Title: Paul Christiano on how you might get consequentialist behavior from large language models
Author: Paul Christiano
URL: https://forum.effectivealtruism.org/posts/dgk2eLf8DLxEG6msd/how-would-a-language-model-become-goal-directed?commentId=cbJDeSPtbyy2XNr8E
Why it's good: I think lots of people are very wrong about how LLMs might lead to consequentialist behavior, and Paul's comment here is my favorite attempt at answering this question. I think that this question is extremely important.

EA-Aligned Political Activity in a US Congressional Primary: Concerns and Proposed Changes by Carolina_EA (78 karma)
Why it's good: I am so, so appreciative when people share detailed, good-faith takes on parts of EA from perspectives that we rarely get such insight into. (Similar posts with at least the same karma, covering disagreement, agreement, praise, indifference, justification, advice: Podcast: The Left and Effective Altruism with Habiba Islam, A subjective account of what it's like to join an EA-aligned org without previous EA knowledge, Notes From a Pledger, Some unfun lessons I learned as a junior grantmaker, Some observations from an EA-adjacent (?) charitable effort.)

Title: Epistemic Legibility - "Tl;dr: being easy to argue with is a virtue, separate from being correct."

Author: Elizabeth

URL: https://forum.effectivealtruism.org/posts/oRx3LeqFdxN2JTANJ/epistemic-legibility 

Why it's good: 

Things that make texts or discussions more "epistemically legible" include: making clear what you actually believe, making clear the evidence you're really basing your beliefs on (don't just search for a random hyperlink for a claim — say why you really believe this, even if it's "gut feeling" or "anecdotal evidence"), making the logical steps of your argument clear (don't just list assorted evidence and a conclusion — explain what leads to what and how), use examples, pick a vocabulary that's appropriate to your audience, and write the argument down.^[1] I also think that summaries or outlines help.

So why is this important?

I think a lot of discussions are confused for a bunch of reasons. One of these is that it's hard to understand exactly what other parties are saying and why, both because communicating clearly is hard and because we have a tendency to want to hedge and protect our views — making it harder for others to see how we might be wrong (e.g. because of impostor syndrome). 

For instance, I might want to say, "I think a lot of discussions are confused for a bunch of reasons," and walk away — especially if I find a good hyperlink for "confused" or something. But that would make it very hard to argue with me. I didn't explain what "confused" really means to me, I didn't list specific reasons, I didn't say which discussions, or approximately how many of them. (So what do you argue with? "No, I think very few discussions are 'confused'?") I could, instead, write something more specific, like "I think that too many posts on the Forum (for my taste) lead to discussions that misinterpret the claims of the related post or are arguing about details or logical connections that aren't actually relevant. This happens for a bunch of reasons, some of which I could list, but I'm focusing on a specific thing here that is one of what I'd guess are the top 10 reasons, and here's how that happens..." This is still pretty vague, but I think it's better. You can now say, "Here's my list of 10 reasons that contribute to this phenomenon, and epistemic illegibility doesn't get into the top 10 — which do you think are less important?" And I imagine that this leads to a more productive discussion.

There are downsides and costs to being more specific or epistemically legible like this — Elizabeth's post acknowledges them, and notes that not everything should necessarily be epistemically legible. For instance, the rewritten claim above is messier and longer than the original one. (Although I don't think this always has to be true.) But on the margin, I think I'd prefer more posts that are messier and even longer if they're also more epistemically legible. And I really like the specific suggestions on how to be legible. 

Or, as Elizabeth puts it, 

If I hear an epistemically legible argument, I have a lot of options. I can point out places I think the author missed data that impacts their conclusion, or made an illogical leap. I can notice when I know of evidence supporting their conclusions that they didn’t mention. I can see implications of their conclusions that they didn’t spell out. I can synthesize with other things I know, that the author didn’t include.

If I hear an illegible argument, I have very few options. Perhaps the best case scenario is that it unlocks something I already knew subconsciously but was unable to articulate, or needed permission to admit. This is a huge service! But if I disagree with the argument, or even just find it suspicious, my options are kind of crap. I write a response of equally low legibility, which is unlikely to improve understanding for anyone. Or I could write up a legible case for why I disagree, but that is much more work than responding to a legible original, and often more work than went into the argument I’m responding to, because it’s not obvious what I’m arguing against.  I need to argue against many more things to be considered comprehensive. If you believe Y because of X, I can debate X. If you believe Y because …:shrug:… I have to imagine every possible reason you could do so, counter all of them, and then still leave myself open to something I didn’t think of. Which is exhausting.

I could also ask questions, but the more legible an argument is, the easier it is to know what questions matter and the most productive way to ask them. 

I could walk away, and I am in fact much more likely to do that with an illegible argument. But that ends up creating a tax on legibility because it makes one easier to argue with, which is the opposite of what I want.

I also love reasoning transparency, but feel like it gets at the quality in a different way, with a different emphasis. And I've also been using "butterfly idea" a lot.

Title: Working with the Beef Industry for Chicken Welfare

Author: RobertY

URL: https://forum.effectivealtruism.org/posts/5XKAsEBMuxiycTHL7/working-with-the-beef-industry-for-chicken-welfare

Why it's good: Correct focus on a source of immense, totally unnecessary suffering, with outside-the-box thinking to help mitigate the suffering. Thanks, Robert!

Title: The long reflection as the great stagnation 

Author: Larks

URL: https://forum.effectivealtruism.org/posts/o5Q8dXfnHTozW9jkY/the-long-reflection-as-the-great-stagnation 

Why it's good: Powerful attack on a cherished institution. I don't necessarily agree on the first order, but on the second order people will act up and ruin the Reflection.

Paper summary: The Epistemic Challenge to Longtermism (Christian Tarsney) by Global Priorities Institute, EJT (37 karma)
Why it's good: An attempt to put numbers on the much-disputed tractability of longtermism (I also really appreciate summaries).

The Possibility of Microorganism Suffering by Elias Au-Yeung (46 karma)
Why it's good: I think there should be more investigations into potential welfarist priorities that sound super weird. (Similarly, though higher-karma: Do Brains Contain Many Conscious Subsystems? If So, Should We Act Differently?, Reducing nightmares as a cause area.)

Title: Forecasting Newsletter: April 2222

Author: Nuno

URL: https://forum.effectivealtruism.org/posts/xnPhkLrfjSjooxnmM/forecasting-newsletter-april-2222 

Why it's good: Incredible density of gags. Some of the in-jokes are so clever that I had to think all day to get them; some are so niche that no one except Nuno and the target could possibly laugh.

Some Carl Sagan quotations by finm (77 karma)
Why it's good: I really appreciate inspiring posts. (Higher-karma posts on this topic that I really appreciated: Open call for EA stories, 🎨 Altruist Dreams - a collaborative art piece from EAG SF, Good things that happened in EA this year, The Spanish-Speaking Effective Altruism community is awesome.)

Consequentialism and Cluelessness by Richard Y Chappell (27 karma)
Why it's good: Raises and defends an important point that I think would release a lot of people from cluelessness-induced paralysis if more widely shared, namely that Option A can still have higher expected value than Option B despite us having no clue what many of the consequences will be, because these invisible consequences speak neither for or against either option. (Another important point that I wish was better known is that Person-affecting intuitions can often be money pumped, although this post was essentially a repeat of a post from 2020.)

Space governance - problem profile by finm, 80000_Hours (65 karma)
Why it's good: A new entry to 80,000 Hours' most pressing world problems that is huge in scale, severely neglected from a longtermist perspective, although admittedly not especially tractable (which is why I'd love to see more posts like Influencing United Nations Space Governance).

Observations of community building in Asia, a 🧵 by Vaidehi Agarwalla (36 karma)
Why it's good: Concrete ideas for improving diversity in EA. (Similar posts I particularly appreciated with more karma: Top down interventions that could increase participation and impact of Low and Middle Income Countries in EA, EA career guide for people from LMICs, A few more relevant categories to think about diversity in EA.)

Criticism of the main framework in AI alignment by Michele Campolo (34 karma)
Why it's good: Explores an area of AGI alignment that I think is under-discussed, namely the possibility of using AGI for direct moral progress. (Other under-discussed areas with more karma: AGI and Lock-In, Robert Long on Why You Might Want To Care About Artificial Sentience, ‘Dissolving’ AI Risk – Parameter Uncertainty in AI Future Forecasting, Steering AI to care for animals, and soon.)

Some core assumptions of effective altruism, according to me  by peterhartree (85 karma)
Why it's good: I think people should focus on EA's core assumptions much more than they do and I think this list is pretty accurate (and more accurate than the list it is responding to).

Title:

EA’s brain-over-body bias, and the embodied value problem in AI alignment

Author:

Geoffrey Miller

URL:

https://forum.effectivealtruism.org/posts/zNS53uu2tLGEJKnk9/ea-s-brain-over-body-bias-and-the-embodied-value-problem-in

Why it's good:

Embodied cognition is a hot topic in cognitive science. Are AI safety people overlooking this?

From Geoffrey’s introduction:

Evolutionary biology and evolutionary medicine routinely analyze our bodies’ biological goals, fitness interests, and homeostatic mechanisms in terms of how they promote survival and reproduction. However the EA movement includes some ‘brain-over-body biases’ that often make our brains’ values more salient than our bodies’ values. This can lead to some distortions, blind spots, and failure modes in thinking about AI alignment. In this essay I’ll explore how AI alignment might benefit from thinking more explicitly and carefully about how to model our embodied values.

Big, if there’s something to it. But the piece received one three word comment...

I'll bite on the invitation to nominate my own content. This short piece of mine spent little time on the front page and didn't seem to capture much attention, either positive or negative. I'm not sure why, but I'd love for the ideas in it to get a second look, especially by people who know more about the topic than I do.

Title: Leveraging labor shortages as a pathway to career impact? [note: question mark was added today to better reflect the intended vibe of the post]

Author: Ian David Moss

URL: https://forum.effectivealtruism.org/posts/xdMn6FeQGjrXDPnQj/leveraging-labor-shortages-as-a-pathway-to-career-impact

Why it's good: I think it surfaces an important and rarely-discussed point that could have significant implications for norms and practices around EA community-building and career guidance if it were determined to be valid.