I love AI, programming, open source, video games, economics, motorcycles and continuous self improvement. Staff Engineer @ MCP Manager by Usercentrics

Baltimore, Maryland
Joined November 2012
Hit the dance floor and learn how LLMs work in this 3D tour of Qwen3.8 27B, built on real numbers recorded on my PC. 🎧 The record is the model, and every token rides its groove through all 64 layers. Nine minutes, fully narrated, no math required. youtu.be/fgMa2gdPuKs
1
36
Go deeper: the full interactive version, live in your browser (GPU required) 🎧 A narrator, detailed captions or both, plus an FAQ section for the "wait, why?" moments, like how a 27B model can outscore ones with far more weights. sambatista.com/blog/hit-reco…
13
Samuel Batista retweeted
185
3,259
235
19,679
576,404
Next LLM architecture visualization is going to take you into a Dance Club with Qwen3.8 27B. It's almost done!
1
20
Samuel Batista retweeted
33
240
7
2,427
282,533
Can you find the easter egg on this blog post? sambatista.com/blog/lightnin… It lets you play around with the GLM 5.3 avatar editor I used while creating the LLM architecture presentation.
41
Samuel Batista retweeted
You are told that the market efficiently prices assets. It takes Amazon about 6 days to earn the amount of revenue SpaceX earned in the past 12 months. And they both trade at about the same market cap. Do you believe this is efficient pricing?
99
35
6
1,118
69,764
It's a flipping car!
Dude. Elon Musk is on another planet.
67
Samuel Batista retweeted
I asked Opus what it thinks 2027 will look like. It made this amazing video.
98
258
44
2,019
110,914
I love this take, it resonates with me.
Alexandr Wang on what matters in an age of abundant intelligence: “The scarce resource isn’t going to be intelligence or agency. I really think it’s going to be vision and ambition.” “Do you have a clear view of what you want the world to look like in the future? What is the one way in which you wanna put your finger on the scale for how the future of the world will develop?” “AI will make that easier... But the flip side of that is then all of a sudden you can dream bigger.”
39
Say what you want about Mamdani, sometimes he's spot on. And he's spot on here.
🚨 At the U20 Mayors Summit, Mamdani Says the Left Has Ceded “Quality of Life,” “Efficiency,” and “Fraud and Waste” to the Right This was one of the more interesting arguments Zohran Mamdani made today at the U20 Mayors Summit, and a reminder that he can be genuinely sharp when he drops the “central casting” routine. “In New York City, there are certain words that have taken on ideological associations.” “Quality of life has come to be seen as somehow a right-wing concern. Efficiency… has been taken on as a right-wing concern, rooting out fraud and waste, a right-wing concern.” Mamdani’s argument is that refusing to engage those issues actually helps the Right. “If you are staring at an example of social disorder or social inefficiency, and the left refuses to explain either why it has happened or how they will fix it, then the right gets to point at someone in the immediate area and say, ‘That’s who did it.’” The harder question is how much of the broader Left is actually prepared to follow the advice.
34
Skip the diagrams. This is what a 743B-parameter model looks like when it's working. ⚡ Every spark is the model choosing what to read. Eight minutes, fully narrated, no math required. youtu.be/p9HiHYIWTho
1
18
Go deeper: the full interactive version, live in your browser (GPU required) ⚡ Detailed captions, a study mode to go at your own pace, and an FAQ section for the "wait, why?" moments. sambatista.com/blog/lightnin…
15
Ever wonder how an LLM guesses what to say when you ask it a question? If you have interest and 10 minutes, dive into GLM-5.3's architecture with this engaging 3D visualization into how a large LLM works. Watch it in 4K on YouTube: youtu.be/p9HiHYIWTho Or run it live in your browser (GPU required): sambatista.com/blog/lightnin… I followed a conversation through Z.ais GLM-5.3, an open-weight AI model with 743 billion parameters, and built its architecture as a lightning machine. Each of its 78 layers is a tower, and every time the model picks something from the conversation to read, a spark strikes it. This is the first one with a narrator, so you can sit back and watch. Enjoy!
1
69
Follow a conversation through GLM-5.3, a 743 billion parameter AI. ⚡ This 3D animation was built with several LLMs and runs on your browser (link in comment). It's designed to teach and entertain and help build accurate mental models of how LLMs work during inference. Enjoy!
1
1
98
I'm cooking with AIs. Another 3D animation on LLM architecture is coming. Anyone want to guess what model I'm showcasing next?
1
1
34
I've always loved this interview with Steve Jobs. He as a special human.
Jobs called it a disease: thinking the idea is 90% of the work, then handing it to other people. Jobs says no. Electrons will not do what you want. Plastic will not. Glass will not. You are holding 5,000 things that do not fit. His picture for a real team is an old man, a coffee can, and ugly rocks. Motor on. Noise all night. Next day the same stones are polished. The polish was them hitting each other. The idea that went in is not the one that came out.
26
Samuel Batista retweeted
1. Oh shit, AI can do my job better than me 2. Software engineering is dead! 3. I can vibe code as much slop as I want! 4. Oh, things are not going well. Fable/Sol/Opus all suuuck 5. But what if I constrain the agent? Put it in a straitjacket? 6. Oh, the results are pretty good now. How do I go deeper? 7. *Orders 12 software engineering books from Amazon* 8. Software engineering is alive!
325
470
81
7,566
699,555