“That means you can run them (or “posttrain” them), but, crucially, users can’t customize or improve them to remotely the same degree (more about that in a second), and outsiders can’t investigate them to nearly the same degree.”
Our reading, the sentence that was checked:
Open-weight models can be run and post-trained but cannot be customized, improved or investigated nearly as much as open-source models.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Open-source is NOT the same as open-weight, paragraph 10.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-open-source-is-not-the-same-as-open-038fbb-a1
We recorded 8 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 2026-08-10. 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)
Conditions
- compared with open-source models (qualifier, stated by the speaker)
From the speaker
- you can run them
- posttrain
- can't customize or improve them to remotely the same degree
- can't investigate them to nearly the same degree
Our interpretation
- Assess capabilities available as of August 10, 2026.
- The comparison with open-source models is a comparator, not a forecast antecedent.
- Read the claim primarily through the examples in context: changing original training inputs, repeating original training stages and investigating how training choices affect behavior.
- Do not interpret it as saying that released weights cannot be changed, fine-tuned or examined.
- Remotely the same degree means a substantial practical gap across these activities, not a specified difference in benchmark scores.
Counts as supported if: Comparable release records show that weight-only users can run and post-train models but lack materials necessary for a substantial range of original-training interventions and audits that fully open releases enable.
Counts as contradicted if: Weight-only releases provide substantially equivalent access to those interventions and audits, or documented comparisons show no substantial practical gap.
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
Ai2 documents that its fully open tooling enables controlled changes to data mixtures and training objectives. That supports the mechanism behind Marcus's comparison. It does not, alone, establish how much less customization, improvement and investigation the open-weight category permits. The retrieved records do not supply a sufficiently complete comparison for the post-date state, so the broad comparative claim remains unresolved.
Limitations
- The cited comparator predates the target period and is context only.
- More available training materials do not by themselves establish larger achievable performance gains.
- No conclusion here treats fine-tuning or internal inspection as impossible.
Evidence (1 item)
Context only: Allen Institute for AI,
controlled ablations
Why it matters here: Documents the specific experimental access Marcus attributes to fully open releases. It establishes one side of the comparison, not the breadth of the disadvantage across open-weight releases on the post date.
- Publisher
- Allen Institute for AI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A dated comparison of representative weight-only and fully open releases showing which original-training interventions and audits outsiders could actually perform, with documented examples of the resulting practical gap.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Open-source is NOT the same as open-weight
That means you can run them (or “posttrain” them), but, crucially, users can’t customize or improve them to remotely the same degree (more about that in a second), and outsiders can’t investigate them to nearly the same degree.
- Where
- paragraph 10
- Published
- 2026-08-10
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-open-source-is-not-the-same-as-open-038fbb-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-open-source-is-not-the-same-as-open-038fbb-a1.