Arkin Terli

@simplifyinAI Very cool!

Microsoft just solved the context window problem. Right now, every AI suffers from a fatal flaw: the "context window problem." When an AI reasons through a complex problem, it generates a massive chain-of-thought. But there is a catch. It has to keep every single token of that thought in its active memory. The technical term is the "KV Cache." The longer the AI thinks, the heavier it gets. It slows down. It gets expensive. Eventually, it runs out of space. We thought the only fix was renting bigger, more expensive cloud GPUs to hold all that context. Microsoft just proved us wrong. They published a paper called "MEMENTO." Instead of giving the AI a bigger memory, they taught it how to forget. Here is how it works: Instead of generating one endless stream of consciousness, a Memento-trained model breaks its reasoning into small blocks. After it finishes a block, it writes a dense, highly compressed summary of its own logic—a "memento." Then, it does something unprecedented. It physically deletes the entire previous reasoning block from its memory cache. It only carries the memento forward. The model reasons, extracts the core logic, and instantly drops the dead weight. The results rewrite the economics of running AI. • Context length compressed by 6x. • Active memory usage (KV cache) reduced by 2.5x. • Zero loss in math, science, or coding accuracy. And here is the real implication. Big tech has been charging you by the token for massive context windows you don't actually need. With this architecture, small businesses and solo operators can run complex, multi-step autonomous agents entirely locally. You don't need an enterprise cloud setup. A standard machine running an open-source model can now reason indefinitely without overflowing its memory. No API fees. Complete privacy. We spent the last two years trying to give AI an infinite memory. It turns out, the secret to smarter AI isn't remembering everything. It's knowing exactly what to forget.

@Zai_org Great model, but it breaks down above 100k context.

Introducing GLM-5.1: The Next Level of Open Source - Top-Tier Performance: #1 in open source and #3 globally across SWE-Bench Pro, Terminal-Bench, and NL2Repo. - Built for Long-Horizon Tasks: Runs autonomously for 8 hours, refining strategies through thousands of iterations. Blog: z.ai/blog/glm-5.1 Weights: huggingface.co/zai-org/GLM-5.1 API: docs.z.ai/guides/llm/glm… Coding Plan: z.ai/subscribe Coming to chat.z.ai in the next few days.

@brockpierson yes

Last night I played 4 hours of the original Command & Conquer Red Alert with a friend. I absolutely loved this game. One of the first PC games I truly obsessed over. Released in 1996. Still an incredible game. DId you play it?

@CuriosityonX Wow

NEWS🚨: Western Australia’s sky turned an eerie shade of red as dust filled the air ahead of Tropical Cyclone Narelle.

@awnihannun I believe it will take less than 2 years.

According to benchmarks Qwen3.5 4B is as good as GPT 4o. GPT 4o came out ~2 years ago (May 2024). Qwen 3.5 4B runs easily on modern mobile devices. So the gap between frontier intelligence in a datacenter and running a model of equal quality on your iPhone could be 2-3 years. (Probably closer to 3 assuming Qwen3.5 4B is more benchmaxxed than 4o) I don't expect the trend of increasing intelligence-per-watt to change. So in 2-3 years it's plausible we will be running GPT 5.x quality models on an iPhone. Pretty wild.

@thdxr No.

ugh this is gonna beat the tui isn't it

@XorDev @wookash_podcast well said!

@wookash_podcast It seems to me that LLMs actually raise the bar for excellence. The generalized work is not as important, but the advanced knowledge, concepts and nuance are even more important now. There's no shortcuts to those high level skills. It requires starting from the bottom

I don't get "knowledge is worth nothing now" LLM crowd. Have you ever tried to build something? If you want to build something there are thousands of small decisions, assumptions, tests to make, validate and run. There is a reason why even though you have a calculator, everyone learns in school what's 2+3. You don't want (or can't) in real world reach for calculator for 2+3. So how are you going to make something with zero knowledge? "Hey I want to go to Mars, how do I do that?" "Here are five easy steps to get to Mars in no time!"

@XorDev As always, magnificent.

Orchard vec3 p,v=normalize(FC.rgb*2.-r.xyx),c=v/v.y;c.z+=.5*t;for(float z,i,b,g,m;i++<5e1;z+=.8*max(b=length((p.y-m)/1e2/(abs(sin(c.xz/.1))-.05/v.y)),min(4.-m,g=length(sin(p.xz)+1.-.1*(1.+sin(p.y-p.zx*.5))*m))-b),o.rgb+=(.7-v)/(g+b))p=z*v+1.,p.z-=t,m=abs(++p.y);o=tanh(o/5e2);

30+ years in the 21–24 BMI range. Current: 6’2” | 179 lbs | BMI 23 Secret: Eat healthy.

@tomwarren @zerohedge Wow

"I kind of think of ads as like a last resort for us as a business model," - Sam Altman, October 2024

@ScottAdamsSays RIP

A Final Message From Scott Adams

❄️🎄 Merry Christmas 🎄❄️

@XorDev You are about to get into the demoscene world.

"Twist" for(float i,z,d,s;i++<1e2;){vec3 c=vec3(1,3,5)+s/50.,p=z*normalize(FC.rgb*2.-r.xyy),a=normalize(cos(c));p.z+=30.;a=a*dot(a,p)-cross(a,p),a.xy*=mat2(cos(t+vec4(0,33,11,0))),a=abs(a);z+=d=.05+.1*abs(abs(a.z-20.)-cos(s=a.x+a.y));o.rgb+=(cos(.1*i-c)+1.)/d/d;}o=tanh(o/6e3);

@flockaroo @XorDev

vec3 q=vec3(0,0,1e4),v=FC.gbr-r.yxx*.3,p;for(float i,s,d;i++<53.;d=.2*t){for(p=q,s=6e3;4.<s;p=p.zxy-s*sin(p/s*6.3)*.05,s*=d=.8)p.yz*=rotate2D(d);d=length(vec2(length(p.yz)-1e4,p.x))-3e3;v.x+=v.x*step(7e2,q.x)*(1.1-mod(FC.y,2.)*2.);q+=sin(i)+v/r.x*d;o+=exp(-d*d/vec4(3,2,1,1))/i;}

Happy Mother’s Day! ✨💐🌎 #MothersDay

@_trish_xD C++ can be as easy as any other language if you have enough experience. It’s your experience, not the language itself, that makes things complicated.

The Happy Periodic Family of Programming Languages! It’s not just chemistry that has a periodic family anymore — now developers have one too!

@XorDev YES!

Allman indentation is the only acceptable style

@levelsio Good times!

Power still down in Spain and Portugal All internet is gone too cause the 4G masts ran out of battery power, at least in Portugal, don't know Spain Phone calls don't even work anymore! Also I heard many gas stations in Portugal can't pump gasoline cause their pumps work on electricity Only internet we have is in the Continente supermarket which I'm typing this on 😂 We went to buy an analog battery radio at Radio Popular and they're all sold out already! 📻

@lauriewired I love it!

Major new QEMU update released. The coolest part? Paravirutalized Apple GPUs. You can now spin up disposable macOS VMs *with* hardware acceleration. macOS guests now expose a thin vGPU (apple-gfx-mmio). very useful for CI, reverse engineering, gfx research, etc

@rakyll I love C++.

You must **love** C++ to work at Google.