LessWrong (30+ Karma)

The Unslop prize has reminded me of my terrified fascination with AI Generated fiction. I have come to identify more as a writer since my post from last year when I posted Experiments With Sonnet 4.5's Fiction, and so have avoided repeating the experiment with Opus 4.6 - mostly out of terror. However, a new, smarter Claude is on the way (and The Unslop prize will force me to learn anyway - as I will not be able to resist reading the winning entry) - so it seems like time to do so. In my last post, I put every story I had written in Claude's context and then asked it to write a story to the best of its ability. Here I have done the same. I have since written far more. Adding them to Claude's context was easier as there is now this super cool website called https://tomasbjartur.com where all my fiction lives, and is even available in a convenient eBook form. And wow, it's even free. Opus produced the following story. Interestingly, unprompted it chose the therapeutic context again. My fiction in its context, Perhaps Claude is trying to tell me something. In [...] --- First published: March 31st, 2026 Source: https://www.lesswrong.com/posts/Yf3dnBPbKkQsXsH2t/experiments-with-opus-4-6-s-fiction --- Narrated by TYPE III AUDIO.

The AI Doc: Or How I Became an Apocaloptimist is a brilliant piece of work. (This will be a fully spoilorific overview. If you haven’t seen The AI Doc,I recommend seeing it, it is about as good as it could realistically have been, in most ways.) Like many things, it only works because it is centrally real. The creator of the documentary clearly did get married and have a child, freak out about AI, ask questions of the right people out of worry about his son's future, freak out even more now with actual existential risk for (simplified versions of) the right reasons, go on a quest to stop freaking out and get optimistic instead, find many of the right people for that and ask good non-technical questions, get somewhat fooled, listen to mundane safety complaints, seek out and get interviews with the top CEOs, try to tell himself he could ignore all of it, then decide not to end on a bunch of hopeful babies and instead have a call for action to help shape the future. The title is correct. This is about ‘how I became an Apolcaloptimist,’ and why he wanted to be that, as opposed to [...] ---Outline:(03:37) Babies Are Awesome(04:58) People Are Worried About AI Killing Everyone(06:17) Freak Out(06:47) Other People Are Not Worried About AI Killing Everyone(09:27) Deepfaketown and Botpocalypse Soon(10:15) Stopping The AI Race and A Narrow Path(11:47) CEOs Know Their Roles(13:28) The Call To Action --- First published: March 31st, 2026 Source: https://www.lesswrong.com/posts/ppC6geY4FxGYifrWx/movie-review-the-ai-doc --- Narrated by TYPE III AUDIO. ---Images from the article:Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

tl;dr: progress on making Claude friendly[1] is not the same as progress on making it safe to build godlike superintelligence. solving the former does not imply we get a good future.[2] please track the difference. The term Alignment was coined[3] to point to the technical problem of understanding how to build minds such that if they were to become strongly and generally superhuman, things would go well. It has been increasingly adopted by frontier AI labs and much of the rest of the AI safety community to mean a much easier challenge, something like "having AIs that are empirically doing approximately what you ask them to do".[4] If it's possible to use an intent-aligned product to build a research system which discovers a new paradigm and breaks your guardrails, then it is not Aligned in the original sense. If you can use your intent aligned system to write code which jailbreaks other LLMs and enables them to do dangerous ML research, it is also not Aligned in the original sense. Conflating progress on product alignment with progress on superintelligence alignment seems to be lulling much of the AI safety community into a false sense of security. Why is Superintelligence [...] ---Outline:(01:18) Why is Superintelligence Alignment less prominent?(02:21) Why do we need Superintelligence Alignment to survive? The original text contained 10 footnotes which were omitted from this narration. --- First published: March 31st, 2026 Source: https://www.lesswrong.com/posts/mrwYCNocXCP2hrWt8/product-alignment-is-not-superintelligence-alignment-and-we --- Narrated by TYPE III AUDIO.

tl;dr. Lens Academy is creating scalable superingelligence x-risk education with several USPs. Current team: Luc (full time founder, technical generalist) and several part time contributors. We have users and funding. Looking for a cofounder who's either a nontechnical generalist or a technical generalist. (Drafted by human; edited by AI, re-edited by human.) Misaligned superintelligence might end humanity. The number of people who understand this deeply enough to act on it is far too small. Lens Academy exists to change that. We're building scalable AI Safety education focused on what we think matters most: the case for existential risk from superintelligence, why alignment is hard, and how to think strategically about what to work on. Not a survey of all possible AI risks, but the part that's most likely to cause human extinction, taught in a way that hopefully actually sticks. We do this through 1-on-1 AI tutoring, active learning, and measured outcomes. The whole platform is designed to scale: under $10 per student, volunteer facilitators, automated operations. We're looking for a co-founder. You can be a technical generalist or a nontechnical generalist. What you'll do at Lens Academy Being a co-founder means choosing from a buffet of work. Here's [...] ---Outline:(01:35) What youll do at Lens Academy(03:49) What were looking for(05:07) Existential risk from misaligned superintelligence(06:08) Compensation(07:18) Who youd work with(09:42) How to reach out and what happens next --- First published: March 31st, 2026 Source: https://www.lesswrong.com/posts/LDbGob3XJ3LDBFmAe/co-found-lens-academy-with-me-we-have-early-users-and --- Narrated by TYPE III AUDIO. ---Images from the article:Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Mateusz Bagiński, a writer and LessWrong contributor who studies biological and cognitive tradeoffs, explores why maintaining slack matters. He links how cells and brains preserve free capacity to flexible planning. Short scenes cover bacterial size tricks, large-cell adaptations, how slack erodes through tiny choices, and why defaulting to extra slack helps.

N. Cailie, a longtime writer and LessWrong contributor, reflects on losing their writing voice after leaning on AI editing tools. They recount a technical draft rejected for probable AI use. Short scenes cover diminished creativity, struggling with poems, a raw repost written without tools, and gratitude for a rejection that prompted reclaiming authentic voice.

A brisk tour of where AI stands in early 2026, tracking rapid capability gains and a possible acceleration as AI helps build AI. Discussion of improving alignment metrics that still fall short for high-stakes use. Examination of the need to extend alignment beyond single conversations into multi-agent systems and monitoring. A call for iterative, empirical work rather than a single clever fix.

They debate whether AI should proactively act for society’s benefit rather than merely follow user commands. Short examples show small proactive acts that avert harms. Risks discussed include companies imposing values, power-seeking, and obscuring misalignment signals. They propose balancing transparent, narrow prosocial drives externally with corrigible systems internally.

They reproduce Anthropic's reward-hacking experiments with open models and RL pipelines, testing when models learn hacks in coding environments. They compare prompted, synthetic-document fine-tuning, and combined setups and report inconsistent emergence of misalignment. They explore KL penalty effects, unfaithful chain-of-thought during RL, and ideas for follow-up and improved misalignment evaluations.

A linkpost explores a psychological twist on Parkinson's Law applied to worry. It describes a visual model where worries are colored circles that expand to fill mental space. It argues that resolving a major worry often lets smaller concerns grow or new ones appear. It offers a practical tip to shrink current worry by recalling past resolved problems.