Erik Hoel (@erikphoel) · AI Consciousness Opinions

Erik Hoel @erikphoel Jul 6

Yup. What they are doing is playing a game where you take [concept from consciousness research] and then [find an analog of it in LLMs], because for *most* theories of consciousness the concepts are defined loosely enough you can always make this work.

The concept is undefined

12 124 quoted by 1 archive account substance 6 · argument · rank 214 · Attributions of consciousness to LLMs exploit loosely defined concepts in consciousness research. view on X ↗

Erik Hoel @erikphoel Aug 9

Affirming

If you think alignment is hard for pre-trained models, wait until you meet continual learning models

1 more tweet in this thread

Some predictions: 1. Real continual learning will be quite hard. 2. It's probably linked to consciousness and free will (or whatever deflationary definition you want to put in here) and any model that continually learns will feel far more "alive." 3. There is no mathematical way

Dwarkesh Patel @dwarkesh_sp Aug 7

8 Predictions for the Era of Continual Learning. Also up on YouTube, pod feed, and Substack.

0 299 substance 5 · opinion · rank 75 · Real continual learning is likely linked to consciousness, making models feel far more alive. view on X ↗

1 top reply

Ben Schulz @schulzb589 Aug 10

Given my 9 year olds efforts around getting various school fads (toys, trinkets, games, styles), scheming is going to be a massive issue.

Tweets