“Worse, we are no closer to a solution to the alignment problem now than we were then.”
Our reading, the sentence that was checked:
There has been no progress toward solving the AI alignment problem between 2022 and mid-2024.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, 26 months of ridicule and failure, paragraph 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-26-months-of-ridicule-and-failure-39298b-a1
We recorded 4 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2024-06-01. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- no closer to a solution
- the alignment problem
- now than we were then
- 26 months later
Our interpretation
- The comparison runs from approximately April 2022 to June 1, 2024, inferred from the post's 26-month comparison.
- The ordinary reading concerns progress toward reliably keeping AI behavior aligned with human intentions, including under unfamiliar conditions and increasing capability.
- Being unsolved does not establish that no progress occurred. Conversely, improving a particular safety benchmark does not by itself establish progress toward a general solution.
- Research addressing specific alignment obstacles is material counterevidence, but judging the overall distance to a solution requires a link between those advances and the broader problem.
Counts as supported if: Comparable evidence shows that the principal barriers to dependable alignment were no nearer resolution in June 2024 than around April 2022, accounting for relevant advances and their limitations.
Counts as contradicted if: Validated theoretical or empirical advances demonstrably remove or reduce a substantive barrier to a general alignment solution relative to the 2022 baseline.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Research during the comparison period reports advances in behavioral control and supervision, so a literal claim that alignment research made no progress faces substantial counterevidence. But Marcus says no closer to a solution to the alignment problem. The sources do not establish how those advances change the distance to a general solution. The international assessment explicitly separates improved control of some systems from uncertainty about future control. Neither continuing failures nor these limited advances settle the broader comparison.
Limitations
- A narrower interpretation covering any improvement in present-model alignment would favor contradiction.
- The original studies report results from their developers and do not establish dependable alignment across future systems.
- The assessment does not equate an unsolved problem with a lack of progress.
Evidence (3 items)
Context only: Anthropic researchers, via arXiv,
far fewer human labels
Why it matters here: The authors report progress in controlling assistant behavior while reducing human supervision. This challenges a literal claim of no alignment research progress, but does not demonstrate a general solution or establish its distance from completion.
- Publisher
- Anthropic researchers, via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI researchers, via arXiv,
empirical progress
Why it matters here: The study finds that weaker supervisors can elicit some stronger-model capabilities, addressing a relevant supervision obstacle. Its experiments are an analogy to future superhuman alignment, and the authors acknowledge that naive methods fail to recover full capabilities.
- Publisher
- OpenAI researchers, via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: International Scientific Report on the Safety of Advanced AI, published by the UK government,
more or less controllable
Why it matters here: The assessment notes improved control in some recent systems while finding the direction of future controllability uncertain. It supports separating demonstrated local advances from an established trajectory toward a general solution.
- Publisher
- International Scientific Report on the Safety of Advanced AI, published by the UK government
- Kind of source
- Secondary summary
- Describes events of
- Public from
What would settle it
A specified alignment objective and a comparable 2022–2024 assessment showing whether validated advances reduced its principal barriers, including evidence that those advances generalize beyond the tested settings.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, 26 months of ridicule and failure
Worse, we are no closer to a solution to the alignment problem now than we were then.
- Where
- paragraph 7
- Published
- 2024-06-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-26-months-of-ridicule-and-failure-39298b-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-26-months-of-ridicule-and-failure-39298b-a1.