State of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI | Lex Fridman Podcast #490
Nathan Lambert and Sebastian Raschka are machine learning researchers, engineers, and educators. Nathan is the post-training lead at the Allen Institute for AI (Ai2) and the author of The RLHF Book. Sebastian Raschka is the author of Build a Large Language Model (From Scratch) and Build a Reasoning Model (From Scratch). Thank you for listening ❤ Check out our sponsors: https://lexfridman.com/sponsors/ep490-sb See below for timestamps, transcript, and to give feedback, submit questions, contact Lex, etc. *Transcript:* https://lexfridman.com/ai-sota-2026-transcript *Correction:* Here's an updated image listing a collection of recent open & closed AI models with some improvements & fixes: https://lexfridman.com/wordpress/wp-content/uploads/2026/01/ai_models_2025.png *CONTACT LEX:* *Feedback* - give feedback to Lex: https://lexfridman.com/survey *AMA* - submit questions, videos or call-in: https://lexfridman.com/ama *Hiring* - join our team: https://lexfridman.com/hiring *Other* - other ways to get in touch: https://lexfridman.com/contact *EPISODE LINKS:* Nathan's X: https://x.com/natolambert Nathan's Blog: https://interconnects.ai Nathan's Website: https://natolambert.com Nathan's YouTube: https://youtube.com/@natolambert Nathan's GitHub: https://github.com/natolambert Nathan's Book: https://rlhfbook.com Sebastian's X: https://x.com/rasbt Sebastian's Blog: https://magazine.sebastianraschka.com Sebastian's Website: https://sebastianraschka.com Sebastian's YouTube: https://youtube.com/@SebastianRaschka Sebastian's GitHub: https://github.com/rasbt Sebastian's Books: Build a Large Language Model (From Scratch): https://manning.com/books/build-a-large-language-model-from-scratch Build a Reasoning Model (From Scratch): https://manning.com/books/build-a-reasoning-model-from-scratch *SPONSORS:* To support this podcast, check out our sponsors & get discounts: *Box:* Intelligent content management platform. Go to https://lexfridman.com/s/box-ep490-sb *Quo:* Phone system (calls, texts, contacts) for businesses. Go to https://lexfridman.com/s/quo-ep490-sb *UPLIFT Desk:* Standing desks and office ergonomics. Go to https://lexfridman.com/s/uplift_desk-ep490-sb *Fin:* AI agent for customer service. Go to https://lexfridman.com/s/fin-ep490-sb *Shopify:* Sell stuff online. Go to https://lexfridman.com/s/shopify-ep490-sb *CodeRabbit:* AI-powered code reviews. Go to https://lexfridman.com/s/coderabbit-ep490-sb *LMNT:* Zero-sugar electrolyte drink mix. Go to https://lexfridman.com/s/lmnt-ep490-sb *Perplexity:* AI-powered answer engine. Go to https://lexfridman.com/s/perplexity-ep490-sb *OUTLINE:* 0:00 - Introduction 1:57 - China vs US: Who wins the AI race? 10:38 - ChatGPT vs Claude vs Gemini vs Grok: Who is winning? 21:38 - Best AI for coding 28:29 - Open Source vs Closed Source LLMs 40:08 - Transformers: Evolution of LLMs since 2019 48:05 - AI Scaling Laws: Are they dead or still holding? 1:04:12 - How AI is trained: Pre-training, Mid-training, and Post-training 1:37:18 - Post-training explained: Exciting new research directions in LLMs 1:58:11 - Advice for beginners on how to get into AI development & research 2:21:03 - Work culture in AI (72+ hour weeks) 2:24:49 - Silicon Valley bubble 2:28:46 - Text diffusion models and other new research directions 2:34:28 - Tool use 2:38:44 - Continual learning 2:44:06 - Long context 2:50:21 - Robotics 2:59:31 - Timeline to AGI 3:06:47 - Will AI replace programmers? 3:25:18 - Is the dream of AGI dying? 3:32:07 - How AI will make money? 3:36:29 - Big acquisitions in 2026 3:41:01 - Future of OpenAI, Anthropic, Google DeepMind, xAI, Meta 3:53:35 - Manhattan Project for AI 4:00:10 - Future of NVIDIA, GPUs, and AI compute clusters 4:08:15 - Future of human civilization *PODCAST LINKS:* - Podcast Website: https://lexfridman.com/podcast - Apple Podcasts: https://apple.co/2lwqZIr - Spotify: https://spoti.fi/2nEwCF8 - RSS: https://lexfridman.com/feed/podcast/ - Podcast Playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4 - Clips Channel: https://www.youtube.com/lexclips *SOCIAL LINKS:* - X: https://x.com/lexfridman - Instagram: https://instagram.com/lexfridman - TikTok: https://tiktok.com/@lexfridman - LinkedIn: https://linkedin.com/in/lexfridman - Facebook: https://facebook.com/lexfridman - Patreon: https://patreon.com/lexfridman - Telegram: https://t.me/lexfridman - Reddit: https://reddit.com/r/lexfridman
Read Video · Transkripsi & Wawasan
Episode ini punya transkripsi lengkap + AI insights
Akun gratis · tanpa kartu · 150 kredit saat daftar, cukup untuk episode ini
- 📄 Transkripsi lengkap dengan stempel waktu
- ✨ Ringkasan AI, kata kunci & peta pikiran
- 💡 Poin utama & kutipan
Garis waktu episode
The DeepSeek moment and who is winning the global AI race between China and the US
- DeepSeek-R1's early 2025 release surprised everyone with near state-of-the-art performance at allegedly much lower compute cost, kicking off an intense acceleration in AI competition.
- Sebastian argues there will be no clear winner because researchers rotate between labs, so ideas won't stay proprietary—the real differentiator will be budget and hardware constraints.
- Nathan notes Anthropic's Claude 3.5 Opus hype is organic and culturally Anthropic is known for betting hard on code with Claude Code, while China now has many labs beyond DeepSeek like Zhipu, MiniMax, and Kimi Moonshot releasing strong open-weight models.
Why Chinese companies keep releasing open-weight models and how long it will last
- Nathan predicts Chinese open-weight releases will continue for a few years because top US tech companies won't pay for Chinese APIs due to security concerns, so open weights are a way to influence the huge US AI expenditure market.
- DeepSeek is secretive in communication but open in technical reports, while startups like MiniMax and Moonshot AI have filed IPO paperwork and seek Western mindshare.
- Sebastian notes DeepSeek isn't getting worse—others are adopting its ideas, creating a leapfrogging dynamic where the most recent model is usually the best.
Which models win 2024 and 2025: ChatGPT, Gemini, and Anthropic's positioning
- Nathan sees the consumer chatbot race as a bet on Gemini versus ChatGPT, and while 2024 momentum was on Gemini's side, it's hard to bet against OpenAI because they consistently land things despite operational chaos.
- Nathan predicts Gemini will keep making progress on ChatGPT due to Google's scale and ability to separate research from product, while Anthropic will continue succeeding on the software and enterprise side.
- Google's TPU advantage comes from avoiding NVIDIA's insane chip margins and building its full stack top to bottom, though new paradigms most likely still come from OpenAI.
Konsep utama
- DeepSeek moment— The early-2025 release of DeepSeek-R1 that shocked the AI world with near-frontier performance at allegedly far lower cost.
- open-weight models— Models whose weights are publicly released, driving much of the China vs. US competition debate.
- scaling laws— The core question of whether more pre-training compute keeps making models smarter.
Kutipan penting
I don't think nowadays, in 2026, that there will be any company having access to a technology that no other company has access to.
💡— Overturns the assumption that any single lab holds a durable proprietary technological edge.
These models, the cost of training them is really low relative to the cost of serving them to hundreds of millions of users.
🤯— Reveals that the real financial burden is inference at scale, not the one-time training run.
Tindakan nyata
🧠Understanding AI Progress
Pre-training still consumes most compute and bigger models are still expected, despite talk that pre-training is dead.
This week, read the OLMo-1 section 2.4 paper to see real training cost breakdowns and update your mental model of scaling.
RLVR unlocks latent knowledge rather than teaching new math, so gains can be misleading due to contamination.
Pick one RLVR paper this week and check whether its benchmark predates the model's training cutoff.
🛠️Career and Skills
The best way to learn machine learning is to build models from scratch, as Sebastian's books advocate.
Start building a small language model from scratch this week using Sebastian Raschka's book as a guide.
Researchers rotate between labs, so proprietary ideas don't stay proprietary; resources and culture are the real moats.
Follow three AI researchers on X or Substack this week and note which lab cultures they praise.
Transkrip dan wawasan dihasilkan oleh AI dan mungkin mengandung kesalahan. Akurasi bergantung pada kualitas audio dan kejelasan pembicara — jika ada yang tidak tepat, audio asli selalu menjadi sumber kebenaran.
Episode dan video siap dibaca
Episode podcast
Pulling Back the Curtain on Sportsbooks & VIP Programs w/ Dillon Borgida | Ep 63
The Risk Takers Podcast
14 Mar 20241:38:10EN
Stock Market EMERGENCY: Sell Your Stocks Now, The Collapse Is Weeks Away!
The Diary Of A CEO with Steven Bartlett
25 Jun 20261:45:21EN
How I turn joy into art | Yinka Ilori
TED Talks Daily
5 Agu 202611:20EN
#30 Die herrscherlose Zeit - Das Interregnum
Wer wir sind und warum das nicht klappte ...
29 Okt 202542:29DE
EP678 | 🎮
Gooaye 股癌
11 Jul 202652:46ZH-Hant
163.- “Mujeres, dinero y el miedo a incomodar” con Maca Riva
LA MAGIA DEL CAOS con Aislinn Derbez
26 Mei 20261:27:54ES
Video

Neuroscientist: You Will NEVER Feel Stressed Again | Andrew Huberman
RESPIRE
27 Feb 202311:05EN
No Priors Ep. 80 | With Andrej Karpathy from OpenAI and Tesla
No Priors: AI, Machine Learning, Tech, & Startups
5 Sep 202444:15EN
Steve Jobs' 2005 Stanford Commencement Address (with intro by President John Hennessy)
Stanford
14 Mei 200822:10EN
INDIA’S GOT LATENT S2 EP4 ft. Karan Aujla, Tanmay Bhat, Gurleen Pannu, Rahul Dua
Samay Raina
2 Agu 202653:54HI
#32 Die Schlacht von Worringen - Der Freiheitskampf der Kölner
99 mal Geschichte
12 Nov 202553:29DE