--- title: "Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI" source_name: "AI Engineer" original_url: "https://www.youtube.com/watch?v=fLUtUkqYHnQ" canonical_url: "https://www.traeai.com/articles/e82027ca-c98b-48d3-be81-7a53e07b68a2" content_type: "video" language: "其他" score: 0 tags: [] published_at: "2026-04-29T12:00:06+00:00" created_at: "2026-04-30T09:27:11.202168+00:00" --- # Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI Canonical URL: https://www.traeai.com/articles/e82027ca-c98b-48d3-be81-7a53e07b68a2 Original source: https://www.youtube.com/watch?v=fLUtUkqYHnQ ## Summary 该链接指向的是一个YouTube视频页面，而非一篇技术文章。无法从中提取有效信息。 ## Key Takeaways - 该链接指向的是一个YouTube视频页面，而非一篇技术文章 - 无法从中提取有效信息 ## Content Title: Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI URL Source: http://www.youtube.com/watch?v=fLUtUkqYHnQ Markdown Content: # Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI - YouTube Back [![Image 1](http://www.youtube.com/watch?v=fLUtUkqYHnQ)](http://www.youtube.com/ "YouTube Home") Skip navigation Search Search with your voice [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [Sign in](https://accounts.google.com/ServiceLogin?service=youtube&uilel=3&passive=true&continue=https%3A%2F%2Fwww.youtube.com%2Fsignin%3Faction_handle_signin%3Dtrue%26app%3Ddesktop%26hl%3Den%26next%3Dhttps%253A%252F%252Fwww.youtube.com%252Fwatch%253Fv%253DfLUtUkqYHnQ&hl=en&ec=65620) [![Image 2](http://www.youtube.com/watch?v=fLUtUkqYHnQ)](http://www.youtube.com/ "YouTube Home") [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) Tap to unmute 2x [![Image 3](http://www.youtube.com/watch?v=fLUtUkqYHnQ)](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ## Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI AI Engineer 10,130 views 21 hours ago [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) Search Copy link Info Shopping ![Image 4](http://www.youtube.com/watch?v=fLUtUkqYHnQ) [![Image 5](http://www.youtube.com/watch?v=fLUtUkqYHnQ)](http://www.youtube.com/watch?v=fLUtUkqYHnQ) If playback doesn't begin shortly, try restarting your device. • You're signed out Videos you watch may be added to the TV's watch history and influence TV recommendations. To avoid this, cancel and sign in to YouTube on your computer. Cancel Confirm [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) Share [](http://www.youtube.com/watch?v=fLUtUkqYHnQ "Share link")- [x] Include playlist An error occurred while retrieving sharing information. Please try again later. ![Image 6](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 0:00 [](http://www.youtube.com/watch?v=fLUtUkqYHnQ)[](https://www.youtube.com/watch?v=96jN2OCOfLs "Next (SHIFT+n)") 0:00 / 0:00 Live •Watch full video • Start • [29:49 Andrej Karpathy: From Vibe Coding to Agentic Engineering Sequoia Capital 149K views • 18 hours ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=96jN2OCOfLs)[40:57 Demis Hassabis: Agents, AGI & The Next Big Scientific Breakthrough Y Combinator 50K views • 19 hours ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=JNyuX1zoOgU)[2:13:41 How GPT-5, Claude, and Gemini are actually trained and served – Reiner Pope Dwarkesh Patel 80K views • 16 hours ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=xmkSf5IS-zw)[18:26 "Software Fundamentals Matter More Than Ever" — Matt Pocock AI Engineer and Matt Pocock 409K views • 6 days ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=v4F1gFy-hqg)[19:03 Open Models at Google DeepMind — Cassidy Hardin, Google DeepMind AI Engineer 10K views • 2 days ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=_A367W_qvc8)[9:50 They Lied to You About AI (This Study Proves It)Caleb Ulku 233K views • 1 month ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=AIYQp1n51ZI)[23:16 GitHub is facing HUGE problems!Maximilian Schwarzmüller 44K views • 22 hours ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=pekbl3Yz02g)[27:03 Agentic Engineering: Working With AI, Not Just Using It — Brendan O'Leary AI Engineer 57K views • 3 weeks ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=BEKc4P87XKo)[1:06:32 Skill Issue: Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AI No Priors: AI, Machine Learning, Tech, & Startups 772K views • 1 month ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=kwSVtQ7dziU)[1:36:30 Full Walkthrough: Workflow for AI Coding — Matt Pocock AI Engineer and Matt Pocock 216K views • 5 days ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=-QFHIoCo-Ko)[19:51 What Is Yann LeCun Cooking? JEPA Explained Simply bycloud 89K views • 9 days ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=oM4neOyZOi0)[26:46 Everything We Got Wrong About Research-Plan-Implement - Dexter Horthy MLOps.community 102K views • 1 month ago Live Playlist ()Mix (50+)](https://www.youtube.com/watch?v=YwZR6tc7qYg) Sign in to confirm you’re not a bot This helps protect our community. [Learn more](https://support.google.com/youtube/answer/3037019#zippy=%2Ccheck-that-youre-signed-into-youtube) [Sign in](https://accounts.google.com/ServiceLogin?service=youtube&uilel=3&passive=true&continue=https%3A%2F%2Fwww.youtube.com%2Fsignin%3Faction_handle_signin%3Dtrue%26app%3Ddesktop%26hl%3Den%26next%3D%252Fwatch%253Fv%253DfLUtUkqYHnQ&hl=en) # Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI [![Image 7](https://yt3.ggpht.com/ajVemEB89DAOemsbfuMY6ZOWXJAACx3cbty9z21jeqRKODaVkDBSRun1b1xfQJljEsziOWS_Mg=s48-c-k-c0x00ffffff-no-rj)](http://www.youtube.com/@aiDotEngineer) [AI Engineer](http://www.youtube.com/@aiDotEngineer) AI Engineer 435K subscribers Subscribe Subscribed 336 Share Save Download Download 10K views 21 hours ago 10,130 views • Premiered 21 hours ago A new class of small models is emerging with the ability to reliably follow instructions and call tools while running on-device under 1 GB of memory. In this talk, we'll break down how to post-train frontier small models using the LFM2.5 recipe: on-policy preference alignment, agentic reinforcement learning, and curriculum training with iterative model merging. We'll cover trai…...more ...more How this was made Auto-dubbed Audio tracks for some languages were automatically generated. [Learn more](http://support.google.com/youtube/answer/15569972?hl=en) ## Chapters View all [![Image 8](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Start #### Start 0:00](http://www.youtube.com/watch?v=fLUtUkqYHnQ&pp=0gcJCREDDuyUWbzu) #### [Start](http://www.youtube.com/watch?v=fLUtUkqYHnQ&pp=0gcJCREDDuyUWbzu) 0:00 [![Image 9](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Introduction to frontier small models at Liquid AI #### Introduction to frontier small models at Liquid AI 0:14](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) #### [Introduction to frontier small models at Liquid AI](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) 0:14 [![Image 10](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Characteristics: memory-bound, task-specific, latency-sensitive #### Characteristics: memory-bound, task-specific, latency-sensitive 1:02](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) #### [Characteristics: memory-bound, task-specific, latency-sensitive](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) 1:02 [![Image 11](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Architecture: why large embedding layers are inefficient #### Architecture: why large embedding layers are inefficient 2:20](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) #### [Architecture: why large embedding layers are inefficient](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) 2:20 [![Image 12](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### LFM2 architecture: using gated short convolutions for speed #### LFM2 architecture: using gated short convolutions for speed 4:01](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=241s) #### [LFM2 architecture: using gated short convolutions for speed](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=241s) 4:01 [![Image 13](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### LFM 2.5 recipe: 28T tokens and post-training stages #### LFM 2.5 recipe: 28T tokens and post-training stages 6:09](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=369s) #### [LFM 2.5 recipe: 28T tokens and post-training stages](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=369s) 6:09 [![Image 14](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Post-training: SFT, preference alignment, and RL best practices #### Post-training: SFT, preference alignment, and RL best practices 8:34](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=514s) #### [Post-training: SFT, preference alignment, and RL best practices](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=514s) 8:34 [![Image 15](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Identifying "doom loops" in reasoning models #### Identifying "doom loops" in reasoning models 10:43](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=643s&pp=0gcJCREDDuyUWbzu) #### [Identifying "doom loops" in reasoning models](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=643s&pp=0gcJCREDDuyUWbzu) 10:43 Transcript Follow along using the transcript. Show transcript [![Image 16](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### AI Engineer 435K subscribers](http://www.youtube.com/@aiDotEngineer) [Videos](http://www.youtube.com/channel/UCLKPca3kwwd-B59HNr-_lvA/videos)[About](http://www.youtube.com/channel/UCLKPca3kwwd-B59HNr-_lvA/about) Show less [](http://www.youtube.com/watch?v=fLUtUkqYHnQ) # Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI 10,130 views 10K views Premiered 21 hours ago 336 Share Save Download Download ## 6 Comments Sort comments Sort by [Top Show featured comments](http://www.youtube.com/watch?v=fLUtUkqYHnQ)[Newest Show recent comments, including potential spam](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ![Image 17: Default profile photo](https://yt3.ggpht.com/a/default-user=s48-c-k-c0x00ffffff-no-rj) Add a comment... ![Image 18](https://yt3.ggpht.com/ajVemEB89DAOemsbfuMY6ZOWXJAACx3cbty9z21jeqRKODaVkDBSRun1b1xfQJljEsziOWS_Mg=s88-c-k-c0x00ffffff-no-rj) Pinned by @aiDotEngineer ### [@aiDotEngineer](http://www.youtube.com/@aiDotEngineer) [@aiDotEngineer @aiDotEngineer](http://www.youtube.com/@aiDotEngineer)[15 hours ago](http://www.youtube.com/watch?v=fLUtUkqYHnQ&lc=Ugyu-0G9RLXz3QvNJqF4AaABAg) find Maxime and slides for this presentaiton below! [https://x.com/maximelabonne/status/2049549363181977958?s=20](https://www.youtube.com/redirect?event=comments&redir_token=QUFFLUhqazJUMm1oN2RCdDdLbDUyeDduR01ublRlbnFWUXxBQ3Jtc0ttNnY5NjcxdlZIRGpPTTd6d1oyVjl4c3otT3pwLWRjU2VMZkZvdEdqYUNhdXBCdDJpQ1Fxa2RRazJZRjkwVzBSUXB0YnhWSEc2cXFPMUJtd0hwSDBldGlBLVBRWGhkR1pVR0diOFRwWVJyZ3RqTGtqVQ&q=https%3A%2F%2Fx.com%2Fmaximelabonne%2Fstatus%2F2049549363181977958%3Fs%3D20) Show less Read more Like Dislike Reply ![Image 19](https://yt3.ggpht.com/ytc/AIdro_mf2D2VCwHdQqXvsJxAAAVsgXxXUsUqpwIRGOA7R62mAxg=s88-c-k-c0x00ffffff-no-rj) ### [@loicbaconnier9150](http://www.youtube.com/@loicbaconnier9150) [21 hours ago](http://www.youtube.com/watch?v=fLUtUkqYHnQ&lc=UgxrOdJjmjIpdH9tafx4AaABAg) Salut Maxime, content que tu sois la, merci Show less Read more Like 7 Dislike Reply 1 reply Hide replies 1 reply ![Image 20](https://yt3.ggpht.com/ytc/AIdro_mw38HRwBWr8mTXZ69IRwjMHDKrj9jZ_V04rhMpC4Oc3qY=s88-c-k-c0x00ffffff-no-rj) ### [@jakespeakz](http://www.youtube.com/@jakespeakz) [19 hours ago](http://www.youtube.com/watch?v=fLUtUkqYHnQ&lc=UgwFhM8m59orqY9mHsl4AaABAg) Badly waiting for a liquid lfm video! Thanks ![Image 21: 😊](https://www.youtube.com/s/gaming/emoji/7ff574f2/emoji_u1f60a.png) Show less Read more Like 4 Dislike Reply ![Image 22](https://yt3.ggpht.com/ytc/AIdro_nMTu7wvdc3PSUAflnWSSKDrtwGgjy1Y-XbGxtQJ3oIEzo=s88-c-k-c0x00ffffff-no-rj) ### [@saad3838](http://www.youtube.com/@saad3838) [16 hours ago](http://www.youtube.com/watch?v=fLUtUkqYHnQ&lc=UgyT3WMVSs8AhObKyRV4AaABAg) For real i've enjoyed listen to you Maxime, it was informative lecture ![Image 23: 👌](https://www.youtube.com/s/gaming/emoji/7ff574f2/emoji_u1f44c.png) Show less Read more Like 1 Dislike Reply ![Image 24](https://yt3.ggpht.com/ytc/AIdro_mHmuU06D7d5sBRgDRpRTvVd8vsSTtRz8SeQaLfhG4HVx1vDFIuKij6U4A3NXKz4jBOYA=s88-c-k-c0x00ffffff-no-rj) ### [@ausmiku](http://www.youtube.com/@ausmiku) [10 hours ago](http://www.youtube.com/watch?v=fLUtUkqYHnQ&lc=UgzMYmeajQ60TBH05XB4AaABAg) Why aren't the AI influencers on their super yachts ? LOL. The AI fanboys don't get it. AI companies are shonky grifters. They will tank. The cost of tokens is greater than the return from subscriptions. 99% of AI is used as a huge free encyclopaedia . Future AI will be open source models on our PCs. Gemma 4 is Google's local LLM. It's free, no licence, no subscriptions, no cloud. The upcoming M5 Mac Mini will be like having a personal data centre in your home. So ChatGPT, Claude, Grok etc. will not be needed. Good riddance to these fraudsters and charlatans. Show less Read more Like 1 Dislike Reply Top is selected, so you'll see featured comments ## Comments 6 [Top Show featured comments](http://www.youtube.com/watch?v=fLUtUkqYHnQ)[Newest Show recent comments, including potential spam](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ## In this video Chapters Transcript ## Chapters [![Image 25](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Start #### Start 0:00](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### [Start](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 0:00 [![Image 26](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Introduction to frontier small models at Liquid AI #### Introduction to frontier small models at Liquid AI 0:14](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) #### [Introduction to frontier small models at Liquid AI](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) 0:14 [![Image 27](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Characteristics: memory-bound, task-specific, latency-sensitive #### Characteristics: memory-bound, task-specific, latency-sensitive 1:02](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) #### [Characteristics: memory-bound, task-specific, latency-sensitive](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) 1:02 [![Image 28](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Architecture: why large embedding layers are inefficient #### Architecture: why large embedding layers are inefficient 2:20](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) #### [Architecture: why large embedding layers are inefficient](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) 2:20 [![Image 29](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### LFM2 architecture: using gated short convolutions for speed #### LFM2 architecture: using gated short convolutions for speed 4:01](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=241s) #### [LFM2 architecture: using gated short convolutions for speed](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=241s) 4:01 [![Image 30](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### LFM 2.5 recipe: 28T tokens and post-training stages #### LFM 2.5 recipe: 28T tokens and post-training stages 6:09](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=369s) #### [LFM 2.5 recipe: 28T tokens and post-training stages](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=369s) 6:09 [![Image 31](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Post-training: SFT, preference alignment, and RL best practices #### Post-training: SFT, preference alignment, and RL best practices 8:34](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=514s) #### [Post-training: SFT, preference alignment, and RL best practices](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=514s) 8:34 [![Image 32](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Identifying "doom loops" in reasoning models #### Identifying "doom loops" in reasoning models 10:43](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=643s) #### [Identifying "doom loops" in reasoning models](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=643s) 10:43 [![Image 33](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Solutions: mitigating loops via preference alignment and RL #### Solutions: mitigating loops via preference alignment and RL 11:34](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=694s) #### [Solutions: mitigating loops via preference alignment and RL](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=694s) 11:34 [![Image 34](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Future focus: using agentic tools to overcome memory limits #### Future focus: using agentic tools to overcome memory limits 15:29](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=929s) #### [Future focus: using agentic tools to overcome memory limits](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=929s) 15:29 [![Image 35](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Q&A: real-world applications for small vs. large models #### Q&A: real-world applications for small vs. large models 17:58](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=1078s) #### [Q&A: real-world applications for small vs. large models](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=1078s) 17:58 Sync to video time ## Description Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI A new class of small models is emerging with the ability to reliably follow instructions and call tools while running on-device under 1 GB of memory. In this talk, we'll break down how to post-train frontier small models using the LFM2.5 recipe: on-policy preference alignment, agentic reinforcement learning, and curriculum training with iterative model merging. We'll cover training challenges unique to the 1B scale, like doom loops, capability interference, and how to fix them. The goal is to give you a concrete playbook to fine-tune and deploy small models for your own use cases, from structured data extraction to multi-turn tool use. Speaker info: * [https://x.com/maximelabonne](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqa2V2T3F6Y0Y2dzVEMk1sbDBjalRpRGtzSXo2QXxBQ3Jtc0tsNzdLUGM2TjBweURCNEMyZ3dGNUtpODJDb3l2QzViQ1p0NzVNSkJNSFdIekhqUjdEMlprTUtJYXhVaDRVdDZOLVBGQ3Zaalhvcm1DVXI2UEo0QmlnZ1cyMlJCQWoydWdZX0ZTNTVQcENsOVlsVXd3Zw&q=https%3A%2F%2Fx.com%2Fmaximelabonne&v=fLUtUkqYHnQ) * [![Image 36](http://www.youtube.com/watch?v=fLUtUkqYHnQ)/maxime-labonne](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqbWkxWEdGQWJGdkJ1Zk5fMEZodlpHOGpKWHNMQXxBQ3Jtc0tublJydDdYZU9fWDhCNEVNYkQ2dlZ4b0R4SU45alQ3RWc0d0JQVjZ2aWlJSWQwczNfTVZfUFRVTWlsbF9OaURmUDA5QTAwbFE4RTQ4RWNManBvLXh4eG5DbzhGMkdEd0QwWXRGdlhsNS14Rk1CVlhmSQ&q=https%3A%2F%2Fwww.linkedin.com%2Fin%2Fmaxime-labonne%2F&v=fLUtUkqYHnQ) * [https://github.com/mlabonne](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqbGRwQXNrbUxlZnRpQUEtSXY1VjFocW1KaExJUXxBQ3Jtc0trRmVXSW1UYmduY3V4Tm9pbVdJY3NobHJSdG5tUGhWU1F4blZWRERGZ3ViZGdwMi1rTlVpRFVoODZNckdTNDRIb2xmZ3U1YjJwQVREYUVyREtYUnY1am5ITEpXX2t5WUM1V3llYlFuRVM4aXBIT2xMRQ&q=https%3A%2F%2Fgithub.com%2Fmlabonne&v=fLUtUkqYHnQ) Timestamps: [0:00:00](http://www.youtube.com/watch?v=fLUtUkqYHnQ) - Start [0:00:14](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) - Introduction to frontier small models at Liquid AI [0:01:02](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) - Characteristics: memory-bound, task-specific, latency-sensitive [0:02:20](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) - Architecture: why large embedding layers are inefficient [0:04:01](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=241s) - LFM2 architecture: using gated short convolutions for speed [0:06:09](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=369s) - LFM 2.5 recipe: 28T tokens and post-training stages [0:08:34](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=514s) - Post-training: SFT, preference alignment, and RL best practices [0:10:43](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=643s) - Identifying "doom loops" in reasoning models [0:11:34](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=694s) - Solutions: mitigating loops via preference alignment and RL [0:15:29](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=929s) - Future focus: using agentic tools to overcome memory limits [0:17:58](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=1078s) - Q&A: real-world applications for small vs. large models…...more ...more Show less How this was made Auto-dubbed Audio tracks for some languages were automatically generated. [Learn more](http://support.google.com/youtube/answer/15569972?hl=en) ## Chapters View all [![Image 37](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Start #### Start 0:00](http://www.youtube.com/watch?v=fLUtUkqYHnQ&pp=0gcJCREDDuyUWbzu) #### [Start](http://www.youtube.com/watch?v=fLUtUkqYHnQ&pp=0gcJCREDDuyUWbzu) 0:00 [![Image 38](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Introduction to frontier small models at Liquid AI #### Introduction to frontier small models at Liquid AI 0:14](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) #### [Introduction to frontier small models at Liquid AI](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=14s) 0:14 [![Image 39](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Characteristics: memory-bound, task-specific, latency-sensitive #### Characteristics: memory-bound, task-specific, latency-sensitive 1:02](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) #### [Characteristics: memory-bound, task-specific, latency-sensitive](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=62s) 1:02 [![Image 40](http://www.youtube.com/watch?v=fLUtUkqYHnQ) #### Architecture: why large embedding layers are inefficient #### Architecture: why large embedding layers are inefficient 2:20](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) #### [Architecture: why large embedding layers are inefficient](http://www.youtube.com/watch?v=fLUtUkqYHnQ&t=140s) 2:20 Transcript Follow along using the transcript. Show transcript [![Image 41](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### AI Engineer 435K subscribers](http://www.youtube.com/@aiDotEngineer) [Videos](http://www.youtube.com/channel/UCLKPca3kwwd-B59HNr-_lvA/videos)[About](http://www.youtube.com/channel/UCLKPca3kwwd-B59HNr-_lvA/about) ## Transcript NaN / NaN Chat Replay is disabled for this Premiere. [![Image 42](https://i.ytimg.com/vi/96jN2OCOfLs/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLBZdp1cX62GE_0F_7p8NNyecd0HIQ) 29:49](http://www.youtube.com/watch?v=96jN2OCOfLs&pp=ugUEEgJlbg%3D%3D) ![Image 43](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Andrej Karpathy: From Vibe Coding to Agentic Engineering](http://www.youtube.com/watch?v=96jN2OCOfLs&pp=ugUEEgJlbg%3D%3D) Sequoia Capital 149K views • 18 hours ago New [![Image 44](https://i.ytimg.com/vi/JNyuX1zoOgU/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLDn0yBys-Of1HHHC6hqYyutUUpWPg) 40:57](http://www.youtube.com/watch?v=JNyuX1zoOgU&pp=ugUHEgVlbi1VUw%3D%3D) ![Image 45](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Demis Hassabis: Agents, AGI & The Next Big Scientific Breakthrough](http://www.youtube.com/watch?v=JNyuX1zoOgU&pp=ugUHEgVlbi1VUw%3D%3D) Y Combinator 50K views • 19 hours ago New [![Image 46](https://i.ytimg.com/vi/xmkSf5IS-zw/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLAhYxJtTkwgBGvlbXSHSit0X390ZQ) 2:13:41](http://www.youtube.com/watch?v=xmkSf5IS-zw) ![Image 47](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [How GPT-5, Claude, and Gemini are actually trained and served – Reiner Pope](http://www.youtube.com/watch?v=xmkSf5IS-zw) Dwarkesh Patel 80K views • 16 hours ago New [![Image 48](https://i.ytimg.com/vi/v4F1gFy-hqg/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLCd08R5OZiXI3579qZ8XJrdyjcTEA) 18:26](http://www.youtube.com/watch?v=v4F1gFy-hqg&pp=ugUEEgJlbg%3D%3D) ![Image 49](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ![Image 50](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### ["Software Fundamentals Matter More Than Ever" — Matt Pocock](http://www.youtube.com/watch?v=v4F1gFy-hqg&pp=ugUEEgJlbg%3D%3D) AI Engineer and Matt Pocock 409K views • 6 days ago New [![Image 51](https://i.ytimg.com/vi/_A367W_qvc8/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLD66Vs6AQ0AOK3gterMwHyjieWr1Q) 19:03](http://www.youtube.com/watch?v=_A367W_qvc8&pp=ugUEEgJlbg%3D%3D) ![Image 52](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Open Models at Google DeepMind — Cassidy Hardin, Google DeepMind](http://www.youtube.com/watch?v=_A367W_qvc8&pp=ugUEEgJlbg%3D%3D) AI Engineer 10K views • 2 days ago New [![Image 53](https://i.ytimg.com/vi/AIYQp1n51ZI/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLAZOErFB73cGmEYC6hzAqbxzljtkA) 9:50](http://www.youtube.com/watch?v=AIYQp1n51ZI) ![Image 54](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [They Lied to You About AI (This Study Proves It)](http://www.youtube.com/watch?v=AIYQp1n51ZI) Caleb Ulku 233K views • 1 month ago [![Image 55](https://i.ytimg.com/vi/pekbl3Yz02g/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLBb9QmpXGAgda8Cz7rnKQ76RFPkEA) 23:16](http://www.youtube.com/watch?v=pekbl3Yz02g&pp=ugUEEgJlbtIHCQnfCgGHKiGM7w%3D%3D) ![Image 56](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [GitHub is facing HUGE problems!](http://www.youtube.com/watch?v=pekbl3Yz02g&pp=ugUEEgJlbtIHCQnfCgGHKiGM7w%3D%3D) Maximilian Schwarzmüller 44K views • 22 hours ago New [![Image 57](https://i.ytimg.com/vi/BEKc4P87XKo/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLAPrqrd1Sn6YCY73dbd1zrSjz6p1A) 27:03](http://www.youtube.com/watch?v=BEKc4P87XKo&pp=ugUEEgJlbg%3D%3D) ![Image 58](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Agentic Engineering: Working With AI, Not Just Using It — Brendan O'Leary](http://www.youtube.com/watch?v=BEKc4P87XKo&pp=ugUEEgJlbg%3D%3D) AI Engineer 57K views • 3 weeks ago [![Image 59](https://i.ytimg.com/vi/kwSVtQ7dziU/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLCBZVADa7hzE5H0FMp70mD4AUxJvQ) 1:06:32](http://www.youtube.com/watch?v=kwSVtQ7dziU) ![Image 60](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Skill Issue: Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AI](http://www.youtube.com/watch?v=kwSVtQ7dziU) No Priors: AI, Machine Learning, Tech, & Startups 772K views • 1 month ago [![Image 61](https://i.ytimg.com/vi/-QFHIoCo-Ko/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLAjqfhVXCym1J3M1Q5FFsmLv37FBg) 1:36:30](http://www.youtube.com/watch?v=-QFHIoCo-Ko&pp=ugUEEgJlbg%3D%3D) ![Image 62](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ![Image 63](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Full Walkthrough: Workflow for AI Coding — Matt Pocock](http://www.youtube.com/watch?v=-QFHIoCo-Ko&pp=ugUEEgJlbg%3D%3D) AI Engineer and Matt Pocock 216K views • 5 days ago New [![Image 64](https://i.ytimg.com/vi/oM4neOyZOi0/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLAMn0XUHUblj7e0zlWPQr7TIqbmtw) 19:51](http://www.youtube.com/watch?v=oM4neOyZOi0&pp=ugUEEgJlbg%3D%3D) ![Image 65](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [What Is Yann LeCun Cooking? JEPA Explained Simply](http://www.youtube.com/watch?v=oM4neOyZOi0&pp=ugUEEgJlbg%3D%3D) bycloud 89K views • 9 days ago [![Image 66](https://i.ytimg.com/vi/YwZR6tc7qYg/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLAoxboHs4rBFF89Zym3dYyejoNbsA) 26:46](http://www.youtube.com/watch?v=YwZR6tc7qYg&pp=ugUEEgJlbg%3D%3D) ![Image 67](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Everything We Got Wrong About Research-Plan-Implement - Dexter Horthy](http://www.youtube.com/watch?v=YwZR6tc7qYg&pp=ugUEEgJlbg%3D%3D) MLOps.community 102K views • 1 month ago [![Image 68](https://i.ytimg.com/vi/rmvDxxNubIg/hqdefault.jpg?sqp=-oaymwEmCKgBEF5IWvKriqkDGQgBFQAAiEIYAdgBAeIBCggYEAIYBjgBQAE=&rs=AOn4CLDueMyAvIdVvQCXgASCCFD4hu7p4A) 20:31](http://www.youtube.com/watch?v=rmvDxxNubIg&pp=ugUEEgJlbg%3D%3D) ![Image 69](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [No Vibes Allowed: Solving Hard Problems in Complex Codebases –Dex Horthy, HumanLayer](http://www.youtube.com/watch?v=rmvDxxNubIg&pp=ugUEEgJlbg%3D%3D) AI Engineer 512K views • 4 months ago [![Image 70](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 1:25:49](http://www.youtube.com/watch?v=SVOrURyOu_U&pp=ugUHEgVlbi1VUw%3D%3D) ![Image 71](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Designing Data-intensive Applications with Martin Kleppmann](http://www.youtube.com/watch?v=SVOrURyOu_U&pp=ugUHEgVlbi1VUw%3D%3D) The Pragmatic Engineer 206K views • 7 days ago [![Image 72](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 26:52](http://www.youtube.com/watch?v=AFpeWo1GTeg&pp=ugUEEgJlbg%3D%3D) ![Image 73](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Demis Hassabis: We're Three Quarters of the Way to AGI](http://www.youtube.com/watch?v=AFpeWo1GTeg&pp=ugUEEgJlbg%3D%3D) Sequoia Capital 45K views • 18 hours ago New [![Image 74](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 17:27](http://www.youtube.com/watch?v=epzzALZ8oYo) ![Image 75](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Deepseek is a problem](http://www.youtube.com/watch?v=epzzALZ8oYo) Matthew Berman 52K views • 9 hours ago New [![Image 76](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 14:37](http://www.youtube.com/watch?v=mtiOK2QG9Q0) ![Image 77](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [The 7 Skills You Need to Build AI Agents](http://www.youtube.com/watch?v=mtiOK2QG9Q0) IBM Technology 291K views • 2 weeks ago [![Image 78](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 39:58](http://www.youtube.com/watch?v=E0Q96IKXx6Q&pp=0gcJCd8KAYcqIYzv) ![Image 79](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ![Image 80](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Ex-Google Exec: How to Position Yourself Now Before the Next AI Phase (2026–2027) | Mo Gawdat](http://www.youtube.com/watch?v=E0Q96IKXx6Q&pp=0gcJCd8KAYcqIYzv) Silicon Valley Girl and Mo Gawdat 760K views • 4 weeks ago [![Image 81](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 19:53](http://www.youtube.com/watch?v=uBWuKh1nZ2Y&pp=ugUEEgJlbtIHCQnfCgGHKiGM7w%3D%3D) ![Image 82](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [The only AutoResearch tutorial you’ll ever need](http://www.youtube.com/watch?v=uBWuKh1nZ2Y&pp=ugUEEgJlbtIHCQnfCgGHKiGM7w%3D%3D) David Ondrej 200K views • 1 month ago [![Image 83](http://www.youtube.com/watch?v=fLUtUkqYHnQ) 32:23](http://www.youtube.com/watch?v=SSya123u9Yk&pp=ugUHEgVlbi1VUw%3D%3D) ![Image 84](http://www.youtube.com/watch?v=fLUtUkqYHnQ) ### [Demis Hassabis: Why AGI is Bigger than the Industrial Revolution & Where Are The Bottlenecks in AI](http://www.youtube.com/watch?v=SSya123u9Yk&pp=ugUHEgVlbi1VUw%3D%3D) 20VC with Harry Stebbings 314K views • 3 weeks ago Show more [](http://www.youtube.com/watch?v=fLUtUkqYHnQ)