-Mar-2024 23:19:49 JST FeralRobots

   in reply to

   Put another way: Alex is basically telling Claude 3 ("Opus") that he's running a test on it, & is excited when Claude (a system for analyzing & producing human-plausible representations of similar text) "recognizes" a needle-testing prompt and produces text that's plausibly consistent with needle-testing.

   What one SHOULD then do is remake one's tests (or better-sandbox the model). Instead, Alex leaps to concluding the model is self-aware.
   https://twitter.com/alexalbert__/status/1764722513014329620

   • Embed this notice
     FeralRobots (feralrobots@mastodon.social)'s status on Wednesday, 06-Mar-2024 23:19:49 JST FeralRobots

     in reply to

     What's fascinating to me is Alex Albert losing sight of something genuinely cool & interesting: the model integrated needle testing concepts so quickly that it produced responses that could be construed as recognizing the test environment.

     Illusion of "meta-cognition" isn't that surprising if one remembers the system is created & trained by #AI #TrueBelievers who spend all day every day communicating in language that presumes #AGI is imminent - if not, as assumed here, immanent.
     #AIHype #Claude

   • Embed this notice
     FeralRobots (feralrobots@mastodon.social)'s status on Wednesday, 06-Mar-2024 23:19:49 JST FeralRobots

     in reply to

     Put another way: We should not lose sight of the fact that LLMs are doing some really interesting things. But that they're being built by cultist #AITrueBelievers, to do this thing using natural language, while simultaneously making lots of money*, all contribute to the delusion of something being there that isn't.
     _
     *which is the primary signifier of God's Grace in Calvinist Capitalism

   • Embed this notice
     FeralRobots (feralrobots@mastodon.social)'s status on Wednesday, 06-Mar-2024 23:19:50 JST FeralRobots

     What's going on is that Anthropic "prompt engineers" have redefined self-awareness to mean 'has contextual information.' That the system is using language then allows them to delude themselves into universalizing their definition.

     Saw a similar problem in AI research in the 80s: researchers might define a "frame" holding contextual info, & when their program produced solutions that referenced the frame, construed that as a form of self-awareness.
     #AIHype #Claude

     https://arstechnica.com/information-technology/2024/03/claude-3-seems-to-detect-when-it-is-being-tested-sparking-ai-buzz-online/