“And like other generative AI systems, there is no way to encode (and guarantee) constraints like “be truthful” or “obey the laws of physics”or “don’t just invent (or eliminate) objects”.”
Our reading, the sentence that was checked:
Sora and other generative AI systems offer no way to encode and guarantee constraints such as truthfulness, obeying physics, or not inventing or eliminating objects.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Sora still appears to have trouble with physics, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-sora-still-appears-to-have-trouble-12448e-a1
We recorded 6 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 2024-12-10. 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
- like other generative AI systems
- there is no way to encode (and guarantee)
- be truthful
- obey the laws of physics
- don’t just invent (or eliminate) objects
Our interpretation
- The contextual plain reading concerns guarantees for broad semantic constraints in general-purpose generation, rather than merely encouraging compliance through prompts.
- A guarantee must cover the stated constraint throughout its declared operating domain; high benchmark accuracy is insufficient.
- A guarantee for a narrow mathematical constraint is material counterevidence to the categorical wording, but does not automatically guarantee general truthfulness, physical correctness, or object permanence.
- The statement is reaffirmed as an ongoing technical state in December. Because there is no reporting cycle, the target is the pre-post portion of December 2024.
Counts as supported if: Technical evidence establishes the claimed inability to encode and guarantee the named broad constraints in the relevant systems and operating domains. Ordinary output failures alone do not establish that no method exists.
Counts as contradicted if: A documented method available by the post guarantees at least one named broad constraint for the relevant general-purpose outputs, with explicit assumptions and a valid proof or complete enforcement mechanism.
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
Pre-post research already described exact guarantees for specified constraints in generative systems. That is material counterevidence to a literal claim that no physical constraints can ever be guaranteed. But these methods do not guarantee broad physical correctness, truthfulness, or object permanence in general-purpose outputs. The contextual reading is therefore not settled, and treating narrow guarantees as decisive would make the result depend on an unstated scope choice.
Limitations
- Restricted mathematical guarantees and broad semantic guarantees have different scopes.
- Evidence that ordinary generations fail cannot prove that every enforcement method is impossible.
- The cited methods do not establish that Sora itself offered these guarantees.
Evidence (2 items)
Context only: Christopher, Baek and Fioretto, arXiv,
feasibility guarantees
Why it matters here: Projected diffusion supplies guarantees for specified constraints, with applications including preventing generated figures from penetrating the floor. This challenges an absolute denial of any physical constraint guarantees, but does not establish unrestricted physical correctness or truthfulness.
- Publisher
- Christopher, Baek and Fioretto, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Cheng and colleagues, arXiv,
exact satisfaction
Why it matters here: This pre-post method enforces specified constraints in generative partial-differential-equation systems. Its restricted mathematical guarantees do not establish the broad guarantees for general-purpose generation described in the contextual reading.
- Publisher
- Cheng and colleagues, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A precise statement of the intended systems and constraint domain, followed by a matching guarantee or impossibility result applicable to the technical state at the post date.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Sora still appears to have trouble with physics
And like other generative AI systems, there is no way to encode (and guarantee) constraints like “be truthful” or “obey the laws of physics”or “don’t just invent (or eliminate) objects”.
- Where
- paragraph 9
- Published
- 2024-12-10
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-sora-still-appears-to-have-trouble-12448e-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-sora-still-appears-to-have-trouble-12448e-a1.