“But putting LLMs into production is hard; most work to date has been preliminary.”
Our reading, the sentence that was checked:
Putting LLMs into production is hard and most deployment work to date has been preliminary.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Happy 1st Birthday, GPT-4, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-happy-1st-birthday-gpt-4-fe0f67-a1
We recorded 7 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 2024-03-16. 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
- putting LLMs into production is hard
- most work to date has been preliminary
Our interpretation
- 'To date' and the birthday context describe cumulative deployment work during GPT-4's first year, ending at publication.
- The plain reading of 'hard' is recurring material engineering or organizational barriers to dependable production use.
- 'Most' means more than half of LLM deployment efforts; 'preliminary' means experiments, proofs of concept or pilots.
- The share of companies with any production deployment is not the share of projects in production.
Counts as supported if: Engineering evidence documents recurring production barriers, and reasonably representative evidence shows more than half of LLM deployment efforts remained preliminary.
Counts as contradicted if: Representative project-level evidence shows that most LLM deployment efforts had reached routine production, or credible broad evidence shows that production deployment generally lacked material barriers.
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
The difficulty component is supported by documented hallucination and mitigation problems. Surveys also show substantial experimentation, but they count organizations or executives rather than LLM deployment efforts. Their results cannot establish that most work remained preliminary by March 16. The combined claim therefore remains unresolved.
Limitations
- Partial support for deployment difficulty does not settle the majority claim.
- Successful individual deployments can coexist with most projects remaining preliminary.
- Survey scope, timing and sampling limit generalization.
Evidence (3 items)
Supports the claim: Tonmoy et al., arXiv,
addressing and mitigating hallucinations.
Why it matters here: The review documents hallucinations and the limitations of mitigation methods as obstacles to practical LLM use. It supports the difficulty component, but supplies no project-level production share.
- Publisher
- Tonmoy et al., arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: The New Stack,
only used LLM for experimentation purposes.
Why it matters here: Reporting on 150 respondents surveyed from May through July says 44% described experimentation only and 13% reported at least one production LLM. This counts organizations, not deployment efforts, and is secondary reporting because the primary report was inaccessible.
- Publisher
- The New Stack
- Kind of source
- News report
- Describes events of
- Public from
Context only: Boston Consulting Group,
Some 90% of leaders
Why it matters here: The categories combine leaders expecting AI to take time to move beyond hype with leaders pursuing limited experiments and pilots. This broader AI and GenAI survey counts leaders, overlaps categories and does not establish the share of LLM projects in production; its surrounding references to 'next year' place the responses in 2023 without precise field dates.
- Publisher
- Boston Consulting Group
- Kind of source
- Independent measurement
- Describes events of
- Public from
What would settle it
A representative survey or project inventory covering the target period, identifying LLM projects and classifying each as experimental, pilot or routine production, alongside documented deployment barriers.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Happy 1st Birthday, GPT-4
But putting LLMs into production is hard; most work to date has been preliminary.
- Where
- paragraph 4
- Published
- 2024-03-16
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-happy-1st-birthday-gpt-4-fe0f67-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.