“Software like LaMDA simply doesn’t; it doesn’t even try to connect to the world at large, it just tries to be the best version of autocomplete it can be, by predicting what words best fit a given context.”
Our reading, the sentence that was checked:
Software like LaMDA does not connect to the world; it is just autocomplete predicting which words fit a context.
Assessment: Contradicted Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Nonsense on Stilts, paragraph 17.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-nonsense-on-stilts-52d256-a1
We recorded 12 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 2022-06-12. 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
- Software like LaMDA
- doesn’t even try to connect to the world at large
- just tries to be the best version of autocomplete
- predicting what words best fit a given context
Our interpretation
- Assess the publicly described LaMDA dialogue system, including its grounding tools, rather than only its pretrained base model.
- The ordinary technical reading of connecting to the world includes consulting external information to ground factual answers. It does not require consciousness or human understanding.
- For this ongoing technical observation, use the current-year documentation window from the beginning of 2022 through the post date. There is no standard reporting quarter for this system.
- Next-token prediction can coexist with external retrieval. Establishing the former alone does not establish the claim's restrictive 'just' and 'doesn’t even try'.
Counts as supported if: Technical documentation establishes that LaMDA only generates contextual text and contains no mechanism or training objective for consulting external information to ground its answers.
Counts as contradicted if: Technical documentation establishes that LaMDA explicitly attempts factual grounding through external knowledge sources or tools.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Contradicted
The dated evidence goes against 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
Google's published LaMDA system explicitly learned to consult external knowledge and tools to ground responses. That contradicts 'doesn’t even try' under the ordinary technical reading. Next-token prediction remains part of its operation, but does not establish that the complete system is only autocomplete. This finding does not establish understanding or sentience.
Limitations
- The result depends on reading external grounding as a connection to the world.
- The evidence concerns the documented research system, not the configuration of every private demonstration.
Evidence (2 items)
Goes against the claim: Google Research authors / arXiv,
external knowledge resources and tools.
Why it matters here: The paper describes retrieval, calculator and translation tools, and training that routes queries to those tools during inference. This directly contradicts the assertion that LaMDA does not even attempt external grounding.
- Publisher
- Google Research authors / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Google Research,
pre-training and fine-tuning.
Why it matters here: The surrounding sections confirm next-token pretraining, but also describe safety classification, response ranking and external factual grounding. Marcus's description captures the generative training objective without capturing the complete system.
- Publisher
- Google Research
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Nonsense on Stilts
Software like LaMDA simply doesn’t; it doesn’t even try to connect to the world at large, it just tries to be the best version of autocomplete it can be, by predicting what words best fit a given context.
- Where
- paragraph 17
- Published
- 2022-06-12
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-nonsense-on-stilts-52d256-a1: Current. Outcome: Contradicted. 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.