AI progress has plateaued at GPT-4 level
Or, "Search isn't the same thing as intelligence"
Does this new reliance on search-based techniques obviate all concerns about AI safety and existential risk from these technologies? No, it doesn't.
However, the news that baseline general intelligence is plateauing, is an early indication that—at least barring further fundamental breakthroughs—timelines have just been pushed back in our lifetimes.
Erik Hoel
https://www.theintrinsicperspective.com/p/ai-progress-has-plateaued-at-gpt
@corlin @Zevon another article that dances all around embodiment, describes a bunch of failures that stem from the fact that intelligence doesn't come from a brain in a jar. We have bodies and they have senses.
Deprive a human of (only) hearing from birth and language acquisition is at risk.
I bet a whole annoyed psychology literature is developing around calling LLMs "intelligence".
Shorter article: there are tiny (30B) LLMs that perform close to as well as gpt4. Threshold reached.
oh yes! many branches of psychology (neuro, cognitive, et al.) have warned for decades about attempts to limit the def of "intelligence" as several in the AI field do. Now with their backs to the wall, they are forced to concede, but the AGI marketing machine and philosophical dishonesty rolls on @b4cks4w @corlin
it is and I respect your work and opinion. The more I study the technical aspects of AI the clearer it becomes most of what the general public and "influencers" actually know about AI, NLP, etc is almost nil leading to distorted impressions. GenAI and LLMs offer tremendous opportunities for innovation, but it won't drop out of the sky, there's a learning curve no matter how automated the application @b4cks4w @corlin