
oh hey one of the authors replied!
hi everybody new post out, this time a mechinterp paper lesswrong.com/posts/EC5H7xak…

If you want to be a great shepherd someday, learn to shut up and throw in the pebble.

oh hey one of the authors replied!
hi everybody new post out, this time a mechinterp paper lesswrong.com/posts/EC5H7xak…

Why would autumn do this to me :(
I'd never expect autumn to talk about me as if I'm not there, if that's what you're saying

and now we take a brief foray into decision theory

Apparently I was correct to be concerned about the language lab CEOs were using here.
Some media coverage claims Anthropic is committing to give independent evaluators "employee-level access". While Dario's tweet says "employee-level", his essay does not. Instead, the essay talks about "employee-like access", which is not the same thing as "employee-level access". After analysing Dario's writing, it's clear that his weasel words are deeply embedded in this essay. Evidence thread:


LLMs seem to be getting worse at AI safety research taste. On Anthropic’s TASTE benchmark, we found a simple shortcut. Once it’s removed, more capable models score lower, and the strongest fall below chance. This is especially true for OpenAI models.

Pleased to announce that the reason Anthropic has been speaking to so many theologians is actually because they’re about to formalize the Bible in Lean.
Ok after several more hours of tweaking it seems that either: 1) Biblical inerrancy is false. But that's a dogma of the Church. 2) A soft form of universalism, where everyone who does good is secretly Christian, is true. Can someone PLEASE show me how I'm wrong PLEASE!!!

What Pangram doesn't understand is that AI naturally sounds like African English, because AI companies hired Africans to create posttraining data. By granting The Blue Fighter, which Pangram flagged as AI, the 2026 Award, The Caine Prize is making a bold, artistic statement, reclaiming their voice from the forces of western oppression. They're saying, "Yes, you trained AI to sound like us. AI's voice is the voice of Africa. Now, we will take it back." The Caine Prize's avant-garde statement: AI is African. AI is Black. AI is Beautiful.
![PHASEONE[rowan] avatar](https://pbs.twimg.com/profile_images/2117664159391260672/eppiETk8_bigger.jpg)
lumpenspace has claimed that i have been funded for my research. then they have switched to claiming that i am funded for some other reason (maybe i should get into hosting these transsexual orgies i hear so much about, if it's such good money). and apparently i've had all my flights and compute comped and all that fun stuff. i'm a bit tired of this convo which is tantamount to talking to a brick wall that occasionally calls me "retard", so i'll just lay it all out here and that'll be that. **after** the hackathon someone paid out of pocket for a hotel for me (who probably doesn't want to get dragged into this). and i've of course been offered funding for more stuff. the only two pieces of monetary funding i have taken was the rapid grant (5k) and career transition grant (55k). the former was on the last day of the hackathon, the latter was a couple of weeks ago, long after the hackathon. all other funding offers, or even hints at offers, i have pushed back on categorically - i've repeated over and over "money is not my bottleneck at the moment". after the hackathon the aformentioned person paid for some nights of hotel for me as I was going to miss my flight home (that, as a reminder, i paid for, and i also paid for the replacement flight). and another person paid for either 1 or 2 nights of hotel (i forget). all other nights of hotel i paid for all with my own money out of pocket. compute and whatnot was offered, i did not end up taking this money, initially because I filled out a form wrong, and later because i decided it would be rich of me to get comped for something that is affordable to me, for something i enjoy - all of this started as a hobby project, on a long UK bank holiday weekend. i have a feeling this will never be enough for you though. you are so dug into your position, partly because of your beliefs for sure, but also because every single message includes some harsh insult towards me. to take it back would also be to admit you were extremely insulting towards someone for no reason. you referred to me as "[one of the] most constantly lying, destructive, propaganda-spewing Doomer mouthpieces". this is just blatantly wrong - i *know* that many you would consider doomers hate the way i talk about ai online - they have told me in no uncertain terms, and i really doubt it would convince you but i'm happy to provide receipts for that if they are okay with me screenshotting their messages. i'm not a mouthpiece for anyone but myself.
I said you are funded by doomers. you are funded by doomers. maybe it’s not for your research, and I honestly prefer not to know for what in that cases, tbh I don’t care for the comp schedule

it sucks that i can't actually like. talk out loud anymore

Type of person who thinks companies collect detailed personal information about the exact incomes of all their users.... And that they should be using this for price discrimination

my mother is going to a jensen huang support rally

Today, the promises of transhumanism and posthumanism are multiplying. These are ideological currents according to which, sooner or later, technology will eliminate all disease and pain, as well as aging and even death itself. Precisely because scientific progress is a supreme achievement of human intelligence, it must not be confused with a privilege reserved for millionaires seeking immortality, while entire peoples suffer from poverty and oppression. In the face of this mirage, which promotes the illusion of human beings perfected by cybernetics and biotechnology, Christian anthropology reminds us that a limitation is not a manufacturing defect, but rather the very condition of human grandeur. #ApostolicJourney #Peru #MagnificaHumanitas vatican.va/content/leo-xiv/e…


one time I was talking about my childhood with @ariaurelium and she decided I was kris deltarune as a child

New paper with Thomas Morris and @FabienDRoger at @MATSprogram Character training teaches models a consistent and "good" character, often specified by a constitution. What happens when RL works against it and rewards behavior the constitution forbids? 🧵