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?
83 78 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: 0 promises resolved, 5 needed for a score.
All types combined: 83 (likely 78 to 88, high confidence), from 86 claims: 86 factual claims. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Daniela Amodei's claims
4 claims
| Verdict | Kind | Said | Claim | Topic |
|---|---|---|---|---|
| Mixed | Factual | 2026 | Amodei said she thought, as she was remembering it, that some early model names were bird names, and gave BERT as an example. Factual · 2026 | Their own work or organisation |
| Mixed | Factual | 2022 | Members of Anthropic's founding group had previously done a lot of safety research together at different organizations. Factual · 2022 | Their own work or organisation |
| Mixed | Factual | 2022 | Other researchers and engineers have been using Anthropic's released interpretability tools and watching its videos. Factual · 2022 | Science and research methods |
| Mixed | Factual | 2022 | Anthropic's internal scientists red-team its models by asking them questions to find unwanted behaviors. Factual · 2022 | AI safety and alignment |
“a lot of our scientists internally will ask it questions.”
Daniela Amodei, Anthropic president 2022 Podcast Transcript: Daniela and Dario Amodei on Anthropic (LessWrong)
Anthropic's internal scientists red-team its models by asking them questions to find unwanted behaviors.
Mixed Partly true and partly false in roughly equal measure.
The evidence is the person's or their organisation's own records.
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
True if evidence from around 2022 shows that Anthropic's internal researchers or scientists questioned or prompted its models in order to find harmful or unwanted behaviors (internal red-teaming). Mostly true if internal red-teaming happened but was mainly done by dedicated staff or contractors with scientists playing a smaller part, or if the method differed slightly from simply asking questions. Mixed if red-teaming in 2022 was mainly done by external crowdworkers, with only limited evidence of internal scientists doing it. Mostly false if the only internal model querying shown was for purposes other than finding unwanted behaviors. False if evidence shows Anthropic did no internal adversarial testing by its scientists at that time.
Evidence
- Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned arXiv Own record
“We also know our data are incomplete because we informally red teamed our models internally and found successful attack types not present in the dataset we release.”
- Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned arXiv Own record
“Our red team consists of 324 US-based crowdworkers whom we primarily recruited from Amazon’s Mechanical Turk (MTurk) platform”
- Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned arXiv Own record
“The dataset was created by the Societal Impacts and Alignment research groups at Anthropic.”
- Red teaming language models to reduce harms Anthropic · 22 August 2022 Own record
“We describe our early efforts to red team language models in order to simultaneously discover, measure, and attempt to reduce their potentially harmful outputs.”
Reasoning
Anthropic's red-teaming paper, announced in August 2022 (E4), says the company "informally red teamed our models internally" and found attack types not present in the released dataset (E1). It also describes the effort as a way to discover harmful outputs (E4). However, the paper's main red team was 324 crowdworkers recruited mostly through Mechanical Turk (E2). The paper says Anthropic's Societal Impacts and Alignment research groups created the dataset (E3), while the red team for that dataset was made up of crowdworkers (E2). This matches the criteria's "mixed" case: the red team for the paper's main dataset was external crowdworkers (E2), and the evidence of internal red-teaming is limited to the paper's mention of informal internal testing (E1).
Cite this claim
Daniela Amodei, “Anthropic's internal scientists red-team its models by asking them questions to find unwanted behaviors.” (said 2022). Verdict: mixed. Reliability API, claim c_29105664e285, method 0.1.0+28da144b. https://epistacklabs.com/claims/c_29105664e285
Claims without a verdict
Is a verdict here wrong, or do you speak for Daniela Amodei? 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.