The inventor of the bitter lesson has a bitter lesson for LLM scaling enthusiasts
The Decoder
Turing Award winner Richard Sutton and other researchers warn that raw compute and transformer scaling alone are insufficient for true artificial intelligence.
Turing Award winner Richard Sutton has reaffirmed his "bitter lesson" theory, arguing that raw compute alone is insufficient to achieve true intelligence and that fundamentally new learning methods are required. Former OpenAI researcher Tworek supported this view, stating that blind transformer scaling has negatively influenced a generation of scientists. Meanwhile, a growing wave of startups is beginning to invest in alternative approaches to address this technological gap.
Why it matters: It highlights a growing shift in the artificial intelligence community away from reliance solely on raw compute and transformer scaling toward fundamentally new learning methods.

