“that models will continue to get more efficient and less expensive, but that hallucinations and reliability problems will persist.”
Our reading, the sentence that was checked:
Models have gotten more efficient and cheaper, but hallucinations and reliability problems have persisted.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, China has all but caught up. The US is not going to “win” the AI war. Here’s what we should do instead., paragraph 19.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-china-has-all-but-caught-up-the-us-284a22-a1
We recorded 15 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 2026-07-20. 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
- models will continue to get more efficient and less expensive
- hallucinations and reliability problems will persist
- That’s basically what has happened.
Our interpretation
- Assess the retrospective trend from the January 2025 prediction through July 20, 2026.
- The ordinary reading is declining resources and price for comparable useful capability, alongside continuing factual errors and unreliable outputs.
- This does not require every model, the latest flagship, or total customer spending to become cheaper.
- Efficiency and price are separate propositions; a price cut alone does not prove computational efficiency.
Counts as supported if: Within-period evidence documents lower prices and improved computational efficiency at broadly comparable capability, while contemporary research documents continuing hallucinations or reliability failures.
Counts as contradicted if: Representative evidence rejects the comparable-capability cost and efficiency trend, or establishes that hallucinations and reliability failures no longer persist.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
The ordinary comparable-capability reading is supported. DeepSeek documented improved efficiency and sharply lower prices while retaining comparable benchmark performance during the claimed period. Broader research corroborates the direction of price-performance and efficiency gains. Peer-reviewed research in April 2026 explicitly finds that hallucinations still persist. This does not establish that the newest frontier models or complete agent tasks always become cheaper.
Limitations
- The concrete within-period efficiency comparison is vendor-reported.
- The broad independent cost study also includes months before January 2025 and is context only.
- Falling unit or capability-adjusted costs can coexist with rising total inference spending.
Evidence (3 items)
Supports the claim: DeepSeek,
reducing compute cost
Why it matters here: DeepSeek reports improved long-context efficiency, benchmark parity with V3.1-Terminus and an immediate API price reduction exceeding 50%. This is a concrete within-period example supporting both efficiency and price, although the performance comparison is vendor-reported.
- Where
- Efficiency Gains
- Publisher
- DeepSeek
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: Nature,
the problem persists
Why it matters here: The paper identifies continuing confident falsehoods and their consequences for reliability. It directly supports persistence, without implying that error rates never improve.
- Where
- Abstract
- Publisher
- Nature
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: arXiv,
algorithmic efficiency improvements
Why it matters here: The April 2024–November 2025 dataset finds substantial price reductions at fixed benchmark capability and algorithmic efficiency gains. It also finds increasing costs for running the advancing frontier, so it supports the interpretation while materially qualifying an unadjusted-price reading.
- Publisher
- arXiv
- 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, China has all but caught up. The US is not going to “win” the AI war. Here’s what we should do instead.
that models will continue to get more efficient and less expensive, but that hallucinations and reliability problems will persist.
- Where
- paragraph 19
- Published
- 2026-07-20
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-china-has-all-but-caught-up-the-us-284a22-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-china-has-all-but-caught-up-the-us-284a22-a1.