There's a long list of content-based signals we might want to rank on. We are limited by the classifiers we actually have. BUT we are working on LLM scoring with arbitrary prompts! The challenge is performance, but we have a cunning plan docs.google.com/document/d/1...

How does GreenEarth determine "constructiveness"? We use the Google Perspective API but not the classic "toxicity" score. Instead we combine many of their "bridging" attributes. The weights are from our previous experiment. github.com/greenearth-s... rankingchallenge.substack.com/p/its-possib...

I had the robots make a chart of AI funding (essentially private) and fusion funding (mix of public and private) and you are definitely right.

GreenEarth (@greenearth-social.bsky.social) will be the first feed to explain the algorithm, tell you why you got each post, and let you adjust all the parameters. Here's an early version running on our staging server. We're ~2 weeks away from public beta.

I'm not gonna say he did this because of us, but I know for a fact that his boss read my essay, "Making Social Media Algorithms Social" Competition is good, actually! www.greenearth.social/p/making-soc...

Really I'm just rehashing some classic stuff. There's a modern book which re-covers this territory very well, "Mind, Meaning, and Mental Disorder" (Bolton and Hill, 2003). In general, I find the "functional semantics" view of meaning, representation, behavior and causality very compelling.

I like this one best so far. I think it's clearest, gives the right mental model even if it doesn't match how the parameters are actually implemented

Another attempt. This isn't anywhere near our final feature set btw -- we're adding prompt-based scorers and other things -- and yet it's still more control than any social media user has ever had. Which is absurd.

GreenEarth will be the first fully transparent, controllable, open-source feed. Does this diagram help you understand how the core algorithm works? Try the pre-release version now! survey.qualtrics.com/jfe6/form/SV...

Similar feels

IT'S WORKING! After months of engineering and tuning (thanks team!) the new @greenearth-social.bsky.social feed is reliably serving me an excellent social media experience. It's already engaging and constructive. Transparency and controls coming soon! Try it! survey.qualtrics.com/jfe6/form/SV...

You may remember we did a 10,000 person experiment testing AI-powered healthier feed algorithms. Well, now we’ve built it as a product for BlueSky/ATProto. We are currently recruiting pre-release users to test it out and give us feedback. Want to try it? survey.qualtrics.com/jfe/form/SV_...

If you've ever wanted to mainline the Internet, try our new Random feed. www.greenearth.social/p/welcome-to...

FINALLY managed to load my M4 Macbook enough to get Spotify to stutter. Occasionally. Modern hardware is very powerful, hard to max out.

Is this a good logo for the GreenEarth feed? It's a healthier, user-controllable, open-source, LLM-powered, transparent feed we're building -- now in alpha testing. Try it? Tell us what you think! bsky.app/profile/did:...

X head of product says the most monetized poasters are a bad user experience, actually.

and yet

Feedback Loops -- I particularly worked on this part. Human-AI and AI-AI feedback loops are narrowing the epistemic space from which humans and AI draw. This already drives homogenization, and may lead to fragmentation and more self-referential information environments.

Epistemic risks are threats to humanity's collective capacity to know things accurately, reason well, form beliefs, and maintain a healthy information environment. We focus on three mechanisms: • Persuasion & Manipulation • Cognitive Offloading • Feedback Loops

Humanity's ability to know, reason, judge, and act well is the foundation of science, democracy, crisis response, & management of AI itself. AI poses serious risks to that foundation. New paper on epistemic risks by 30 experts calls for attention and proposes solutions. Link in thread.

Why do we ask if people “approve” of the answer? Should we ask if it’s fair, or informative, or trustworthy instead? It turns out all of these correlate — you get the same answer. This is consistent with previous work on perceptions of news credibility. 7/

Why maximize equal approval? There’s a long history of arguments for pluralistic debate, but “neutrality” also has a key purpose in managing conflict: it maintains trust across parties. Without broadly trusted AI, we’ll end up self-selecting into fragmented AI realities. 6/

Unlike previous work, our definition doesn’t assume left vs. right politics. It looks at the conflict around each issue separately, so it generalizes to any political context or culture. Here’s how our 20 issues align and don’t align with the liberal-conservative axis. 5/

Each participant rated four Reddit question / LLM answer pairs. The questions were intentionally chosen to be leading (in both directions) and the answers were model defaults, plus the models prompted to be “for” or “against” and our experimental “balanced” response. 4/

Here’s how we tested. For 20 controversial issues in US politics, we collected 200 charged questions from Reddit, and tested 8 different model/prompt combinations. 7,434 participants on all sides of these issues rated the responses. 3/

Key finding: neutrality isn't a fantasy. Our experimental “balanced” response wins high approval from both sides at once — even when the sides strongly disagree on the substance. And costs less than 10% in approval vs. answers that agree with you (red and blue arrows). 2/

What could it mean for an AI to be "politically neutral”? And can we measure it? New paper + dataset. We propose a definition that applies to any type of conflict on any topic: a neutral response should maximize approval on both sides of an issue, while keeping that approval balanced. 1/🧵

Did "hate" go up on X/Twitter after they changed their content moderation rules post-Musk? What is "hate" anyway, and what data do we actually have? I've got all the charts and analysis I know you crave! New by me for @techpolicypress.bsky.social www.techpolicy.press/what-are-the...

A peer-reviewed competition, 5 winning algorithms, 9,386 users, three platforms, 6 months, and 43 authors -- that's what it took to prove that social media doesn't have to work the way it does now. Here's the paper. rankingchallenge.substack.com/p/its-possib... /FIN