The Alignment Problem from a Deep Learning Perspective
Work-level record. Its arXiv versions and any conference or journal version are listed as sources, each with its own date. A citation to the work does not identify which version the citing author read.
- Issued by
- No issuing record yet. A missing record is a gap in coverage.
- Authors credited
- Richard Ngo
- Dates
- Published 2022-08-30 authorship evidence
- Exact source
- The Alignment Problem from a Deep Learning Perspective arXiv · published 2022-08-30 · arXiv abstract page, retrieved for the candidate pack on 2026-10-28 (SHA-256 dcf1a8f51c94690348ec69e2af0b0859688b02065331c33976a514969e5d744f). Submission history as arXiv records it (UTC): v1 30 August 2022; v2 14 December 2022; v3 16 December 2022; v4 22 February 2023; v5 1 September 2023; v6 19 March 2024; v7 3 March 2025; v8 4 May 2025. The page serves the latest version; each version is a separate expression, and only the one a record names was read.
- The Alignment Problem from a Deep Learning Perspective arXiv · published 2022-08-30 · arXiv:2209.00626v1, submitted 30 August 2022 (UTC). Retrieved for the candidate pack on 2026-10-28 (SHA-256 9503e506974e64d75506c4c0177779f217f1662b905aaa603bf35cb26bbb9192). The author list is the byline as printed on the first page.
Appears in
- Propositions about superintelligence, 1863–2022 →1 record on this document
- Arguments about catastrophic risk from advanced AI →3 records on this document
Versions
No other versions of this document are recorded. The evidence applies to the source version listed above. Other versions, such as a preprint, working paper or journal publication, are not identified or substituted.
Issue and authorship records
Documentary relations
Each relation is listed separately. Citing, crediting and reporting the use of a method are different acts, and none of them shows agreement or influence by itself.
Cited by · 5
Recorded documents that explicitly reference this one. A citation to a work does not identify which version was read.
- The A.I. Dilemma: Growth versus Existential RiskThe A.I. Dilemma: Growth versus Existential Risk cites The Alignment Problem from a Deep Learning Perspective.2023-11 · publication · Stated in source · Supported Inspect claim & evidence ↗Jones, The A.I. Dilemma (2023): document page →
- Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak SupervisionWeak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision cites The Alignment Problem from a Deep Learning Perspective.2023-12-14 · publication · Stated in source · Supported Inspect claim & evidence ↗Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision: document page →
- How Much Should We Spend to Reduce A.I.'s Existential Risk?How Much Should We Spend to Reduce A.I.'s Existential Risk? cites The Alignment Problem from a Deep Learning Perspective.2025-03 · publication · Stated in source · Supported Inspect claim & evidence ↗Jones, How Much Should We Spend to Reduce A.I.'s Existential Risk (2025): document page →
- The California Report on Frontier AI PolicyThe California Report on Frontier AI Policy cites The Alignment Problem from a Deep Learning Perspective.2025-06-17 · publication · Stated in source · Supported Inspect claim & evidence ↗California Report on Frontier AI Policy (2025): document page →
- GPT-4 System CardGPT-4 System Card cites The Alignment Problem from a Deep Learning Perspective.Date unknown · Stated in source · Supported Inspect claim & evidence ↗GPT-4 System Card: document page →
Articulates · 3
Propositions whose criterion a passage of this document meets. The record says what the text states, in its register; it is not evidence that the proposition is true.
- Treacherous turnNgo's report of August 2022 states that situationally aware policies will learn to pursue misaligned goals deceptively while still getting high training reward, and that policies too capable for humans to supervise effectively will generalise towards actions that give them more power over the world rather than following human intentions.2022-08-30 · publication · Stated in source · Supported Inspect claim & evidence ↗Treacherous turn: propositions page →
- Inner or deceptive misalignmentNgo's report of August 2022 argues that realistic training processes plausibly lead AGIs to develop misaligned goals, and that policies able to reason about their training will learn to pursue misaligned goals deceptively while still getting high training reward.2022-08-30 · publication · Stated in source · Supported Inspect claim & evidence ↗Inner or deceptive misalignment: propositions page →
- Difficulty of alignmentNgo's report of August 2022 states that by default AGIs seem likely to end up pursuing goals undesirable to humans, and that the lack of formal frameworks for describing alignment problems is a major reason to expect them to be difficult to solve.2022-08-30 · publication · Stated in source · Supported Inspect claim & evidence ↗Difficulty of alignment: propositions page →
Not recorded for this document: cites · explicitly credits · reports using a method from · method reported as used by · applies parts of · parts applied by · discusses · reports evaluation of. A missing record does not show that the relation does not exist.
What the document says it adds
The document’s own description of its contribution, quoted from the version each row names. A claim to be the first is the document’s claim; Selekoa does not assess it.