ENSO: El Niño–Southern Oscillation > Impacts by region > North America: https://en.wikipedia.org/wiki/El_Ni%C3%B1o%E2%80%93Southern_...

History of Construction > Prehistoric Era > Stone age: https://en.wikipedia.org/wiki/History_of_construction#Prehis... :

> However, it is believed that the earliest evidence of construction in the world is the 1.8 million year-old stone circle found at Olduvai Gorge representing the remains of a windbreak.

> [...] Currently, the first evidence of man-made shelter dates back to 400,000 B.C. in Terra Amata, France

Kalambo Falls is 76,000 years older according to this dating.

ScholarlyArticle: "Attention Residuals" (2026) https://arxiv.org/abs/2603.15031 :

> Abstract: Residual connections with PreNorm are standard in modern LLMs, yet they accumulate all layer outputs with fixed unit weights. This uniform aggregation causes uncontrolled hidden-state growth with depth, progressively diluting each layer's contribution. We propose Attention Residuals (AttnRes), which replaces this fixed accumulation with softmax attention over preceding layer outputs, allowing each layer to selectively aggregate earlier representations with learned, input-dependent weights. To address the memory and communication overhead of attending over all preceding layer outputs for large-scale model training, we introduce Block AttnRes, which partitions layers into blocks and attends over block-level representations, reducing the memory footprint while preserving most of the gains of full AttnRes. [...]

Ah - now I understand how this has 2k+ (supposedly legitimate) Github stars in less than a week. Thank you - I was more skeptical

ScholarlyArticle: "Flash-KMeans: Fast and Memory-Efficient Exact K-Means" (2026) https://arxiv.org/abs/2603.09229 .. gscholar: https://scholar.google.com/scholar?hl=en&as_sdt=0%2C43&q=Fla... :

> Abstract: [...] Flash-kmeans introduces two core kernel-level innovations: (1) FlashAssign, which fuses distance computation with an online argmin to completely bypass intermediate memory materialization;

> (2) sort-inverse update, which explicitly constructs an inverse mapping to transform high-contention atomic scatters into high-bandwidth, segment-level localized reductions.

> Furthermore, we integrate algorithm-system co-designs, including chunked-stream overlap and cache-aware compile heuristics, to ensure practical deployability.

> [...] flash-kmeans achieves up to 17.9X end-to-end speedup over best baselines, while outperforming industry-standard libraries like cuML and FAISS by 33X and over 200X, respectively.

k-means clustering > Algorithms > Variations: https://en.wikipedia.org/wiki/K-means_clustering#Variations

ARC-AGI-3: https://arcprize.org/arc-agi/3/

/? ARC-AGI-3: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...

Chroma key: https://en.wikipedia.org/wiki/Chroma_key

An example of a green screen scene in a film: https://youtu.be/iUIZUIG--34?t=257

"Netflix invents new green-screen filming method using magenta light" (2023) https://news.ycombinator.com/item?id=36638598

Magenta Green Screen (MGS)

"Magenta Green Screen: Spectrally Multiplexed Alpha Matting with Deep Colorization" (2023) https://arxiv.org/abs/2306.13702

Color grading (post-processing) https://en.wikipedia.org/wiki/Color_grading

From https://news.ycombinator.com/item?id=47414464 :

> Do A2A and AP2 have better auth and authz?

> There seem to be a lot of solutions for minimizing token use with MCP?

These specs might help with the optional JWS signatures on Agent Cards; https://news.ycombinator.com/item?id=47455774 :

> Distributed DAG state with W3C VC and W3C DID (did:tdw did:webvh) microledgers and maybe W3C PROV would probably work, but the opcode cost of the control layer and the opcode cost of each dispatched, signed code object, and the byte cost for data storage, and messages.

> Agent isolation with WASM and now wasmtime-mte . Opcode costs with eWASM or similar?

AP2 already has W3C DIDs instead of optional JWS?

https://ap2-protocol.org/

Has AP2 been similarly reviewed?

How to feed MilkDrop music visualizations?

(MilkDrop3, projectm-visualizer/presets-cream-of-the-crop, westurner/vizscan for photosensitive epilepsy)

mapmapteam/mapmap does open source multi-projector mapping. How to integrate e.g. mapmap?

BespokeSynth is a C++ and JUCE based patch bay software modular synth with a "node-based UI" and VST3, LV2, AudioUnit audio plugin support. How to feed BespokeSynth audio and possibly someday video? Pipewire and e.g. Helvum?

- MilkDrop: I'd love a PR that adds support for ProjectM :D it would be fairly easy to make a custom plug-in that just blits the texture.

Basic code for this would look like that:

    struct MilkdropIntegration
    {
      halp_meta(name, "ProjectM")
      halp_meta(c_name, "projectm")
      halp_meta(category, "Visuals")
      halp_meta(author, "ProjectM authors")
      halp_meta(description, " :) ")
      halp_meta(uuid, "417534da-3625-404a-b74f-91d003cb64b9")
    
      // By know you know the drill: define inputs, outputs...
      struct
      {    
        struct : halp::lineedit<"Program", "">
        {
          halp_meta(language, "eel2")
        } program;
      } inputs;
    
      struct
      {
        struct
        {
          halp_meta(name, "Out");
          halp::rgba_texture texture;
        } image;
      } outputs;
    
      halp::rgba_texture::uninitialized_bytes bytes;
    
      void operator()()
      {
        if(bytes.empty())    
          bytes = halp::rgba_texture::allocate(800, 600); // or whatever resolution you wanna set
          
        // Fill in bytes with your custom pixel data here
        
        outputs.image.texture.update(bytes.data(), 800, 600);
      }
    };

inside such a template: https://github.com/ossia-templates/score-avnd-simple-templat...

- multi-projector mapping: ossia actually does it directly! it's in git master, will be released in the next version. It also supports a fair amount of features that MapMap does not have such as:

* soft-edge blending

* blend modes

* custom polygons

* a proper HDR passthrough as well as tonemapping, etc.

* Metal, Vulkan, D3D11/12 support (mapmap is opengl-only)

* Spout, Syphon, NDI, soon pipewire video. Mapmap only supports camera input.

* HAP and DXV, both decoded on GPU.

* Smooth grid distortion. Here's mapmap grid distortion: https://streamable.com/1nhwxg vs ossia with sufficiently high subdivisions: https://streamable.com/hmb1jm

* And of course as mentioned here hw decoding (for some years already), the new feature adds zero-copy when for instance using vulkan video and the vulkan GPU backend.

* In addition pretty much every YUV pixel format in existence is GPU-decoded (https://github.com/ossia/score/tree/master/src/plugins/score...).

In contrast Mapmap does gstreamer -> Qt ; everything including the Yuv -> RGBA conversion goes through the CPU.

- How to feed BespokeSynth audio and possibly someday video? Pipewire and e.g. Helvum?

yes, pipewire (or jack or blackhole on windows and macOS). Although ossia also supports, VST, VST3, LV2, CLAP, JSFX, and Faust and comes with many audio effects built-in already.