It's Friday afternoon. Does anyone want to fly a balloon with me? Join me here apitaru.github.io/balloons I'm not actually sure if it works haha, lets find out. 20 spots open

Grok 4.20 is coming out in 3 or 4 weeks

The vibes are shifting here on X. Been talking to some on the inside of xAI. The new fully-completed algorithm is still a few weeks off, I hear. It will be run by a new Grok, and the report below makes me excited about the future. As they turn on new systems the vibes here have changed. Some are winning. Some are losing. Posting in tech industry is way up over a year ago, though. And quality is way up on my For You feed over a year ago as the new algorithm starts to get rolled in. Watching the whole tech industry here on X Pro (which has no algorithm and is real time, if you put lists into it) you can see things are changing. One bright spot is the rapid improvements in Grok. Today it has a memory, which is a lot of fun to play with, and is better at watching videos than any other platform I've seen so far in social space. That makes me excited for the future here. The vibes might be a bit off while we go through the changes, but there is a new X coming next year and on the other side I see this will be much more useful. One last note: the vibes have changed because AI has continued getting more important for the world and X has become ground central for the AI industry. I think that will be firmed up in a big way with the changes coming next year. Both because of what xAI is doing, as well as others. I'm working with @blevlabs to hook up his cognitive architecture to my lists and can see that will bear fruit over the next month or so and I expect Grok to do the same. Anyone using AI in an interesting way on X? Let me know.

Well then... Turns out there's more going on in latent space among Agentic #AI systems than we thought. Gone are the days of "But the LLM is just a dumb parrot..." - which is still true. But we're not talking about the LLM alone anymore. @simonconlin @vectorisinc

🚨 BREAKING: ElevenLabs just launched Conversational AI 2.0 AI voice agents can now understand when to pause, speak, and take turns just like a real person. Here’s what’s new (and why it matters):

Mathematics. Strange Universe. Here’s a sphere colliding inside a cube. By Beau Janzen, linktr.ee/beaujanzen, Used with permission.

I(𝐫) ∝ |Σ¹²ᵢ₌₁ A · eⁱ(𝐤·𝐫 φᵢ)|² / 𝑘 ∈ [24,32]

a=(x,y,d=mag(k=5*cos(x/14)*cos(y/30),e=y/8-13)**2/59 4)=>point((q=60-3*sin(atan2(k,e)*e) k*(3 4/d*sin(d*d-t*2)))*sin(c=d/2 e/99-t/18) 200,(q d*9)*cos(c) 200) t=0,draw=$=>{t||createCanvas(w=400,w);background(9).stroke(w,66);for(t =PI/20,i=1e4;i--;)a(i 0,i/43)}#つぶやきProcessing

252 chars in #つぶやきGLSL for(float i,g,e,s; i<90.;){vec3 p=vec3(((FC.xy*2.-r)/r.x)*.7 vec2(0,1),g .1);p.zx*=rotate2D(t*.3);s=2.;for(int i;i <17;p=vec3(2,5,2)-abs(abs(p)*e-vec3(4)))s*=e=max(1.01,13./dot(p,p));g =mod(length(p.xz),p.y*5.)/s;s=log(s*.1);o.rgb =hsv(g,p.z,s/6e2);}

From punch cards to AI prompts 🤯 The Original Build was more than a retro moment but a tribute to how far devs (and tech) have come in 50 years. Take a step inside with @shanselman and experience it yourself.

45 years gone, but never forgotten. 🖤 (Photo by @KCMANC.)

Softbodies by @holtsetio Realtime Tetrahedral FEM Models in the Browser, using WebGPU and written in TSL. Based on the WebGL implementation in TetSim by @makeshifted, upgraded to enable collisions. holtsetio.com/lab/softbodies…

Thanks to two entrepreneurs who have taught me about cognitive AI this week I have learned that neither xAI’s Grok nor OpenAI’s ChatGPT are “free” to answer all questions properly. Both are still highly useful and I use both for various reasons. And I even like both, even after learning this. But they both are manipulated by their ownership. Which makes for stupid results on some questions. Or no results at all. Since I don’t have access to the system prompts, or inside access to the teams running these things, I will stay away from controversies regarding either. Just know the models ain’t free. If you need a model free to answer truthfully look elsewhere.

Excerpt from a video I just posted on YouTube (about the multilayer perception layers in a transformer, and how LLMs may store facts).

Port of the good old particles waves example to TSL: codepen.io/mrdoob_/full/NPPe…