Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
84 79 88 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
Insufficient record: 2 predictions resolved, 5 needed for a score.
-
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 3 promises resolved, 5 needed for a score.
-
Guidance
A company's own figures for its results over a period, checked against what it then reported.
Insufficient record: 1 guidance claim resolved, 5 needed for a score.
All types combined: 84 (likely 79 to 88, high confidence), from 90 claims: 90 factual claims. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Jeff Bezos's claims
43 claims
“The diversity of products demands that we employ modern regression techniques like trained random forests of decision trees to flexibly incorporate thousands of product attributes at rank time.”
Jeff Bezos, CEO, in shareholder letter 2011 2010 Letter to Shareholders (SEC (Amazon))
Amazon uses trained random forests of decision trees to incorporate thousands of product attributes at rank time.
Unresolvable The public record could not settle it.
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
True: there is evidence that around 2011 Amazon's product search or ranking used trained random-forest or decision-tree-ensemble models that incorporate many product attributes at ranking time. Mostly true: Amazon used tree-ensemble machine-learning ranking (for example gradient-boosted trees) with many features, but the exact method or the attribute count differs slightly. Mostly false: Amazon used machine-learned ranking but not tree ensembles, or only on a small scale. False: the evidence shows Amazon did not use such techniques at the time. If the only source is Amazon's own statement, the claim is consistent with it, but some independent or technical corroboration (A9 papers, engineers' talks, job postings) is preferred.
Evidence
- Amazon's Secret Ranking Algorithm Exposed: How I Tricked an A.I. into Revealing the Hidden Truth! Signalytics · 27 March 2023
“Ranking Models: Gradient boosted trees are used as the main method for ranking because they can discover complex feature interactions and work well without significant tuning.”
- Amazon's Secret Ranking Algorithm Exposed: How I Tricked an A.I. into Revealing the Hidden Truth! Signalytics · 27 March 2023
“Amazon has over a hundred machine-learned models in their system, with each model using about 20 of the 150 available features.”
- The 8 Best Papers on eCommerce Search Algorithms Constructor · 3 March 2020
“This paper dives deeper into the algorithms and ML frameworks Amazon’s A9 team have implemented to rank products in categories, blend separate rankings in All Product Search, implement NLP techniques used for matching queries and products, and more.”
Reasoning
E1 says gradient boosted trees are Amazon's main ranking method, and E2 says Amazon has over a hundred machine-learned models, each using about 20 of 150 available features. Neither quote gives a time period, so neither shows what Amazon used around 2011. The source for both is also weak: a 2023 blog post that says it got its information by prompting an AI. E3 only notes that a paper covers A9's ranking algorithms, without saying what methods were used or when, so the evidence does not settle how Amazon ranked products in 2011.
Cite this claim
Jeff Bezos, “Amazon uses trained random forests of decision trees to incorporate thousands of product attributes at rank time.” (said 2011). Verdict: unresolvable. Reliability API, claim c_a68cebaeec1c, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_a68cebaeec1c
Claims without a verdict
Is a verdict here wrong, or do you speak for Jeff Bezos? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.