Videos/AI Models
Small and large language models
Teaching Gemini to Speak YouTube: Adapting LLMs for Video Recommendations to 2B+DAU
Google's AI transformation is reshaping YouTube Google's embrace of AI is fundamentally changing how its flagship products operate, with YouTube—the world's second most visited website—serving as a critical testing ground for its Gemini models. At a recent AI conference, Google's Devansh Tandon unveiled how the company has adapted large language models to power YouTube's recommendation systems, potentially transforming how over two billion daily active users discover content on the platform. The transformation represents one of the largest-scale deployments of AI in a consumer product, with Google carefully balancing innovation against the risks of disrupting a platform that accounts for significant...
watch Jul 16, 2025Transforming search and discovery using LLMs — Tejaswi & Vinesh, Instacart
AI search revolution reshapes e-commerce discovery In a world where customers expect instant, accurate results, traditional keyword-based search is rapidly becoming obsolete. Instacart's engineering team recently demonstrated how large language models (LLMs) are fundamentally transforming how users discover products online. The integration of AI-powered semantic search doesn't just enhance the technical backend—it completely reimagines the customer experience by understanding intent rather than merely matching text patterns. Key Points Instacart transitioned from keyword-based search to semantic search powered by LLMs, allowing their system to understand user intent and context rather than just matching terms. They implemented a hybrid approach combining traditional...
watch Jul 16, 2025Netflix’s Big Bet: One model to rule recommendations: Yesu Feng, Netflix
Netflix's recommendation system evolution In the world of streaming content, Netflix stands as a towering example of how sophisticated recommendation systems can transform a business. At a recent tech conference, Yesu Feng, a key player in Netflix's recommendation engineering team, pulled back the curtain on how the streaming giant has fundamentally reimagined its approach to keeping subscribers engaged. The transformation from multiple specialized models to a unified recommendation system represents one of the most significant shifts in Netflix's technical architecture in recent years. Key insights from Netflix's recommendation evolution Architectural shift: Netflix moved from dozens of specialized models serving different...
watch Jul 16, 2025360Brew: LLM-based Personalized Ranking and Recommendation – Hamed and Maziar, LinkedIn AI
LinkedIn's AI turns the tables on search LinkedIn's pursuit of an AI-powered recommendation system represents a fascinating shift in how we experience professional content online. In a recent technical deep dive, LinkedIn AI researchers Hamed Zamani and Maziar Sanjabi unveiled 360Brew, their latest advancement in personalized content ranking and recommendation technology. This system promises to fundamentally transform how LinkedIn's 900+ million users discover everything from job opportunities to learning content across the platform. Key Points LinkedIn's 360Brew is a novel Large Language Model (LLM) approach to recommendation systems that moves beyond traditional methods by incorporating diverse user signals including views,...
watch Jul 16, 2025RL for Autonomous Coding
RL transforms how machines write code As AI increasingly infiltrates software development, a quiet revolution is unfolding at the intersection of reinforcement learning and code generation. In a recent presentation, Aakanksha Chowdhery from Reflection.ai shared groundbreaking insights into how reinforcement learning techniques are transforming the way machines write code. Her talk illuminates how autonomous coding systems are evolving beyond traditional supervised learning approaches to create more reliable, efficient programming tools. Key points from Chowdhery's presentation: Beyond imitation learning: While current code generation models are primarily trained on human-written code repositories, reinforcement learning introduces novel approaches allowing AI to learn from...
watch Jul 16, 2025A new course on Retrieval Augmented Generation (RAG) is live!
RAG transforms AI into your data expert In the rapidly evolving landscape of artificial intelligence, staying current with the latest techniques isn't just advantageous—it's essential. Retrieval Augmented Generation (RAG) has emerged as a transformative approach for organizations looking to harness their proprietary data in AI applications. DeepLearning.AI's new course on RAG, developed in collaboration with industry leaders, offers practitioners a comprehensive toolkit to implement these powerful systems. Key Points RAG fundamentally solves AI hallucination problems by grounding large language models with retrievals from reliable knowledge sources, creating more accurate and trustworthy outputs. The technique bridges the gap between pre-trained LLMs...
watch Jul 16, 2025Open Weight models are finally getting good at coding…
Open weight models rival closed AI for coding In the rapidly evolving landscape of AI development, a significant shift is taking place that could reshape how developers interact with coding assistants. Open-weight models are finally coming into their own, challenging the dominance of closed AI systems like ChatGPT and Claude in the programming domain. This advancement marks a potential inflection point where freely available, open-source models begin to rival their commercially restricted counterparts. Key Points Open-weight models have shown dramatic improvement in coding capabilities, with some now approaching or matching closed models in specific programming tasks The gap between open...
watch Jul 15, 2025Using OSS models to build AI apps with millions of users
Building AI apps that scale with open source models In a world increasingly dominated by proprietary AI systems, Hassan El Mghari offers a refreshing counternarrative. His recent talk explores how open source models can power applications serving millions of users while maintaining sustainable economics. As businesses contemplate their AI strategy, El Mghari's insights provide a compelling case for considering open source alternatives alongside the dominant proprietary options from OpenAI and Anthropic. El Mghari draws from his experience at Replit, where their AI coding assistant "Ghostwriter" serves millions of developers. His presentation walks through the practical considerations of building AI applications...
watch Jul 14, 2025Kimi K2 in 6 minutes
Kimi K2: AI assistant you might actually want In the rapidly evolving landscape of AI assistants, Anthropic has quietly launched a new iteration that deserves your attention. The recently unveiled Kimi K2 represents a significant leap forward in conversational AI technology, offering business professionals a more intuitive, capable assistant that might finally deliver on the promise of AI-powered productivity. While tools like ChatGPT and Claude have dominated headlines, Kimi K2 introduces a fresh approach worth exploring. Kimi K2 positions itself as a genuinely helpful AI assistant, designed specifically for knowledge work and creative tasks with enhanced reasoning capabilities. The system...
watch Jul 13, 2025How LLMs work for Web Devs: GPT in 600 lines of Vanilla JS
GPT from scratch: coding AI in vanilla JavaScript In the world where artificial intelligence powers everything from customer service chatbots to code generation tools, understanding how these systems actually work has become increasingly valuable. Ishan Anand's tutorial video offers something refreshingly different: a ground-up implementation of a GPT-like language model using nothing but vanilla JavaScript. For web developers looking to demystify the AI black box, this practical walkthrough bridges the gap between theoretical machine learning concepts and practical coding applications. The video demonstrates that while large language models appear magical in their capabilities, the underlying architecture follows comprehensible patterns that...
watch Jul 11, 2025AI Engineering with the Google Gemini 2.5 Model Family – Philipp Schmid, Google DeepMind
Gemini 2.5: breaking AI engineering barriers Google's Gemini 2.5 marks a significant leap forward in how developers can build with multimodal AI models. In his presentation, Philipp Schmid from Google DeepMind unveils how Gemini 2.5's architecture eliminates previous constraints around context windows and input processing, offering a new paradigm for AI engineering that combines unprecedented flexibility with simplified development approaches. The video delves into Google's latest Gemini model family, emphasizing how these advances are transforming how developers build AI applications. Schmid, clearly enthusiastic about these developments, walks through the architectural improvements that address persistent challenges in working with large language...
watch Jul 11, 2025Grok 4 Fully Tested (INSANE)
AI's breakthrough moment is finally here In a stunning leap forward, Grok 4 has arrived with capabilities that force us to rethink what's possible in artificial intelligence. As someone who's watched the AI space evolve from clever parlor tricks to genuine cognitive tools, I'm witnessing what appears to be a genuine inflection point—one where the gap between human and machine intelligence has narrowed dramatically. The recent demonstration of Grok 4's capabilities shows a system that doesn't just follow instructions but seems to understand context, nuance, and creativity in ways previous models simply couldn't approach. This isn't just another incremental improvement;...
watch Jul 10, 2025A year of Gemini progress + what comes next — Logan Kilpatrick, Google DeepMind
Google's strategic AI vision unfolds quietly Google's approach to AI is markedly different from the splashy product launches we've come to expect in Silicon Valley. As Logan Kilpatrick, Developer Relations Lead at Google DeepMind, recently outlined, the tech giant has been methodically building its Gemini ecosystem with a focus on responsible development and creating genuine value. While competitors race to claim headlines, Google appears to be playing a longer, more deliberate game with artificial intelligence. Key Points Google is intentionally taking a measured approach to AI development, prioritizing responsibility and real-world utility over rushing products to market The Gemini ecosystem...
watch Jul 9, 20252025 in LLMs so far, illustrated by Pelicans on Bicycles
AI progress races ahead while humans pedal to catch up The pace of development in large language models has accelerated dramatically in early 2025, with breakthroughs arriving almost weekly that are reshaping our expectations of artificial intelligence. Simon Willison's recent talk, whimsically illustrated with pelicans on bicycles, captures this technological vertigo perfectly. As business leaders struggle to keep pace with these developments, Willison offers a clear-eyed assessment of where we are and where we're headed. In his characteristically accessible style, Willison walks us through the current state of LLMs in 2025, highlighting how dramatically the landscape has shifted in a...
watch Jul 9, 2025Learn to post-train LLMs in this free course
Free LLM post-training for businesses In the rapidly evolving landscape of artificial intelligence, large language models (LLMs) have emerged as transformative tools for businesses across sectors. However, the gap between generic, pre-trained models and the specific needs of individual organizations remains a significant challenge. Enter post-training: the process of adapting existing LLMs to perform better on domain-specific tasks. A new free course from DeepLearning.AI and Cohere is offering businesses the knowledge they need to harness this powerful technique, potentially transforming how companies leverage AI. Key Points Post-training allows businesses to customize general-purpose LLMs for specific domains without the massive computational...
watch Jul 7, 2025Training Agentic Reasoners — Will Brown, Prime Intellect
AI agents are now teaching themselves In a move that feels straight out of a sci-fi premise, we're witnessing a crucial shift in artificial intelligence development. Prime Intellect's Will Brown has revealed a fascinating approach to creating AI systems that can genuinely reason and solve complex problems through self-training mechanisms. Rather than the usual method of force-feeding mountains of data to models, this new paradigm lets AI systems essentially teach themselves through exploration and reflection. Key developments worth your attention The technique creates AI agents that learn through a trial-and-error process called "exploration and exploitation," similar to how humans learn...
watch Jul 6, 2025We really need something like opencode to succeed!
OpenCode emerges as AI's GitHub moment The line between artificial intelligence's capabilities and human coding skill continues to blur, with the recent introduction of OpenCode representing a potential watershed moment in software development. This open-source AI coding model from xAI aims to match or exceed the capabilities of proprietary alternatives while offering unprecedented accessibility and community involvement. As someone who's followed the AI coding assistant landscape closely, I'm convinced this development deserves your attention. Key aspects that make OpenCode significant: OpenCode represents one of the first truly powerful, fully open-source AI coding models that developers can freely access, modify, and...
watch Jul 4, 2025LangChain Expression Language (LCEL)
LangChain Expression Language transforms AI workflows In the rapidly evolving landscape of AI development frameworks, LangChain has emerged as a powerful tool for developers building applications with large language models. The recent introduction of LangChain Expression Language (LCEL) marks a significant evolution in how developers can construct and manage their AI application chains. This new declarative approach to building LLM applications promises to streamline development while offering greater flexibility and maintainability. Key Points LCEL introduces a declarative, composable approach to chain building that replaces the old imperative style, allowing developers to construct chains that more clearly express their intent The...
watch Jun 7, 2025Did Google Fix Gemini 2.5 Pro?
Gemini 2.5 Pro: real progress or clever marketing? Google's latest AI model Gemini 2.5 Pro arrives with considerable fanfare and impressive demos, but how much of the hype represents genuine advancement? The new model showcases remarkable multimodal capabilities with claims of handling longer contexts and exhibiting improved reasoning – but the real question is whether these improvements address the fundamental issues that plagued earlier versions or simply represent incremental enhancements wrapped in clever marketing. Key elements of the Gemini 2.5 Pro update Context length expanded to 2 million tokens, theoretically enabling the model to process entire books, lengthy videos, or...
watch Jun 6, 2025AI Accelerates: New Gemini Model + AI Unemployment Stories Analysed
Gemini's growth shows real-world AI impact In the rapidly evolving landscape of artificial intelligence, Google's latest Gemini AI model improvements highlight both the accelerating pace of technological progress and its tangible effects on the workforce. The recent video commentary on Gemini's capabilities and its connection to employment disruption offers a sobering glimpse into our AI-powered future. As generative AI moves from novelty to necessity, businesses must grapple with how these tools are reshaping entire industries and job functions with unprecedented speed. Key insights from the analysis Gemini's latest version demonstrates substantial improvements in reasoning, problem-solving, and multimodal processing that narrow...
watch Jun 5, 2025The AI Image Tool That’s Quietly Beating GPT-4o | FLUX Kontext Deep Dive
AI's new visual frontier explained For months, the AI world has been fixated on OpenAI's GPT-4o and its multimodal capabilities. Yet quietly, a different player has emerged with potentially superior visual understanding: Anthropic's FLUX Kontext. This relatively unheralded model demonstrates surprising capabilities that might signal a significant advancement in how AI systems process and understand visual information. Key Points Visual groundedness: FLUX Kontext demonstrates an impressive ability to understand and relate elements within images, avoiding the "hallucination" problem common in other models that invent details not present in images. Contextual awareness: Unlike GPT-4o which can struggle with precise spatial relationships,...
watch Jun 5, 2025New AI Breakthrough: Most Advanced AI for Science Explained
Gemini AI brings scientific breakthrough for researchers Google's recent release of Gemini, their most powerful AI model yet, represents a significant leap forward in how artificial intelligence can accelerate scientific discovery. The announcement brings remarkable new capabilities to researchers across disciplines through multimodal understanding that processes text, code, audio, images, and video simultaneously. This breakthrough AI promises to transform how scientists work by combining deep reasoning with the ability to understand complex scientific content. Key insights from Gemini's release: Gemini processes multiple types of information simultaneously (multimodal), allowing it to reason across text, images, code, audio and video - making...
watch Jun 4, 2025The Easiest Way to Build an App in 2025 (Claude Code)
Claude Code could reshape app development landscape In a recent video that's been making waves across tech circles, the spotlight falls on what might be the next revolution in app development: Claude Code. This AI-powered development tool from Anthropic promises to dramatically lower the barriers to entry for app creation, potentially allowing anyone with an idea to bring it to life without writing a single line of traditional code. The implications for business users and the broader software ecosystem could be profound. The video explores several groundbreaking aspects of this technology: Claude Code represents a paradigm shift in how we...
watch Jun 4, 2025Goodbye GPT-5… New DeepSeek Update is HERE! AI News EXPLAINED
The rise of DeepSeek and the "model wars" The artificial intelligence landscape continues to evolve at a breakneck pace, with new models challenging established players in unexpected ways. In a recent video, the host explores DeepSeek's latest advancements and what they mean for the broader AI ecosystem, particularly in relation to OpenAI's GPT models. The emergence of DeepSeek represents yet another shift in what's becoming an increasingly competitive field where technical capabilities, open-source philosophy, and business strategy collide. Key points from the video: DeepSeek has released a powerful new model that demonstrates remarkable capabilities in coding, reasoning, and mathematics, positioning...
watch