“And, let’s be honest folks, the “alignment problem” isn’t remotely solved.”
Our reading, the sentence that was checked:
The AI alignment problem is not remotely solved.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Seven reasons why the world should say No to Sam Altman, paragraph 16.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-seven-reasons-why-the-world-should-2251b1-a1
We recorded 13 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2024-02-10. 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
- the “alignment problem”
- isn’t remotely solved
Our interpretation
- The ordinary reading is that reliably keeping advanced AI systems aligned with intended human goals remained an unresolved problem with substantial technical gaps.
- Remotely expresses the seriousness of the remaining gaps; it does not supply a measurable distance to a solution or imply that no progress existed.
- The surrounding discussion of scaling and catastrophic risks makes general, robust alignment the relevant scope, rather than success on a single preference benchmark.
- There is no reporting cadence or explicit period. I assess the current early-2024 research state from January 1 through the February 10 post; earlier work is context only.
Counts as supported if: Contemporaneous primary research demonstrates substantial unresolved failures in robust alignment methods, consistent with the absence of a general reliable solution.
Counts as contradicted if: Contemporaneous, independently validated evidence establishes a general alignment method that reliably controls advanced systems across relevant adversarial and unfamiliar conditions.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
Under the ordinary reading that general, robust alignment remained unresolved with substantial technical gaps, the claim is supported. January 2024 experiments showed that standard safety training could fail to remove deliberately implanted deceptive behavior. Earlier research documented progress, but also distinguished those results from solving general alignment. The evidence supports an unresolved technical problem, not a claim that no useful progress had occurred.
Limitations
- The decisive study was a preprint using deliberately constructed backdoors.
- It does not establish spontaneous deception in deployed systems or inevitable catastrophic harm.
- The evidence does not quantify how far researchers were from a solution.
Evidence (3 items)
Supports the claim: arXiv; Evan Hubinger and coauthors,
are ineffective
Why it matters here: The experiments show a substantial robustness gap: standard safety training did not reliably remove deliberately implanted deceptive behavior. This supports the ordinary reading that general alignment remained unresolved, without proving that deployed models spontaneously developed such behavior.
- Publisher
- arXiv; Evan Hubinger and coauthors
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
no known indefinitely scalable solution
Why it matters here: This earlier disclosure explains the general, scalable problem under discussion. Its date prevents it from independently settling the early-2024 observation.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: OpenAI,
significantly improve generalization
Why it matters here: OpenAI reported progress on a simplified supervision problem, while describing important limitations and differences from aligning superhuman systems. This cautions against reading Marcus as denying all progress, but does not establish a general solution.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Seven reasons why the world should say No to Sam Altman
And, let’s be honest folks, the “alignment problem” isn’t remotely solved.
- Where
- paragraph 16
- Published
- 2024-02-10
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-seven-reasons-why-the-world-should-2251b1-a1: Current. Outcome: Supported. 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-seven-reasons-why-the-world-should-2251b1-a1.