purple_cloud@[email protected] Superintelligent AI capable of going rogue and leading to humanity’s extinction within this decade ? How ? Because OpenAI developed a LLM which is exceptionally good at predicting the next token in a sequence. Does somebody care to elaborate ?
Goutte@purple_cloud @[email protected]
Not sure if he's on the fedi yet, but Eliezer Yudkowski speaks quite convincingly about this.
The gist of it is : we don't get to survive a misaligned AI (no trial-and-error), we never faced a threat more intelligent than us, the emergent capabilities of AIs when we add more layers surprise everyone, optimism bias, and intelligence tends to explode (monkeys on the moon in the blink of an eye).