Linkpost for April
You can hire me for life coaching! You can also hire me for more general writing, editing, and freelancing work— email me at ozybrennan@gmail.com. My fiction Substack is here. Recent releases:
You can make a bomb out of anything, Hunger Games fanfiction.
Who prays for Satan?, a web weave about the redemption of Satan.
Mirror maze, horror about being trapped in an infinite hall of mirrors.
dark ilan, dath ilan fanfic.
Effective Altruism
Global Poverty
Development Innovation Ventures was a program within USAID that piloted new interventions, checked if they worked, and helped organizations scale them up if they turn out to be cost-effective. Unfortunately, it was destroyed during the USAID cuts. But good news! It has spun off into a nonprofit called the DIV Fund, backed by philanthropists including Coefficient Giving [Vox].
People in low-income countries produce a hundred times as much plastic pollution as people in high-income countries [Our World In Data]. The difference? People in high-income countries benefit from appropriate waste management programs like landfills. In high-income countries, the primary cause of plastic pollution is littering.
Animal Advocacy
The evidence for whether electrical stunning humanely stuns fish is surprisingly sparse, and many fish species may still feel pain after being electrically stunned [Rethink Priorities Strategic Animal Insights].
Factory farms are cruel because cruelty is profitable, but most people actually don’t like knowing that their food comes from animals raised in cruel conditions [Lewis Bollard]. So it’s a collective action problem. As long as part of the industry is inhumane, it will outcompete the rest of the industry on price. But if the entire industry raises standards at once, we can have an outcome everyone prefers.
Farmed animals might experience more pain than we think [Rethink Priorities Strategic Animal Insights], because pain tends to be less painful if you’re otherwise happy and more painful if you’re otherwise stressed and miserable.
Even as an animal advocate, I was shocked by this visualization of exactly how much bigger chickens are today compared to 1957 [Our World In Data].
Existential Risk
AI slop as an existential risk [Gwern]: “As we contemplate an AI future based heavily on human corpuses followed by autonomous AIs bootstrapping and recursively self-improving, we have to ask: is that going to preserve human values? Or are they going to lose or omit important things—becoming a photocopy of a photocopy of something that once mattered to us? There is no law that any of that has to be maintained, no law that future AIs have to so much as have a sense of humor and enjoy a good pun. (Animals do not share our values, and even humans differ drastically on any specific value, often for what seem to be biological reasons.) It would be a tragedy if AI development continued on its present course, and we sleepwalked into a future where the AI morality was the ‘ChatGPTese of morality’ or the ‘ChatGPTese of esthetics’: some sort of radically simplified caricature of human values, forever.”
AI won’t speed up clinical trials [Ruxandra Teslo]. AIs can in principle develop drugs, but figuring out if the drug works involves a lot of empirical research in the physical world, which is inherently very slow. The regulatory barriers that make clinical trials so expensive will continue to exist even if AIs can develop effective drug candidates.
Both humans and LLMs often write with a lot of bullet points and section headings [Dynomight], even though most people with good taste also agree that excessive formatting is bad writing. Excessive formatting can cover for poor thinking, because it allows you to gesture at several disconnected points and then allow the reader to compile them into an idea. Since LLMs know everything and suck at synthesis, bullet points and section headers can make them look like they’re doing better thinking than they really are.
Related: A deep dive into why chatbots write so goddamn weirdly [New York Times]. AIs combine multiple dialects: for example, the characteristic overuse of “delve” might come from Nigerian English. AIs also use constructions common in high-status, high-quality prose, like em dashes and groups of three. “The A.I. is trying to write well. It knows that good writing involves subtlety: things that are said quietly or not at all, things that are halfway present and left for the reader to draw out themselves. So to reproduce the effect, it screams at the top of its voice about how absolutely everything in sight is shadowy, subtle and quiet. Good writing is complex. A tapestry is also complex, so A.I. tends to describe everything as a kind of highly elaborate textile.”
I don’t agree with this entire article [Eurydice Lives] but I think it makes an excellent point that doing biological research is much more difficult than a lot of handwaving at “the AI will kill us with idk bioweapons” implies. Many of the constraints on biological research are physical-- related to time, equipment, and reaeants. It’s difficult to improve on these processes, even with a hypothetical superintelligence. And it’s pretty likely that a human would notice an AI (say) attempting to buy up large numbers of reagents it wasn’t supposed to have. I think in general a difficult part of predicting the course of AI is that the people most familiar with AI are AI researchers, who are witnessing AIs’ ability in computer programming-- the single area where AIs are uncontroversially the best.
The Pentagon declaring Anthropic a supply chain risk is not remotely legal [Lawfare]. I am particularly struck by how the Trump administration just won’t stop talking. This isn’t the first case where the Trump administration would have had plausible deniability if they had enough self-control not to post that they’re doing something to own the WOKE LIBS who have BETRAYED AMERICA and are DISLOYAL to the PRESIDENT.
Related: the Department of War threatening Anthropic as yet another step in the fall of our democracy [Hyperdimensional].
Pause AI supporters support a multilateral agreement in which both the U.S. and China pause AI, not unilaterally pausing and letting China win the AI race [Astral Codex Ten].
You shouldn’t counter “LLMs can’t think because of X” with “humans also do X” [Experience Machines]. LLMs legitimately do think very differently than human beings do-- and that doesn’t mean that they don’t think.
OpenAI’s foundation is significantly richer than the country of Luxembourg [Vox]. We still don’t know how it’s going to end up spending its money.
American Democracy
ICE officials arrested a 57-year-old U.S. citizen without a warrant and, when criticized for this, falsely accused him of being linked to sex offenders [The Watch]. They said he resembled two convicted sex offenders.
Meta Effective Altruism
Particularly Good: On patriarchy, the commonality of human evil, and moral progress [Good Bones].
Experiments with a simple model of “the optimizer’s curse” [Effective Altruism Forum] (the mathematical fact that high-performing interventions will always turn out to be worse than you expected, and will be even worse the more speculative they are). These experiments overall show that the optimizer’s curse is pretty robust to different distributions of effectiveness, charities being different amounts of uncertain, pessimism about whether you’ll do any good at all, etc. Interestingly, the optimizer’s curse has a smaller effect as the mistakes you make in estimating charity effectiveness are more correlated with each other. That is, the optimizer’s curse has a smaller effect if you’re deciding between malaria vaccines and malaria nets than it does if you’re deciding between malaria vaccines and AI safety research, because many of the mistakes you’d make about the effectiveness of malaria nets also affect the effectiveness of malaria vaccines and that’s not true about AI safety research.
Lessons learned from being a funder [Effective Altruism Forum]. I particularly liked the discussion of how weird your personal relationships become when you become a funder, because you suddenly have to wonder whether your acquaintances or even your friends are just trhying to get money out of you. It reminds me of some articles I’ve read about what it’s like to be a billionaire. I also liked the discussions of the cursed epistemic state that funders are in-- they know that lots of people are shading the truth to try to get more money from them, and it’s very difficult to figure out what’s most effective.
