“The output of DeepResearch will inevitably feed further models; we will wind up with more of this:”
Our reading, the sentence that was checked:
Deep Research output will inevitably be used to train further models, contributing to model collapse.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Deep Research, Deep Bullshit, and the potential (model) collapse of science, paragraph 32.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-deep-research-deep-bullshit-and-the-5ecd5c-a1
We recorded 13 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 2025-02-03. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
From the speaker
- The output of DeepResearch
- will inevitably feed further models
- we will wind up with more of this
- Model collapse
Our interpretation
- DeepResearch means OpenAI's product named in the post.
- Feed further models means use as training data, following the surrounding discussion of models learning from model outputs.
- The contextual forecast includes degradation attributable to that training, not merely the presence of synthetic data.
- No deadline, training proportion or size of degradation is specified.
Counts as supported if: A later model demonstrably trains on OpenAI Deep Research output and controlled evaluation attributes degradation or model collapse to that training.
Counts as contradicted if: Evidence conclusively excludes the predicted pathway or its claimed degradation. Beneficial synthetic-data training in another system does not by itself falsify this open-ended product-specific forecast.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The searched training disclosure concerns GPT-5-generated trajectories, not documented OpenAI Deep Research outputs. General research also shows that synthetic-data training does not automatically produce collapse. Neither source settles this product-specific forecast. No decisive event establishes both ingestion and attributable degradation, and Marcus gave no deadline.
Limitations
- Training datasets may not disclose product-level provenance.
- The predicted collapse consequence is contextual rather than fully stated in the quoted sentence.
- Generic experiments cannot determine the outcome for an undisclosed later training process.
Evidence (2 items)
Context only: Allen Institute for AI,
distillation SFT stage.
Why it matters here: Ai2 describes training on trajectories generated by GPT-5 using its own tools and prompts. This is relevant synthetic-data training, but does not establish ingestion of OpenAI Deep Research output or resulting collapse.
- Where
- SFT for cold start
- Publisher
- Allen Institute for AI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: arXiv,
data accumulation avoids model collapse.
Why it matters here: The experiments distinguish replacing original data from accumulating synthetic data alongside it. This challenges an automatic inference from synthetic-data ingestion to collapse, but predates and does not test OpenAI Deep Research.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
An inspectable training dataset or disclosure identifying Deep Research outputs, paired with controlled evaluations establishing their effect on later-model degradation.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Deep Research, Deep Bullshit, and the potential (model) collapse of science
The output of DeepResearch will inevitably feed further models; we will wind up with more of this:
- Where
- paragraph 32
- Published
- 2025-02-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-deep-research-deep-bullshit-and-the-5ecd5c-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-deep-research-deep-bullshit-and-the-5ecd5c-a1.