Twitxy Saved

Drake Thomas

Verified account This account is verified. Learn more

@MaskedTorah

System cards, risk reports, and misc safety takes at Anthropic; math; puzzles; spaced repetition. Writes with too many caveats for Twitter.
Berkeley, CA admonymous.co/drake Joined April 2014
536 Following 7.6K Followers 3.8K Posts

Latest posts 5

Drake Thomas Verified account @MaskedTorah
I wish more people knew Rotblat's story: he was the only person to decide to leave the Manhattan project when its original motivation (to beat the Nazis to the bomb) was falsified. I have a "What Would Joseph Rotblat do?" sticker on my work laptop with this image.
Lizka @LizkaVaintrob 5d
I think of November 4th as Rotblat Day.

Joseph Rotblat was the only scientist to leave the Manhattan Project on moral grounds. I have a little “Rotblat doc” inspired by him, where I’ve written down what motivates me and what might cause me to leave a job (or seriously consider it). I think of this as a way to avoid getting swept up or drifting over time in ways I don’t endorse.

At the moment, though, I get even more out of his later work against nuclear weapons and war -- especially what others described as his “enduring optimism” and “pragmatic idealism” in the face of catastrophe.

In his own words:

> I would describe myself as a realistic pacifist although this sounds like an oxymoron. [...] Even events that seem out of this world can be realized if we put enough effort and faith into them. One such event is a world without war.

and:

> A campaign for abolition [of nuclear weapons] based on moral principles will be seen as a fanciful dream by many. But the situation is grim; the way things are moving is bound to lead to catastrophe. If there is a way out, even if seemingly unrealistic, it is our duty to pursue it.
View quoted post on X
11 replies 22 reposts 229 likes 8.1K views View on X
Drake Thomas Verified account @MaskedTorah
Most of the things @aaaronson posts are doable by thinking of creative interesting word properties and asking LLMs to search for / maximize them, which is great, but tweets like these show that he also has a natural talent for observing these things “in the wild”.
Adam Aaronson Verified account @aaaronson Oct 28
Everyone's talking about the Anthropic IPO but no one's talking about how AnthrOPIC IPO ends with a 7-letter palindrome
View quoted post on X
Drake Thomas Verified account @MaskedTorah
Really glad METR looked into the response biases here. It appears that our default prompts for asking Claude to review public alignment artifacts are unusually prone to yielding favorable reviews, compared to when a non-employee asks or to alternate wordings.
Elizabeth Barnes Verified account @BethMayBarnes Oct 21
The differences here are not huge - 5-6pp of definitely-not-p-hacked effect (ie default METR-written prompt vs original Anthropic prompt), 10pp if you take the biggest gap. (I don't know why the French are the most concerned...) But I was a little surprised the original prompt was the most favorable across a few different variations and metrics. Given how good the models are at author attribution there's a good chance they know who wrote each prompt, which would potentially explain that pattern if it's ~Ant ~sycophancy plus eval awareness. Lots of interesting things to look into here!
View quoted post on X
Drake Thomas Verified account @MaskedTorah
Minor safety governance update: Anthropic's Responsible Scaling Officer is now Sam McCandlish (was previously Jared Kaplan). I've seen misc AIS people be unsure about who the RSO is before - anthropic.com/company/leader… has documented it for ages but in a fairly obscure fashion. (At a glance I think this page is out of date in a couple other minor ways, but its labeling of the RSO is correct.)

Anyway I don't think this is super notable but it's probably at least one tweet's worth of notable and I'm not sure anyone was going to notice otherwise so I am tweeting about it. I don't expect there to be major changes in RSP-related decisionmaking as a result of this.
www.anthropic.com Leadership at Anthropic Meet the leadership team at Anthropic. Learn about the experienced leaders guiding our mission to build reliable, interpretable, and steerable AI systems.
2 replies 2 reposts 167 likes 20K views View on X
Drake Thomas Verified account @MaskedTorah
I'm confused about the level of badness of the Australian government "hack". From chatting with Claude about public reporting so far, I can't tell if anything more interesting than "got around a basic anti-scraping measure and read some publicly accessible URLs" happened?