Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Yo LLM bros and sisters, do check out this amazing visual guide this guy just created... Starts from the basics of Attention and the GPT-2 architecture, and introduces all the concepts required to reach Kimi-K3. Very nice!

AVB

13,358 subscribers

22,482 views • 20 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

4 Comments

Ujjwal20 days ago

thank u sir!🛐 getting retweeted by prodigy❤️

AVB20 days ago

The hard work shows man! Keep making these :)

RS20 days ago

Thanks a lot for this

Daniel Varela20 days ago

so cool!

Related Videos

This AI analyst does $60,000 a day of research for about $450 Kimi K3 + GPT-6 Astra, but the real trick is a funnel, not one giant model: signal → plain code → Kimi K3 → GPT-6 Astra → human → sourced report step 1 → plain code does the math first, zero LLM cost step 2 → rules and a light classifier cut 10,000 stocks to 150 step 3 → Kimi K3 reads eight quarters at once and keeps the 40 that don't add up step 4 → GPT-6 Astra argues bull vs bear, every claim needs a source step 5 → a human confirms anything that executes the result: no BUY, no predictions, just what changed and what contradicts it save this setup, watch the terminal above - then read the full architecture with formulas ↓

This AI analyst does $60,000 a day of research for about $450 Kimi K3 + GPT-6 Astra, but the real trick is a funnel, not one giant model: signal → plain code → Kimi K3 → GPT-6 Astra → human → sourced report step 1 → plain code does the math first, zero LLM cost step 2 → rules and a light classifier cut 10,000 stocks to 150 step 3 → Kimi K3 reads eight quarters at once and keeps the 40 that don't add up step 4 → GPT-6 Astra argues bull vs bear, every claim needs a source step 5 → a human confirms anything that executes the result: no BUY, no predictions, just what changed and what contradicts it save this setup, watch the terminal above - then read the full architecture with formulas ↓

leopardracer

42,730 views • 5 days ago

LLM Visualization This is actually pretty amazing! It helps to visualize the core components of LLMs like nano-gpt and GPT-3.

LLM Visualization This is actually pretty amazing! It helps to visualize the core components of LLMs like nano-gpt and GPT-3.

elvis

39,649 views • 2 years ago

$Kimi K3 vs GPT-5.6 Sol — same game build, a fraction of the cost. Kimi K3 just 3-shotted a Fallout Vault-Tec FPS clone. $2.48 in API usage. The same token count runs $5.34 on GPT-5.6 Sol. a full playable FPS for the price of a coffee. K3 dropped this week: 2.8 trillion parameters, open weights July 27. I broke down exactly how to build with it — full Kimi K3 guide below.$

Kimi K3 vs GPT-5.6 Sol — same game build, a fraction of the cost. Kimi K3 just 3-shotted a Fallout Vault-Tec FPS clone. $2.48 in API usage. The same token count runs $5.34 on GPT-5.6 Sol. a full playable FPS for the price of a coffee. K3 dropped this week: 2.8 trillion parameters, open weights July 27. I broke down exactly how to build with it — full Kimi K3 guide below.

Kirill

27,900 views • 2 months ago

A Stanford professor just gave a public lecture on exactly how GPT, Claude, and LLaMA are built under the hood no insider access required just the clearest breakdown of modern LLM architecture I've seen this lecture reveals the framework professors are paid up to $750K a year to teach the gap between "I use ChatGPT" and "I understand how it works" is smaller than most people think the most complete public breakdown of modern LLM architecture I've seen this year

A Stanford professor just gave a public lecture on exactly how GPT, Claude, and LLaMA are built under the hood no insider access required just the clearest breakdown of modern LLM architecture I've seen this lecture reveals the framework professors are paid up to $750K a year to teach the gap between "I use ChatGPT" and "I understand how it works" is smaller than most people think the most complete public breakdown of modern LLM architecture I've seen this year

rewind

165,059 views • 3 months ago

this guy just wired kimi k3 directly into codex. the model is insanely good, but the app isn’t. so instead of using the kimi app, he connected his kimi subscription to codex through oauth. now kimi k3 shows up right inside the model picker alongside sol 5.6. no api key needed. no extra subscription either. just sign in with oauth through kimi code, and it’s ready to use. simple guides in the video and links below 👇

this guy just wired kimi k3 directly into codex. the model is insanely good, but the app isn’t. so instead of using the kimi app, he connected his kimi subscription to codex through oauth. now kimi k3 shows up right inside the model picker alongside sol 5.6. no api key needed. no extra subscription either. just sign in with oauth through kimi code, and it’s ready to use. simple guides in the video and links below 👇

MIKE

58,719 views • 2 months ago

Next up on First Pass: A conversation with Simon Mo from Inferact on Kimi K3. We discuss: - The basics on the K3 model (it's huge!) - Why open weight models are important - The engineering complexity of working with a model this large - Innovative licensing models And more!

Next up on First Pass: A conversation with Simon Mo from Inferact on Kimi K3. We discuss: - The basics on the K3 model (it's huge!) - Why open weight models are important - The engineering complexity of working with a model this large - Innovative licensing models And more!

Jamin Ball

13,707 views • 1 month ago

Kimi K3 vs GPT-5.6 Sol. I reused the 3D globe test from Kimi vs Fable 5. Same prompt. Same reference image. Kimi K3 left. GPT-5.6 Sol right. Kimi edges out Sol on UI fidelity and globe realism. At current API prices, Kimi is about 50% cheaper. For frontend work, it's my pick.

Kimi K3 vs GPT-5.6 Sol. I reused the 3D globe test from Kimi vs Fable 5. Same prompt. Same reference image. Kimi K3 left. GPT-5.6 Sol right. Kimi edges out Sol on UI fidelity and globe realism. At current API prices, Kimi is about 50% cheaper. For frontend work, it's my pick.

Kai

15,769 views • 2 months ago

Tried recreating Fall Guys using the exact same prompt with Claude Opus 5.0 (Claude Code) and Kimi K3 (Kimi CLI). Kimi K3 nailed the gameplay out of the box. Physics, movement, and the overall feel were surprisingly solid Opus 5.0 had a shinier UI, but movement was broken, it kept adding unnecessary visual effects, lagged during gameplay, and spent most of its time reiterating instead of converging Runtime and cost : - Kimi K3: ~9 min, ~$4.4 - Opus 5.0: ~17 min (still iterating), ~$13+ That's roughly 2× the time and 3× the cost, yet Kimi produced the more playable result with the exact same prompt.

Tried recreating Fall Guys using the exact same prompt with Claude Opus 5.0 (Claude Code) and Kimi K3 (Kimi CLI). Kimi K3 nailed the gameplay out of the box. Physics, movement, and the overall feel were surprisingly solid Opus 5.0 had a shinier UI, but movement was broken, it kept adding unnecessary visual effects, lagged during gameplay, and spent most of its time reiterating instead of converging Runtime and cost : - Kimi K3: ~9 min, ~$4.4 - Opus 5.0: ~17 min (still iterating), ~$13+ That's roughly 2× the time and 3× the cost, yet Kimi produced the more playable result with the exact same prompt.

aditya

318,628 views • 2 months ago

SGLang day-0 speed on Kimi K3: 423 tok/s (measured on gsm8k), plus RL support ready in Miles RadixArk! How the largest open-source model runs this fast: we natively implemented and deeply optimized K3’s new architecture with fused KDA decode kernels, DP attention, DSpark, PD disagg, and KDA-aware prefix caching. We've passed Kimi Vendor Verifier and are ready for production! Thanks to Kimi.ai, NVIDIA, AMD, KVCache.AI, Modal, and Baseten for building this with us, and to Google Cloud, Nebius Token Factory, fal, DigitalOcean, Runpod, DeepInfra and GMI Cloud for serving K3 on SGLang. Blog, cookbook, benchmarks in the comments. P.S. This demo video? Kimi K3 made it itself. Play the game 👇

SGLang day-0 speed on Kimi K3: 423 tok/s (measured on gsm8k), plus RL support ready in Miles RadixArk! How the largest open-source model runs this fast: we natively implemented and deeply optimized K3’s new architecture with fused KDA decode kernels, DP attention, DSpark, PD disagg, and KDA-aware prefix caching. We've passed Kimi Vendor Verifier and are ready for production! Thanks to Kimi.ai, NVIDIA, AMD, KVCache.AI, Modal, and Baseten for building this with us, and to Google Cloud, Nebius Token Factory, fal, DigitalOcean, Runpod, DeepInfra and GMI Cloud for serving K3 on SGLang. Blog, cookbook, benchmarks in the comments. P.S. This demo video? Kimi K3 made it itself. Play the game 👇

LMSYS Org

205,325 views • 2 months ago

This is absolutely amazing. The practice and the skill required to do this. They absolutely nailed it.

This is absolutely amazing. The practice and the skill required to do this. They absolutely nailed it.

NancyH

25,162 views • 3 months ago

The family of Mr bangis please reach out to him,he needs you all at this moment and to the German embassy please pay close attention to our brother in your care and to abike dabiri help us reach out to the German embassy on this issue

The family of Mr bangis please reach out to him,he needs you all at this moment and to the German embassy please pay close attention to our brother in your care and to abike dabiri help us reach out to the German embassy on this issue

VERYDARKMAN

47,977 views • 1 year ago

I asked Opus 5.5, GPT-6 Astra, GPT-6 Sol, Kimi K3 to build the Eiffel Tower in Three.js. Opus 5.5: $8.95 | 10 mins GPT-6 Astra: $7.45 | 8 mins GPT-6 Sol: $3.90 | 5 mins Kimi K3: $4.04 | 6 mins Opus 5.5 really impressed me. The frontend and 3D detailing are seriously good, and I preferred it over Astra for this task. GPT-6 Sol was a bit disappointing, while Kimi K3 looks really good for the price. It delivered a strong result at almost half the cost. Kimi K3.1 is going to be a monster for sure. Overall, Anthropic made a great comeback with Opus 5.5.

I asked Opus 5.5, GPT-6 Astra, GPT-6 Sol, Kimi K3 to build the Eiffel Tower in Three.js. Opus 5.5: $8.95 | 10 mins GPT-6 Astra: $7.45 | 8 mins GPT-6 Sol: $3.90 | 5 mins Kimi K3: $4.04 | 6 mins Opus 5.5 really impressed me. The frontend and 3D detailing are seriously good, and I preferred it over Astra for this task. GPT-6 Sol was a bit disappointing, while Kimi K3 looks really good for the price. It delivered a strong result at almost half the cost. Kimi K3.1 is going to be a monster for sure. Overall, Anthropic made a great comeback with Opus 5.5.

Bhavy☄️

162,515 views • 3 days ago

New short course: Attention in Transformers: Concepts and Code in PyTorch. Last week we released a course on how LLM transformers work. This week, go deeper and learn about the technical ideas behind the attention mechanism, and see how to code it in PyTorch. This course is built with Joshua Starmer, Founder and CEO of StatQuest. The attention mechanism was a breakthrough that led to transformers, the architecture powering large language models like ChatGPT. Transformers, introduced in the 2017 paper: "Attention is All You Need" by Viswani and others, took off because of its highly scalable design. In this course, you’ll learn how the attention mechanism, a key element of transformer-based LLMs, works and implement it in PyTorch. You'll develop deep intuition about building reliable, functional, and scalable AI applications. What you will do: - Understand the evolution of the attention mechanism, a key breakthrough that led to transformers. - Learn the relationships between word embeddings, positional embeddings, and attention. - Learn about the Query, Key, and Value matrices, and how to produce and use them in attention. - Walk through the math required to calculate self-attention and masked self-attention to learn why and how they work. - Understand the difference between self-attention and masked self-attention and how one is used in the encoder to build context-aware embeddings and the other is used in the decoder for generative outputs. - Learn the details of the encoder-decoder architecture, cross-attention, and multi-head attention and how they are all incorporated into a transformer. - Use PyTorch to code a class that implements self-attention, masked self-attention, and multi-head attention. There're lots of exciting technical details in this course. Please sign up here:

New short course: Attention in Transformers: Concepts and Code in PyTorch. Last week we released a course on how LLM transformers work. This week, go deeper and learn about the technical ideas behind the attention mechanism, and see how to code it in PyTorch. This course is built with Joshua Starmer, Founder and CEO of StatQuest. The attention mechanism was a breakthrough that led to transformers, the architecture powering large language models like ChatGPT. Transformers, introduced in the 2017 paper: "Attention is All You Need" by Viswani and others, took off because of its highly scalable design. In this course, you’ll learn how the attention mechanism, a key element of transformer-based LLMs, works and implement it in PyTorch. You'll develop deep intuition about building reliable, functional, and scalable AI applications. What you will do: - Understand the evolution of the attention mechanism, a key breakthrough that led to transformers. - Learn the relationships between word embeddings, positional embeddings, and attention. - Learn about the Query, Key, and Value matrices, and how to produce and use them in attention. - Walk through the math required to calculate self-attention and masked self-attention to learn why and how they work. - Understand the difference between self-attention and masked self-attention and how one is used in the encoder to build context-aware embeddings and the other is used in the decoder for generative outputs. - Learn the details of the encoder-decoder architecture, cross-attention, and multi-head attention and how they are all incorporated into a transformer. - Use PyTorch to code a class that implements self-attention, masked self-attention, and multi-head attention. There're lots of exciting technical details in this course. Please sign up here:

Andrew Ng

132,544 views • 1 year ago

GPT-6 Extra High vs. Kimi K3 Swarm - ECHO Start Menu Both models received the exact same prompt. I’m shocked at the quality Kimi K3 was able to output here. Its overall visual quality is above GPT-6’s and even above the menu from the actual game. However, it took significantly longer - around 3.5 hours versus GPT-6’s 42 minutes. Roughly 5× longer. GPT-6’s demo felt more realistic and stayed a lot closer to how the original menu looks. Kimi added something that GPT-6 missed, which was the interactive eye tracking. Kimi’s eyeball follows (albeit blurry and in a slightly weird way) where you click on the start screen. GPT-6’s doesn’t. • GPT-6 Extra High: approximately 46,000 generated output tokens - $2.30 in API-equivalent output cost. • Kimi K3 Swarm: estimated 250,000 generated output tokens - $3.75 in API-equivalent output cost.

GPT-6 Extra High vs. Kimi K3 Swarm - ECHO Start Menu Both models received the exact same prompt. I’m shocked at the quality Kimi K3 was able to output here. Its overall visual quality is above GPT-6’s and even above the menu from the actual game. However, it took significantly longer - around 3.5 hours versus GPT-6’s 42 minutes. Roughly 5× longer. GPT-6’s demo felt more realistic and stayed a lot closer to how the original menu looks. Kimi added something that GPT-6 missed, which was the interactive eye tracking. Kimi’s eyeball follows (albeit blurry and in a slightly weird way) where you click on the start screen. GPT-6’s doesn’t. • GPT-6 Extra High: approximately 46,000 generated output tokens - $2.30 in API-equivalent output cost. • Kimi K3 Swarm: estimated 250,000 generated output tokens - $3.75 in API-equivalent output cost.

Chris

120,153 views • 20 days ago

🚨 Deepseek V4 pro GA is being tested to few users > this is the output on the same prompt that was used for kimi k3 and GPT sol 5.6 in the quoted video > The details are solid , addition of tents , campfire and everything this model looks promising

🚨 Deepseek V4 pro GA is being tested to few users > this is the output on the same prompt that was used for kimi k3 and GPT sol 5.6 in the quoted video > The details are solid , addition of tents , campfire and everything this model looks promising

Chetaslua

172,573 views • 2 months ago

The way this guy paid attention to details is amazing. Amala, gbegs and eweds to the world 🥰🥰

The way this guy paid attention to details is amazing. Amala, gbegs and eweds to the world 🥰🥰

Omotola❤❤❤

23,678 views • 4 months ago

Kimi K3 just beat Fable 5 on the BridgeBench Horror House game test. I did not expect this at all. Kimi K3 is better than Fable 5 at game development and UI design. The two things Fable 5 was supposed to own. The 5x price increase suddenly makes sense.

Kimi K3 just beat Fable 5 on the BridgeBench Horror House game test. I did not expect this at all. Kimi K3 is better than Fable 5 at game development and UI design. The two things Fable 5 was supposed to own. The 5x price increase suddenly makes sense.

BridgeMind

149,482 views • 2 months ago

OUCH: Conservative Dan Albas runs out of patience with Freeland and her lame excuses: "You come to this committee and you talk, talk, talk. You're not willing to do the basics, which is to -let me just finish here - do the basics of standing up for Canadian workers. This was your opportunity. You and your government failed."

OUCH: Conservative Dan Albas runs out of patience with Freeland and her lame excuses: "You come to this committee and you talk, talk, talk. You're not willing to do the basics, which is to -let me just finish here - do the basics of standing up for Canadian workers. This was your opportunity. You and your government failed."

Sheila Gunn Reid

243,215 views • 1 year ago

Kimi K3 vs GPT-5.6 Sol vs Fable 5 Three top-tier models. Same /design prompt across all. Parameters reviewed: Pace, design sense, gameplay feel Result: Kimi K3 nails the design sense and knows exactly what to add in-game. Pace and gameplay held up well. Fable 5 and GPT-5.6 Sol both play too fast, video isn't sped up. Ranking (DX, Features & Cost): → Kimi K3: 9.5/10 · $0.030 → Fable 5: 7.5/10 · $0.38 → GPT-5.6 Sol: 7/10 · $0.11

Kimi K3 vs GPT-5.6 Sol vs Fable 5 Three top-tier models. Same /design prompt across all. Parameters reviewed: Pace, design sense, gameplay feel Result: Kimi K3 nails the design sense and knows exactly what to add in-game. Pace and gameplay held up well. Fable 5 and GPT-5.6 Sol both play too fast, video isn't sped up. Ranking (DX, Features & Cost): → Kimi K3: 9.5/10 · $0.030 → Fable 5: 7.5/10 · $0.38 → GPT-5.6 Sol: 7/10 · $0.11

Command Code

377,252 views • 2 months ago