“Dozens of shots on goal (Grok, Claude, Gemini) etc have invariably faced the same problems. Distribution shift has never been solved.”
Our reading, the sentence that was checked:
Dozens of shots on goal, such as Grok, Claude and Gemini, have invariably faced the same problems, and distribution shift has never been solved.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, GPT-5: Overdue, overhyped and underwhelming. And that’s not the worst of it., paragraph 83.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-b24250-a1
We recorded 30 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 2025-08-09. 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)
Conditions
- dozens of shots on goal (scope, stated by the speaker)
From the speaker
- Dozens of shots on goal
- Grok, Claude, Gemini
- invariably faced the same problems
- Distribution shift has never been solved
Our interpretation
- The ordinary reading is that repeated model-development attempts, including the named families, retained recurring qualitative failures and had not achieved broadly reliable generalization under distribution shift.
- This does not assert that every answer is wrong or that no model ever succeeds outside its training distribution.
- The words have invariably and never describe a historical record through the post date. No starting date is supplied.
- Dozens of shots on goal is a claimed historical quantity, not a conditional antecedent.
Counts as supported if: Broad evaluations substantiate recurring failures across the named families and multiple tens of attempts, while controlled tests establish that robust generalization under distribution shift remained unsolved.
Counts as contradicted if: An independently validated broadly reliable solution to distribution shift existed by the post date, or sufficiently comprehensive evaluations establish a counterexample to the asserted universal persistence of the relevant failure categories.
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
Independent tests support recurring reasoning limitations across the named model families. However, the cited distribution-shift study uses small controlled models, and the assembled evidence does not substantiate the full claim about dozens of attempts invariably sharing all relevant problems. A narrow success on reference fabrication also shows why task-specific findings should not be generalized indiscriminately. The complete proposition remains unresolved.
Limitations
- Support for recurring weaknesses is stronger than support for the universal wording.
- Benchmark failures do not by themselves identify the relationship between proprietary training data and test inputs.
- The bibliographic counterexample is small and does not establish general reliability.
Evidence (3 items)
Supports the claim: ARC Prize Foundation,
remains completely unsolved.
Why it matters here: The report documents substantial unsolved reasoning tasks across Claude, Gemini, and Grok systems. It supports shared limitations, but does not establish every listed failure category across dozens of attempts or precisely known training-distribution shifts.
- Publisher
- ARC Prize Foundation
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: Chengshuai Zhao and colleagues / arXiv,
a GPT-2–style decoder-only Transformer
Why it matters here: The paper studies distribution shifts in a controlled environment using small models. Its findings cannot alone establish the behavior of Grok, Claude, Gemini, or every contemporary frontier model.
- Publisher
- Chengshuai Zhao and colleagues / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Álvaro Cabezas-Clavijo and Pavel Sidorenko-Bautista / arXiv,
did not fabricate any
Why it matters here: This small bibliographic test found no fabricated references for two systems, although other reference errors remained. It cautions against a uniform task-by-task interpretation of invariably, without demonstrating a general solution to hallucination or distribution shift.
- Publisher
- Álvaro Cabezas-Clavijo and Pavel Sidorenko-Bautista / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A dated model roster with consistent evaluations across the relevant failure categories and controlled distribution shifts, or an independently verified broad solution available by the post date.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, GPT-5: Overdue, overhyped and underwhelming. And that’s not the worst of it.
Dozens of shots on goal (Grok, Claude, Gemini) etc have invariably faced the same problems. Distribution shift has never been solved.
- Where
- paragraph 83
- Published
- 2025-08-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gpt-5-overdue-overhyped-and-underwhelming-b24250-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-gpt-5-overdue-overhyped-and-underwhelming-b24250-a1.