Verified65% confidenceOpinionExact time
In an article titled "The Bitter Lesson" published in 2019, Sutton argued that almost all AI methods relying on human domain knowledge have ultimately been surpassed by general methods that scale with compute.
3
Sources
65%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedRich Sutton argued in 'The Bitter Lesson' that approaches which hard-code human knowledge into systems will ultimately always be surpassed by general methods that fully leverage computational scale.77% similarUnverifiedThe core argument of The Bitter Lesson is that methods relying on human knowledge, domain priors, and hand-crafted rules perform well in the short term, but in the long run are always surpassed by general methods that rely solely on compute and data.73% similarUnverifiedIntrinsic misalignment and extrinsic misalignment are two core concepts in AI alignment research, first systematically articulated by Hubinger et al. in their 2019 paper "Risks from Learned Optimization."71% similarUnverified判断力和方向感因高度依赖隐性知识与价值判断而难以被自动化，成为AI时代新的稀缺品68% similarUnverifiedThe author proposes four corollaries relevant to AI alignment: (1) in value learning and value loading, all values with real stakes are rules and contracts, and there is currently a lack of a "translatability theorem for rules"; (2) when a powerful optimizer's ontology diverges from humans', "at which level of the ladder" is the first diagnostic question; (3) children learn vocabulary through very little exposure because new words arrive with pre-types, directly cutting the hypothesis space by several orders of magnitude; (4) "I", "here" and "now" are the agent's coordinates in its world model, and all value references that must be delivered to the system carry indexicals that must be bound within the extension of what is bound.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/528091API
curl https://kongchang.com/api/v1/knowledge/claims/528091MCP
get_claim(id=528091)