Other researchers and engineers have been using Anthropic's released interpretability tools and watching its videos.
Mixed Partly true and partly false in roughly equal measure.
Factual: Statements about the present or the past, checked against the record: did they hold up?
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
Judged as of 2022. True: there is evidence that, by the time of the statement, outside researchers or engineers had used Anthropic's released interpretability tools (e.g. PySvelte, Garcon-related or circuits tooling) and engaged with its interpretability videos, through citations, GitHub forks, issues or stars, follow-up projects, or public discussion. Mostly true: there is clear evidence of outside engagement with either the tools or the videos, and lighter evidence for the other. Mixed: there is evidence of engagement with only one of the two (tools or videos), and none for the other. Mostly false: engagement is negligible, for example only Anthropic staff or a handful of views or forks. False: the tools or videos had not been released, or there is no outside use at all.
Set before the research began.
Evidence (4)
-
E1 GitHub - TomFrederik/PySvelte: A library for bridging Python and HTML/Javascript (via Svelte) for creating interactive visualizations GitHub (user TomFrederik) · 10 February 2022
“Unofficial fork of Anthropic's PySvelte for continuous production and support”
-
E2 GitHub - anthropics/PySvelte: A library for bridging Python and HTML/Javascript (via Svelte) for creating interactive visualizations GitHub (anthropics organisation) · 18 December 2021 Own record
“228 stars”
-
E3 GitHub - anthropics/PySvelte: A library for bridging Python and HTML/Javascript (via Svelte) for creating interactive visualizations GitHub (anthropics organisation) · 18 December 2021 Own record
“This repository was archived by the owner on Dec 23, 2021. It is now read-only.”
-
E4 A Mathematical Framework for Transformer Circuits Transformer Circuits Thread (Anthropic) · 22 December 2021 Own record
“We've released one of the libraries we made for our research, PySvelte (which makes it convenient to use interactive visualizations from Python), and written an article describing another one called Garcon”
Reasoning
An independent GitHub fork of Anthropic's PySvelte dated 2022-02-10, describing itself as an 'Unofficial fork of Anthropic's PySvelte for continuous production and support', shows outside engagement with the released tools (E1). Anthropic's own pages show the tool was released (E4) and that the repository has 228 stars (E2), but the star count is undated. No evidence addresses outside engagement with Anthropic's interpretability videos, so only the tools half of the claim is supported. Under the criteria this fits 'mixed'.
Cite this claim
Daniela Amodei, “Other researchers and engineers have been using Anthropic's released interpretability tools and watching its videos.” (said 2022). Verdict: mixed. Reliability API, claim c_04d63dc6a843, method 0.1.0+28da144b, dataset release v1.1.2. https://epistacklabs.com/claims/c_04d63dc6a843
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Daniela Amodei
-
Resolved true
“There’s medical research showing that people who have a good relationship with their doctor tend to have better clinical outcomes.”
-
Resolved true
“we released a writeup on Garcon, which is the infrastructure tool that our team used to conduct the research”
-
Mostly false
“This was an immovable baseline established from day one.”
-
Mostly true
“In the lab, Claude designed new protein binders from scratch.”
In laboratory work, Anthropic's Claude designed new protein binders from scratch.
-
Mostly true
“including open-source code much of the internet runs on, before attackers could exploit them.”
Verdicts are made automatically; how verdicts are made.