“Certainly, adding these plug-ins does improve things.”
Our reading, the sentence that was checked:
Adding the Code Interpreter and Wolfram Alpha plug-ins improves GPT-4's math performance.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Getting GPT to work with external tools is harder than you think, paragraph 2.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-getting-gpt-to-work-with-external-53ff44-a1
We recorded 14 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 2023-10-23. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- adding these plug-ins
- does improve things
- Code Interpreter
- Wolfram Alpha
Our interpretation
- The plain reading is that either named plug-in can improve GPT-4's mathematical answer accuracy relative to GPT-4 without it.
- The claim does not promise improvement on every problem or specify a minimum effect.
- Assess the 2023 systems and evaluations conducted before the October 23 post.
Counts as supported if: Comparisons on the same mathematical tasks show higher accuracy with each named plug-in than with unaided GPT-4.
Counts as contradicted if: Adequate comparisons show no improvement, or worse performance, for the named plug-ins across the relevant mathematical tasks.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
An independent contemporary comparison found higher mathematical accuracy with both named plug-ins than with vanilla GPT-4 on the same 120 problems. This supports the modest claim that adding them improves performance. It does not establish that every tool integration helps or that tools make answers reliable.
Limitations
- The shared test set comprises difficult competition problems.
- The study compares deployed configurations rather than holding every product detail constant.
- Exact experiment dates are not established beyond the 2023 pre-publication period.
Evidence (2 items)
Supports the claim: arXiv; Qingyun Wu and colleagues,
We perform evaluations over 120 randomly selected level-5 problems and on the entire5 test dataset from MATH.
Why it matters here: On the common 120-problem subset, Figure 4a reports 48.33% success for ChatGPT with Code Interpreter, 45.0% with Wolfram Alpha, and 30.0% for vanilla GPT-4. The commercial products were not evaluated on the entire MATH test set.
- Publisher
- arXiv; Qingyun Wu and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Xiaoxuan Wang and colleagues,
Prompts that utilize Python yield impressive improvements while those using Wolfram diminish performance.
Why it matters here: SciBench found that its own Wolfram Language prompting configuration reduced performance. Because it used custom code generation through the API, this qualifies general claims about tools but does not directly contradict the claim about the commercial Wolfram Alpha plug-in.
- Publisher
- arXiv; Xiaoxuan Wang and colleagues
- 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, Getting GPT to work with external tools is harder than you think
Certainly, adding these plug-ins does improve things.
- Where
- paragraph 2
- Published
- 2023-10-23
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-getting-gpt-to-work-with-external-53ff44-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-getting-gpt-to-work-with-external-53ff44-a1.