Amazon Web Services announced that Mechanical Turk, its crowdsourced marketplace for small digital tasks such as data labeling, transcription, moderation, surveys, and ML evaluation, will close on September 30, 2026, after launching in 2005. Amazon attributed the decision only to a routine assessment and did not identify AI use, economics, or any other specific cause. Separately, a cited 2023 EPFL study estimated that 33–46% of MTurk workers used LLMs for writing tasks, raising doubts about whether marketplace outputs labeled as human-generated reliably have human provenance; the supplied evidence does not establish that this prompted the shutdown.
The reported LLM use by nominally human MTurk workers independently reinforces Scott’s claims that provenance must be captured by construction and that a verifier is not independent when it may share the producer’s model-mediated failure modes. This bears directly on his synthetic-dataset and LLM-grading work because a purchased “human” label or evaluation can no longer be treated as proof of human judgment, although the evidence does not connect that problem causally to Amazon’s shutdown.
ip:framework.provenance-coupled-workip:concept.mechanically-different-verifiersip:concept.correlated-failuredev:concept.synthetic-finetuning-datasetdev:concept.llm-rubric-gradingradar:concept.provenanceradar:concept.model-provenanceradar:concept.human-in-the-loopradar:concept.llm-evaluation
queries asked of Scott's wikis
- human-generated data provenance and verification
- synthetic data contamination of training corpora
- LLM use in human evaluation pipelines
- trusted human-in-the-loop system design
- data-labeling marketplace failure modes
- provenance-aware RAG and knowledge systems
now 0 pts/hpeak 0 pts/hcomments 0/hpeers p0momentum: steady2 platformsage 1154h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
2026-09-10T05:22:54Z
This review adds no evidence: the reported closure and task-specific LLM substitution remain separate claims supported here by reconstructed upstream testimony, not newly inspected primary sources. Keep the case open for closure logistics or fresh provenance findings, without treating the study as evidence of marketplace-wide AI use or the cause of shutdown.
2026-09-08T04:22:48Z
No new evidence changes the reported closure or the task-specific provenance finding; the supplied support remains reconstructed testimony, and neither marketplace-wide LLM prevalence nor a causal link to closure is established. Keep the episode open for operational closure developments, but lengthen review cadence rather than repeatedly revisiting the same synthesis.
2026-09-06T04:22:03Z
No substantive delta warrants changing the interpretation: the reported closure and task-specific LLM substitution remain causally separate, not evidence of marketplace-wide replacement of human labor. The supplied upstream support is reconstructed testimony rather than directly inspected primary sources; further attention should depend on operational closure details or new provenance evidence.
2026-09-04T03:27:46Z
The case remains a pair of established but causally separate facts: MTurk is scheduled to close, and one study found substantial LLM substitution in a specific MTurk task. With no new evidence, the provenance implication stands but the shutdown narrative has not broadened or accelerated.
2026-09-02T02:25:42Z
No new evidence arrived: the scheduled closure and task-specific provenance failure remain established, while any causal connection between LLM use and Amazon’s decision remains unsupported.
2026-08-31T01:29:56Z
The refreshed discussion adds only a generic claim that MTurk long suffered cheating, not evidence about LLM prevalence or Amazon’s reason for closure. The shutdown and task-specific provenance risk remain established but causally separate.
2026-08-30T21:31:02Z
The case now rests on two independent primary lines: Amazon establishes the shutdown, while the EPFL study establishes LLM substitution in a specific MTurk task. This supports the provenance risk but still does not show that AI use caused the closure or characterize the marketplace as a whole.
2026-08-30T21:29:21Z
grounded: converges/medium — The reported LLM use by nominally human MTurk workers independently reinforces Scott’s claims that provenance must be captured by construction and that a verifi
2026-08-30T21:25:40Z
origin walked (codex/luna, conf 0.97): anchor reddit.post.1w2snwd -> echo.other.928e3bde7e by Amazon Mechanical Turk
2026-08-30T21:24:42Z
case created — The reported shutdown is a bounded infrastructure event with relevant implications for human-label supply and training-data provenance, but it currently has only a secondary Reddit artifact as evidence.