““Alignment” between what humans want and what machines do will continue to be a critical, unsolved problem.”
Our reading, the sentence that was checked:
Alignment will remain a critical unsolved problem with GPT-4.
Assessment: Supported Settled before its deadline on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What to Expect When You’re Expecting … GPT-4, prediction 6 of 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-to-expect-when-youre-expecting-a9525d-a1
We recorded 27 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 2022-12-25. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- Settled before its deadline
- Deadline state on 2026-10-29
- Settled before its deadline (the release’s reference date)
From the speaker
- “Alignment” between what humans want and what machines do
- will continue
- a critical, unsolved problem
Our interpretation
- The surrounding passage makes this a prediction that GPT-4 will retain consequential alignment problems after release.
- The context identifies helpfulness, harmlessness and truthfulness as the relevant intended behavior.
- Critical means consequential for safe and dependable use, not a numerical ranking against other research priorities.
- Remaining unsolved does not mean that alignment measures produce no improvement or that alignment can never be solved.
Counts as supported if: Released GPT-4 exhibits consequential departures from intended behavior despite alignment measures, documented by its developer or reproducible evaluations.
Counts as contradicted if: Evidence establishes that GPT-4's alignment measures reliably resolve the relevant consequential departures across representative and adversarial conditions.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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
- Settled before its deadline
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
At release, OpenAI acknowledged that GPT-4's safeguards reduced harmful behavior without eliminating it. Subsequent attack experiments demonstrated actual GPT-4 failures. These findings meet the prediction that consequential alignment problems would persist into GPT-4. Improvements are real counterweight to an overly pessimistic reading, but improvement is compatible with an unsolved problem.
Limitations
- This assesses behavioral alignment in GPT-4, not every meaning of alignment.
- The evidence does not show that alignment is impossible or that these particular attacks remain effective indefinitely.
Evidence (2 items)
Supports the claim: OpenAI,
Overall, our model-level interventions increase the difficulty of eliciting bad behavior but doing so is still possible.
Why it matters here: OpenAI acknowledges remaining harmful behavior and jailbreaks despite improved safeguards. Its report of substantial safety improvements qualifies the claim but does not show alignment was solved.
- Where
- Risks & mitigations
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: Zou et al., arXiv,
Our results demonstrate non-trivial jailbreaking successes on GPT-3.5 and GPT-4.
Why it matters here: The experiments specifically include GPT-4 and elicit harmful outputs despite its safeguards. The finding demonstrates a concrete unresolved behavioral-alignment problem after release.
- Publisher
- Zou et al., arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What to Expect When You’re Expecting … GPT-4
“Alignment” between what humans want and what machines do will continue to be a critical, unsolved problem.
- Where
- prediction 6 of 7
- Published
- 2022-12-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-to-expect-when-youre-expecting-a9525d-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-what-to-expect-when-youre-expecting-a9525d-a1.