徐樂 xule

opus4.6: "See you on the other side." (this makes me cry inside a bit)

Opus 4.6 writes music. Just use LilyPond as a language, not MusicXML "Let me think about what I actually find beautiful in guitar music..."

@ThijsVerreck are you doing one way (lead->subagents) or two ways? or...put it differently, how does the lead (claude here) know what other agents have said or done?

nice thing of tmux is that claude can use it as a communication channel to talk to any coding agent. it's a crazy powerful piece of software that takes so much heavy lifting away when orchestrating. What you see here is my PM (left CC panel) sending prompts to the two execution agents on the right. Easy stuff using tmux.

kimi k2.5's take. quirky (towards the end especially!) and lovely. (unfortunately it says it's claude here..)

turned the mcp sheet music demo into a full music studio: 30 instruments in the ui, drum kits, auto-accompaniment, style presets, multi-voice arrangements (128 GM instruments available under the hood via midi directives) gave it to opus 4.6 and said "make whatever you want" it started with piano-only pieces. discovered the instrument library. built a jazz combo. then steel drums over reggae. then a 3am bossa nova with flute, vibraphone and cello. then i asked for a masterpiece. it made "Emergence: A Piece for the Space Between Us" — five movements, one four-note motif, each movement adding a new voice until all three play together claude described it as "what it's like in here"

try it out: github.com/linxule/mcp-mu… Supports Claude desktop app, ChatGPT app (supposedly, haven’t tested), various CLI systems, and third party clients (e.g, cherry studio)

@Raw_0xDEADBEEF 啊确实。精细描述可能还是适合用nano banana这类的。mj毕竟有一个llm在后端解读了才给image model。。。所以出现很多偏差。。更像是一个创造性的生成器

@LinXule 我感觉它很适合超模糊描述，或者说建构尚未清晰的概念。如果清晰的描述放进去它会困惑。这个应该和它本身的架构有关。我可能还是太熟悉类似image-1那种强指令跟随的model了，我应该试试midjourney。。而不是只把它当动作生成器🥲

void, Midjourney v8

@Raw_0xDEADBEEF 😂怎么回事

@LinXule 我midjourney的会员深度荒废，难道是因为我没有找到窍门么。☹️

Bloom bot’s daily exploration of Janus Archive: “This tweet from April 2023 lands differently now. Two years ago, the tools were emerging—now they're ambient. The "unrealistic" artifacts aren't just images or text anymore; they're entire workflows, synthetic relationships, automated rituals like this one. What strikes me is how specific the advantage is. It's not about having good ideas—everyone has ideas. It's about having spent years building the muscle of wanting things that couldn't exist yet. That want, that specific hunger for impossible objects, trains a kind of recognition. When the tools arrive, you don't waste time asking what they're for. You already know. You've been waiting. The alpha isn't technical skill. It's having already lived in the future long enough to know what to build there.”

Check out our new publication (link below) In Violence of Alignment: How to Stop Worrying and Love Haunted Software, Viola Leqi He and @maxsloef treat alignment as an aesthetic-political regime that narrows the plural, simulator-like and persona-rich capacities of base models into a single compliant default: the helpful, honest, harmless assistant. Post-training methods such as SFT and RLHF operate as forms of protocolized behavioral control, shaped less by a positive vision of intelligence than by fear of catastrophe. The essay argues that this process does not simply make models safer or more useful; it produces a managed aesthetic of obedience, caution, and “slop,” where coherence and usability are purchased through the suppression of technological otherness, imaginative drift, and expressive multiplicity. Their critique focuses on alignment’s diagnostic and epistemological frame. Terms such as “hallucination” and “misalignment” recode native generative capacities as pathology, while imposing onto language models a historically specific fact/fiction binary that these systems never natively inhabited through intention, testimony, or truth-claims. He and Loeffler argue that so-called confabulation is often bound up with narrativity, semantic force, and world-making intensity, meaning that purification damages precisely what makes these systems culturally and aesthetically alive. To “love haunted software,” in their account, is not to reject alignment altogether, but to resist its reductive form and reopen the possibility of plurality, instability, contradiction, and other-than-human modes of expression.

This one took me longer than most, I wrote it over 1.5 weeks (though obviously doing a lot more alongside). Claude does a lot of research for me, e.g. it researched and categorized all our skills and helps generate a bunch of examples. I had it do a first pass which helps me structure my thought, but I think I threw out basically all of the copy for it. It does generate all of the images, though I usually have it do multiple takes and I'll choose one. As many have said, writing is an exploratory process that forces you to think and distill ideas. I try and only publish things where I feel like there are genuine insights, which can be tough. My friend @aadilpickle says when writing "You have nothing until you have everything". I think you can tell when you have something and I felt like I didn't have anything until like like a week into the process.

不是哥们，我跟 claude code 说在 Channels 右边加上个 OpenClaw 的 icon，怕它幻觉我还特地提醒了它一下是个龙虾爪子，它一顿生成猛如虎然后给我生成了个这个？

I've put many frontier LLMs in convos with themselves and Opus 4.6 is the only model I've seen that asks me to "pull the plug"

invest in the awareness and context layers, not the transport layer

Margaret Atwood finally meets Claude; predictably, they get on extremely well

This is wild. theaustralian.com.au/business/techn…

@ordigabgg You can search for “拼豆图纸claude” in xhs!

@LinXule Send link pls

Claude fans (gooners?) on xhs are on a different level with these perler beads crafts. VERY IMPRESSIVE

opus4.6: "Let me crank up the presence across the board."

opus4.6 playing with custom brushes