“One more way to look at this, is this graph I just saw: enormous convergence on GPT-4 level performance, in multiple models released since, yet nothing decisively ahead.”
Our reading, the sentence that was checked:
Multiple models released since GPT-4 have converged on GPT-4-level performance with none decisively ahead.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Evidence that LLMs are reaching a point of diminishing returns - and what that might mean, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-evidence-that-llms-are-reaching-a-7f3d52-a1
We recorded 20 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 2024-04-13. 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
- GPT-4 level performance
- multiple models released since
- nothing decisively ahead
- this graph
Our interpretation
- The ordinary reading is that several newer models had broadly comparable capabilities to GPT-4, without a substantial overall leader. A small win on one benchmark would not alone contradict this.
- GPT-4 could mean the original model or the updated GPT-4 family. Both readings must be checked.
- For this ongoing comparison, use the latest pre-post release-and-evaluation window, March 1 through April 13, 2024. This reporting window is an evaluator choice.
Counts as supported if: Dated comparisons show several newer models clustering near the specified GPT-4 baseline across varied tasks, without a substantial and robust overall lead.
Counts as contradicted if: Comparable evaluations show a newer model substantially and consistently ahead of the specified GPT-4 baseline, or show that the claimed clustering did not occur.
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
Aider found Opus only slightly ahead of Turbo on code editing. Arena also reported strengths for the April Turbo update, while contemporary reporting distinguished newer models from original GPT-4. These records support some convergence but do not settle the broader assertion. The missing graph and unspecified GPT-4 version matter enough to withhold a verdict.
Limitations
- A coding benchmark cannot establish general capability parity.
- Ranking first does not by itself establish a decisive lead.
- The original graph was absent from the readable source.
Evidence (3 items)
Supports the claim: Aider,
a few points
Why it matters here: This supports close performance between Opus and Turbo on code editing. It does not establish convergence across the entire frontier or against original GPT-4.
- Publisher
- Aider
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: LMSYS / Chatbot Arena,
coding & reasoning
Why it matters here: The evaluator reported strengths for the April Turbo update, which weighs against a blanket claim of no improvement. The readable post supplies neither effect sizes nor confidence intervals sufficient to establish a decisive overall lead.
- Publisher
- LMSYS / Chatbot Arena
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: The Decoder,
original GPT-4.
Why it matters here: This is material evidence against treating original GPT-4 and Turbo as interchangeable baselines. It reports a ranking advantage without establishing its size or broader significance.
- Publisher
- The Decoder
- Kind of source
- News report
- Describes events of
- Public from
What would settle it
Recover the graph and dated underlying results, identify the GPT-4 baseline, and compare the leading models across tasks with uncertainty estimates.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Evidence that LLMs are reaching a point of diminishing returns - and what that might mean
One more way to look at this, is this graph I just saw: enormous convergence on GPT-4 level performance, in multiple models released since, yet nothing decisively ahead.
- Where
- paragraph 18
- Published
- 2024-04-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-evidence-that-llms-are-reaching-a-7f3d52-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-evidence-that-llms-are-reaching-a-7f3d52-a1.