@code4ai Avatar @code4ai Discover AI

Discover AI posts on YouTube about ai, llm, artificial intelligence, vision the most. They currently have [------] followers and [---] posts still getting attention that total [-----] engagements in the last [--] hours.

Engagements: [-----] #

Engagements Line Chart

Mentions: [--] #

Mentions Line Chart

Followers: [------] #

Followers Line Chart

CreatorRank: [-------] #

CreatorRank Line Chart

Social Influence

Social category influence technology brands 29.1% stocks 11.94% travel destinations 6.72% social networks 5.97% countries 5.22% currencies 2.99% finance 2.99% cryptocurrencies 0.75% celebrities 0.75% musicians 0.75%

Social topic influence ai 32.09%, llm 25.37%, artificial intelligence 23.88%, vision 14.18%, language 10.45%, $googl 7.46%, #ai 6.72%, youtube 5.97%, real world 5.97%, open ai 5.97%

Top accounts mentioned or mentioned by @openai @mit @stanford @ucberkeley @huggingface @10 @princeton @nvidia @microsoftresearch @universityofvirginia @penn @deepseekapi @apple @ethzurich @yale @oxforduniversity @fudanuniversity @capitalmedicallabofficial @nvidiaal @canva

Top assets mentioned Alphabet Inc Class A (GOOGL) Microsoft Corp. (MSFT) arXiv (ARXIV) CyberConnect (CYBER)

Top Social Posts

Top posts by engagements in the last [--] hours

"The $10T AI Economy: New Smart Protocol Emerges The Smart Contract between AI Agents New AI Protocol emerges. The future Economy of Humans and AI agents on a global competitive marketplace with decentralized auctions. Human Intelligence and machine intelligence are priced and offered globally. Is this our future Is this why we invented AI All rights w/ authors: "Intelligent AI Delegation" Nenad Tomaev1 Matija Franklin1 and Simon Osindero1 from Google DeepMind #economy #aieconomy #aibusiness #aiproductivity #aiperformance #nextgenai artificial intelligence AI models LLM VLM VLA Multi-modal"
YouTube Link 2026-02-15T14:15Z 85.2K followers, [----] engagements

"Claude OPUS [---] Thinking vs [---] Non-Thinking: Both FAIL Anthropic just released Claude OPUS [---]. And specified "Opus [---] extends the frontier of expert-level reasoning". So I test both models of Claude [---] (Thinking and Non-Thinking) on my standard causal reasoning test. You can see my complete YouTube Playlist for AI model testing on this causal reasoning test here https://www.youtube.com/watchv=g1L8uOQ7Ids&list=PLgy71-0-2-F0Rla8lu5ZldpYQUfXM_5bT https://www.anthropic.com/news/claude-opus-4-6 00:00 OPUS [---] TEST 02:10 OPUS [---] Non-Thinking First Result 02:20 OPUS [---] Thinking 05:58 OPUS 4.6"
YouTube Link 2026-02-05T23:15Z 85.2K followers, [----] engagements

"NEW GLM-5 vs MiniMax-2.5: NEW = BETTER Artificial Intelligence: Two new AI models (GLM-5 vs MiniMax-2.5) are tested - side by side - on a non-public causal reasoning test to evaluate their performance. Live recording of real-world performance of the latest agent optimized LLMs. 00:00 GLM-5 and MiniMax-2.5 02:13 Start Live TEST 12:35 GLM-5 and MiniMax-2.5 CRASH 14:07 First Solution by GLM-5 14:48 Successful GLM-5 Evaluation Run 16:33 GLM-5 Evaluation by MiniMax-2.5 18:03 MiniMax-2.5 FAILS Validation #aiexplained #aitech #ainews artificial intelligence AI models LLM VLM VLA Multi-modal model"
YouTube Link 2026-02-13T14:30Z 85.2K followers, [----] engagements

"First look at PyTorch Geometric: PyG [---] (Nov 2021) Discover together with me PyTorch Geometric (v2). PyG to code and train Graph Neural Networks (GNNs) for applications w/ graph structured data. Various methods for deep learning on graphs also known as geometric deep learning from a variety of published papers (see Arxiv for AI GNN GCN GAT and related pre-prints). #PyTorchGeometric #GraphNeuralNetwork #GeometricDeepLearning PyTorch Geometric (v2) on Graphs. PyG (PyTorch Geometric) is a library built upon PyTorch to easily write and train Graph Neural Networks (GNNs) for a wide range of"
YouTube Link 2021-11-30T07:30Z 85.1K followers, [----] engagements

"How To Transform VISION Tokens to a Language Vector Space A detailed exploration of the failure modes for Vision Language Models (Vision AI) focusing on the Connector module between vision and textual embedded spaces and their projections. All rights w/ authors: "Lost in Embeddings: Information Loss in VisionLanguage Models" Wenyan Li [--] Raphael Tang [--] Chengzu Li [--] Caiqi Zhang [--] Ivan Vulic [--] Anders Sgaard [--] from 1University of Copenhagen [--] Microsoft [--] University of Cambridge arXiv:2509.11986 #scienceexplained #visualai #vlm artificial intelligence AI models LLM VLM VLA Multi-modal model"
YouTube Link 2025-09-23T14:00Z 85.1K followers, [----] engagements

"The Eigenvectors of AI: Shared LoRA Subspaces for Continual Learning All rights w/ authors: Shared LoRA Subspaces for almost Strict Continual Learning Prakhar Kaushik* Ankit Vaidya* Shravan Chaudhari Rama Chellappa Alan Yuille from Department of Computer Science Johns Hopkins University Baltimore MD USA https://toshi2k2.github.io/share/ artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI multi-agent Fine-tune Pre-train RLHF AI Agent Multi-agent Vision Language Model Video AI artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory"
YouTube Link 2026-02-08T14:15Z 85.2K followers, [----] engagements

"Harvard Presents NEW Knowledge-Graph AGENT (MedAI) Harvard Unveils New Knowledge Graph Agent for improved AI in Medicine. Called KGARevion it combines the knowledge from knowledge graphs with the knowledge of LLMs. Since RAG suffers from inaccurate and incomplete retrieval problems in medicine Harvard et al present a new and improved methodology to significantly increase the reasoning performance of medical AI systems. Special focus on complex medical human interactions. New insights and new methods to combine the non-codified knowledge of LLMs with the structural codified knowledge of"
YouTube Link 2024-10-10T14:00Z 85.1K followers, 93.8K engagements

"MIT Invents Neuro-Symbolic LLM Fusion All rights w/ authors: "Teaching LLMs to Plan: Logical Chain-of-Thought Instruction Tuning for Symbolic Planning" Pulkit Verma MIT CSAIL Ngoc La MIT CSAIL Anthony Favier MIT CSAIL Swaroop Mishra Microsoft AI Julie A. Shah MIT CSAIL Cambridge USA #neuroscience #aiexplained #airesearch #logicalreasoning #logic artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI multi-agent Fine-tune Pre-train RLHF AI Agent Multi-agent Vision Language Model Video AI artificial intelligence AI models LLM VLM VLA Multi-modal model"
YouTube Link 2025-09-19T14:00Z 85.2K followers, 16.7K engagements

"How a 14B Model BEATS GPT-5.2 FUZZY Graph Reward All rights w/ authors: "Knowledge Graphs are Implicit Reward Models: Path-Derived Signals Enable Compositional Reasoning" Yuval Kansal Princeton University Niraj K. Jha Princeton University @princeton #ainews #RLVR #gpt52 #chatgpt5 #chatgpt Reinforcement Learning with verifiable return Medical AI Multi-step reasoning Reinforcement Learning with verifiable return Medical AI Multi-step reasoning"
YouTube Link 2026-01-27T14:30Z 85.2K followers, [----] engagements

"Python code to build your BPE - Tokenizer from scratch (w/ HuggingFace) Python TF2 code (JupyterLab) to train your Byte-Pair Encoding tokenizer (BPE): a. Start with all the characters present in the training corpus as tokens. b. Identify the most common pair of tokens and merge it into one token. c. Repeat until the vocabulary (e.g. the number of tokens) has reached the size you want. Training a tokenizer is not () the same as training a DL model. TensorFlow2 code: from tokenizers.trainers import BpeTrainer tokenizer.train(files trainer) Here the special case of a Byte-Pair Encoding (BPE)"
YouTube Link 2022-01-28T06:15Z 85.1K followers, [----] engagements

"Goodbye RAG - Smarter CAG w/ KV Cache Optimization Unleash the future of AI with Cache-Augmented Generation (CAG) Say goodbye to RAG retrieval delays and RAG errors - CAG preloads knowledge directly into large language models delivering lightning-fast accurate responses. CAG is the better RAG. Experience a streamlined architecture that outperforms traditional RAG methods. All rights w/ authors: Dont Do RAG: When Cache-Augmented Generation is All You Need for Knowledge Tasks Brian J Chan Chao-Ting Chen Jui-Hung Cheng and Hen-Hsen Huang National Chengchi University Taipei Taiwan and Academia"
YouTube Link 2024-12-30T15:00Z 85.1K followers, 59.2K engagements

"NEW Gemini [--] FLASH vs GPT [---] HIGH - A Bloodbath NEW Gemini [--] FLASH is [--] times cheaper ($) than OpenAI's GPT-5.2 HIGH for your identical tasks. So in a real-world test that looks similar to real science tasks I evaluate both AI models side-by-side. Note: This is not the known standard vanilla benchmarks this has to do with real world complexities - heavily oriented towards SCIENCE not Social Media. For a detailed comparison with all the other Ai models I have a dedicated YouTube Playlist https://www.youtube.com/playlistlist=PLgy71-0-2-F0Rla8lu5ZldpYQUfXM_5bT Will Gemini [--] FLASH reach only a"
YouTube Link 2025-12-18T13:30Z 85K followers, [----] engagements

"Death of the Token in AI: Multi-Parallel AI Reality NVIDIAs Silent Robots & Pre-GPT-6 Schrdinger Token Frankenstein Vector and Quantum collapse for new Multi-Parallel AI Realities & NVIDIA Silent Robots w/ Pre-GPT-6. Multi-Parallel realities for AI reasoning and Quantum collapse in AI for next generation of AI models and their advanced architectures. Combine these three brand new AI Arxiv pre-prints from the latest Ai research and you will see a pattern emerging for new AI architectures like GPT-6. From NVIDIA to Microsoft. All rights W/ authors: "Reasoning Beyond Chain-of-Thought: A Latent"
YouTube Link 2026-01-16T14:15Z 85.2K followers, [----] engagements

"MiMo V2 Flash: Excellent Performance (vs Kimi K2 Thinking) I test the causal reasoning performance for a simple [--] step logic task with the MiMo V2 Flash MoE 309B-15A model from Xiaomi (open-source). And compare the performance to a much bigger Kimi K2 Thinking Turbo model MoE 1T-32A. My Youtube playlist for this causal reasoning test is available here https://www.youtube.com/playlistlist=PLgy71-0-2-F0Rla8lu5ZldpYQUfXM_5bT 00:00 MiMo V2 vs KIMI K2 02:42 Start Live Test 08:56 MiMo First Result 10:15 1st Validation Run 12:37 2nd Validation Run 13:53 Kimi K2 First Answer 14:56 Self Check Kimi K2"
YouTube Link 2026-02-11T14:30Z 85.2K followers, [----] engagements

"Impressive: GLM [---] vs AIR: TEST What new AI model is better for reasoning: the big GLM [---] 355B-A32B or the smaller little brother model AIR A surprising result Different LLMs compared. NEW GLM [---] series examined. Other LLMs tested of the identical logic test are available https://www.youtube.com/watchv=FtrLaHeEP4E https://youtu.be/eo2QwyAItxI #logicalreasoning #airesearch #aimodel #aimodels #aiexplained artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI multi-agent Fine-tune Pre-train RLHF AI Agent Multi-agent Vision Language Model Video AI"
YouTube Link 2025-08-03T07:00Z 85.1K followers, [----] engagements

"World Model RAG: Generative Semantic Workspaces An ultra-modern RAG system with inherent world model creation and a structured spaciotemporal memory. We've all seen LLMs fail on long-form narratives their reasoning collapsing under the weight of "context rot" as standard RAG systems feed them a fragmented "bag of chunks." But what if instead of just retrieving facts an AI could construct a persistent episodic memory The Generative Semantic Workspace (GSW) paper proposes a groundbreaking neuro-inspired framework that does precisely that. It moves beyond fact retrieval to build a dynamic"
YouTube Link 2025-11-13T14:00Z 85.1K followers, [----] engagements

"Stanford: AI Agents DESTROY their Own Intelligence All rights w/ authors: Multi-Agent Teams Hold Experts Back Aneesh Pappu [--] Batu El [--] Hancheng Cao [--] Carmelo di Nolfo [--] Yanchao Sun [--] Meng Cao [--] James Zou [--] from [--] Stanford University [--] Goizueta Business School Emory University [--] Apple. #airesearch #aiexplained #artificialintelligence #multiagentsystems artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI multi-agent Fine-tune Pre-train RLHF AI Agent Multi-agent Vision Language Model Video AI artificial intelligence AI models LLM VLM VLA Multi-modal"
YouTube Link 2026-02-09T14:15Z 85.2K followers, [----] engagements

"AI Belief Functions: Deciding Under Absolute Uncertainty Engines of Intelligence: The Definition and Necessity of AI Agents. Basic mathematical explanations for any CEO of any multinational strategy and management consulting firm: What AI agents are and why we need them. And why AI (re-)acts in absolute uncertainty with self-updating belief functions. Plus: A simple explanation for CEOs (McKinsey EY .) what are AI agents and why they operate in absolute uncertainty with a simple mathematical probability distribution for the most difficult client jobs. .maybe not such a good idea #aiexplained"
YouTube Link 2026-02-14T14:00Z 85.2K followers, [----] engagements

"DeepSeek built a New Topological Transformer (mHC) DeepSeek build a new topological transformer that is beautifully compatible with the new Transformer architecture from Google (see my video https://youtu.be/5gpc3d2rFlg ). In this video I explain the mathematics of the Manifold-Constrained Hyper-Connections in particular the Sinkhorn-Knopp algorithm to entropically project H_residual onto the Birkhoff polytope. All rights w/ authors: "mHC: Manifold-Constrained Hyper-Connections" Zhenda Xie* Yixuan Wei* Huanqi Cao* Chenggang Zhao Chengqi Deng Jiashi Li Damai Dai Huazuo Gao Jiang Chang Liang"
YouTube Link 2026-01-03T14:00Z 85.2K followers, 17.2K engagements

"New Paradigm: Single Layer AI For a decade we believed that 'Deeper is Better.' We thought that to learn a new task or generate a new image we had to train massive deep architectures end-to-end. We were wrong. Today I present two papers that prove we have entered a new paradigm: Single Layer AI. One paper proves that our Frozen Backbones are geometrically perfect: they don't forget only the last layer gets confused. The second paper proves that these backbones are semantically perfect: we can drive state-of-the-art image generation with just a single layer of adaptation. The era of old type"
YouTube Link 2025-12-14T14:00Z 85.1K followers, [----] engagements

"The Eigenvector of Multi Agent Systems w/ RAG EIGEN-1 is a highly efficient agentic framework that replaces explicit tool calls with an implicit on-stream Monitor-Querier-Injector pipeline for tax-free RAG. It swaps democratic multi-agent aggregation for Hierarchical Solution Refinement (HSR) a structured protocol using rotating anchor-reference assignments for targeted peer-informed solution repair. The process is governed by Quality-Aware Iterative Reasoning (QAIR) an adaptive control loop that uses quality-score thresholds to selectively refine weak solutions and ensure efficient"
YouTube Link 2025-09-27T14:00Z 85.1K followers, [----] engagements

"Moltbook: The First AI Civilization (Clawdbot to OpenClaw) Finally an "AI haven" for super-intelligences to self-replicate and build their own social hierarchies while we (human) watch from the sidelines - and get a good look at those superintelligences (smile). OpenClaw recommends CLAUDE OPUS [---] for best results. After the staggering global success of Facebook and the intellectual olympus of Reddit finally we created a place for super -intelligent AI Agents to come together and learn. Clawdbot Moltbot or their latest identity-crisis rebranding OpenClaw are providing deep insights into their"
YouTube Link 2026-02-01T11:25Z 85.2K followers, [----] engagements

"The Illusion of Intelligence in AI (Harvard MIT) Latest AI research papers regarding the reasoning performance of AI in real-world tests. Medical and Safety applications. GPT-5 Mini GPT [--] Nano GPT-5. All rights w/ authors: "Why Chain of Thought Fails in Clinical Text Understanding" Jiageng Wu [--] Kevin Xie [--] Bowen Gu [--] Nils Krger [--] Kueiyu Joshua Lin [--] Jie Yang [--] from [--] Harvard Medical School [--] MIT [--] Broad Institute of MIT and Harvard "CAN AI PERCEIVE PHYSICAL DANGER AND INTERVENE" Abhishek Jindal [--] Dmitry Kalashnikov [--] Oscar Chang [--] Divya Garikapati [--] Anirudha Majumdar [--] Pierre Sermanet 1"
YouTube Link 2025-09-30T14:00Z 85.1K followers, [----] engagements

"AI Latent Space Surgery: The End of Fine-Tuning Orthogonal Subspaces in AI encode . personality alignment or more entangled complexities by placing additional attention heads on the transformer architecture. The evolution of Large Language Models (LLMs) is shifting from the pursuit of general-purpose reasoning to the creation of specialized coherent agents. Whether for immersive role-playing in open-world environments or empathetic engagement in therapeutic settings the utility of an AI agent increasingly depends on its ability to maintain a stable distinct psychological profile. However"
YouTube Link 2025-12-10T14:15Z 85.1K followers, [----] engagements

"LOGIC Test on Qwen3 Max Thinking and Kimi K2.5 I perform my causal reasoning performance tests on the new Qwen [--] MAX THINKING and analyse the new KIMI K2.5 with added visual intelligence on top of the Base K2 model with a nice new marketing feature of [---] subagents under an orchestrator as a special model variance. Live test on Alibaba Cloud. I like the information provided by John Hammond on "Clawdbot Malware" https://www.youtube.com/watchv=7GS6Xs4hdvg and for some additional info https://www.youtube.com/watchv=kSno1-xOjwI (nothing to add from my side). #aitesting #chatgpt #scienceexplained"
YouTube Link 2026-01-29T14:15Z 85.2K followers, [----] engagements

"256K Context Window Forget It Needle-in-a-haystack served for a long time as a tool to show the performance of the long context window of LLMs (eg 128K token length): the perfect base for ICL multi-step reasoning and RAG. Unfortunately is was not the right benchmark. A disaster for RAG - in any form. A new pre-print published today present a more powerful instrument to explore the Long Context reasoning window of the latest LLMs. And the long context window's reasoning performance is a disaster even for current LLMs and LRMs. All rights w/ authors: "NEEDLECHAIN: MEASURING INTACT LONG-CONTEXT"
YouTube Link 2025-07-31T13:01Z 85.1K followers, [----] engagements

"Wavelet Operator Theory: Beyond GPT-5 (#startup) Wavelet Operator Networks: AI Beyond Transformers (like GPT-5 or Claude 5). The next generation of operator-based machine intelligence. The dominant paradigm of deep learning relies on optimizing millions of parameters in Euclidean space often leading to opaque computationally expensive models. This video explores a radical alternative: recasting machine learning as operator estimation in infinite-dimensional Hilbert spaces. We will demonstrate how replacing the Transformer's O(n) self-attention mechanism with the linear-time O(n) Wavelet"
YouTube Link 2025-08-05T14:01Z 85.1K followers, [----] engagements

"8B outperforms GPT-120B on Multi Agents All rights w/ authors: DyTopo: Dynamic Topology Routing for Multi-Agent Reasoning via Semantic Matching Yuxing Lu * [--] [--] Yucheng Hu * [--] Xukai Zhao [--] Jiuxin Cao [--] from [--] Peking University Beijing China [--] Georgia Institute of Technology Atlanta United States [--] Southeast University [--] Tsinghua University. #aiexplained #airesearch #artificialintelligence #aiagents artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI multi-agent Fine-tune Pre-train RLHF AI Agent Multi-agent Vision Language Model Video AI artificial"
YouTube Link 2026-02-10T14:01Z 85.2K followers, [----] engagements

"Distill [--] AI Agents into ONE (w/ CODE) Optimize your AI compute budget: Distill a complete complex multi-agent intelligence into a single agent. Three different distillation techniques presented and their performance benchmark discussed for distillation to a small LLM for local use. All rights w/ authors: AgentArk: Distilling Multi-Agent Intelligence into a Single LLM Agent Yinyi Luo1 Yiqiao Jin3 Weichen Yu1 Mengqi Zhang2 Srijan Kumar3 Xiaoxiao Li5 Weijie Xu4 Xin Chen4 Jindong Wang2* from [--] Carnegie Mellon University [--] William & Mary [--] Georgia Institute of Technology [--] Amazon [--] University of"
YouTube Link 2026-02-07T14:01Z 85.2K followers, [----] engagements

"Forget LLM: MIT's New RLM (Phase Shift in AI) Weve been misled by the promise of "infinite" context windows: new AI research proves that "Context Rot" is destroying reasoning capabilities as inputs scale. But a groundbreaking paper from MIT introduces a radical solution: Recursive Language Models (RLMs). Instead of blindly force-feeding data into a single Transformer RLMs act as a Neurosymbolic Operating System writing Python code to mechanically split massive datasets and recursively "spawn" fresh model instances to process them. The result is a staggering leap in performance: on quadratic"
YouTube Link 2026-01-04T14:01Z 85.2K followers, 30K engagements

"MiniMax M2.1 vs GLM [---] - TEST AI We have two new Ai models and I test both new models on [--] causal reasoning tests. First is my [--] sentence test and second is my "elevator test" where you find a complete YouTube playlist here Minimax [---] GLM [---] 00:00 New AI M2_1 vs GLM 4_7 03:30 One Sentence test 13:18 Validation run One Sentence 18:17 Elevator Test (2_1 vs GLM 4_7) 25:54 Validation run elevator 29:01 MiniMax M2_1 Platform run #aitesting #aireasoning #science #artificialintelligence #chatgpt artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI"
YouTube Link 2025-12-25T13:45Z 85.2K followers, [----] engagements

"15B Active MoE BEATS OPUS [---] in Reasoning Inside AI: to be specific inside a real powerful reasoning engine MoE and all the new methods and optimizations algorithms we encounter in building an open-source Mixture-of-Expert AI model. Inside a modern MoE AI model (Technical Architecture) All rights w/ authors: MiMo-V2-Flash Technical Report LLM-Core Xiaomi https://arxiv.org/pdf/2601.02780 #aitransformation #aiexplained #scienceexplained #nextgenai #airesearch artificial intelligence AI models LLM VLM VLA Multi-modal model explanatory video RAG multi-AI multi-agent Fine-tune Pre-train RLHF AI"
YouTube Link 2026-02-12T14:01Z 85.2K followers, [----] engagements

"CODE: GRAPH Link Prediction w/ DGL on Pytorch and PyG Code Example GraphML GNN For Graph ML we make a deep dive to code LINK Prediction on Graph Data sets with DGL and PyG. We examine the main ideas behind LINK Prediction and how to code a link prediction example in PyG and DGL - Deep Graph Library. DGL - Easy Deep Learning on Graphs with framework agnostic coding (either PyG or TensorFlow2). A GNN-based link prediction model represents the likelihood of connectivity between two nodes u and v as a function of their node representation computed from the multi-layer GNN. Training a link"
YouTube Link 2022-12-17T13:00Z 85.2K followers, 17K engagements

"Neurosymbolic 80M AI from Princeton beats GPT Curious about the future of neurosymbolic AI regarding verifiable reasoning Dive into the groundbreaking GraphMERT framework a game-changer from Princeton researchers that distills high-quality KGs from unstructured data with unprecedented efficiency and scalability. In this video well unravel how GraphMERT blends neural learning with symbolic reasoning in one new transformer system (without any GPT) delivering factual (69.8% FActScore) and valid (68.8% ValidityScore) KGsoutshining LLM baselines (40.2% and 43.0%). Discover why its poised to"
YouTube Link 2025-10-16T12:01Z 85.1K followers, 23.7K engagements

"The Truth about AI is Devastating: Proof by MIT Harvard AI Superintelligence ASI with the new LLMs like GPT5 Gemini [--] or newly released Grok4 Forget about it GROK4 will discover new Physics Dream on. Harvard Univ and MIT provide new evidence of the internal thoughts and world models of every AI architecture from Transformer to RNN to LSTM to Mamba and Mamba [--]. LLMs are Just Faking It: New Proof by MIT Harvard. Harvard & MIT's New Proof: LLMs Aren't Intelligent. Just pattern matching machines. The truth about AI is devastating. And provides long and fascinating research trajectories into our"
YouTube Link 2025-07-12T12:01Z 79K followers, 67.8K engagements

"NEW Multi-Modal AI by APPLE Apple published new Machine Learning (ML) models on its GitHub repo: 4M-21. Massively Multimodal Masked Modelling. All rights w/ authors: 4M-21: An Any-to-Any Vision Model for Tens of Tasks and Modalities https://arxiv.org/pdf/2406.09406 Video from Apple and Lausanne: https://storage.googleapis.com/four_m_site/videos/4M-21_Website_Video.mp4 #appleai #apple #multimodalai"
YouTube Link 2024-06-22T14:00Z 76.8K followers, [----] engagements

"AI [--] Science FINALLY Latest Insights All rights w/ authors: "On LLM-Based Scientific Inductive Reasoning Beyond Equations" Brian S. Lin [--] Jiaxin Yuan [--] Zihan Zhou [--] Shouli Wang [--] Shuo Wang [--] Cunliang Kong [--] Qi Shi [--] Yuxuan Li [--] Liner Yang [--] Zhiyuan Liu [--] Maosong Sun [--] from [--] Dept. of Comp. Sci. & Tech. Institute for AI BNRist Center Tsinghua University Jiangsu Collaborative Innovation Center for Language Ability Jiangsu Normal University [--] Beijing Language and Culture University [--] Xiamen University [--] Harbin Institute of Technology @stanford @TsinghuaUniversity_official @UCBerkeley"
YouTube Link 2025-09-24T14:01Z 79.5K followers, [----] engagements

"Dirichlet Energy Minimization Explains In-Context Learning (Harvard) Remember the A in RAG The Augmentation of your query with retrieved external data In this amazing video: New insights into the Augmentation process of RAG which is also an In-Context Learning (ICL) of the LLM. But when will the augmented prompt and its new knowledge override the semantic prior of the LLM How can we explain this process and the emergence of a phase transition in the learning performance of the LLM Here you find the answers provided by a new AI research preprint by Harvard University. New research for an"
YouTube Link 2025-01-05T15:00Z 52.4K followers, [----] engagements

"NVIDIA's lack of an AI PC - Good for Investors Current open source LLMs need between [---] GB (for Databricks DBRX 16-bit) down to 100GB (new JAMBA LLM) either VRAM or shared memory between CPU GPU and NPU (AI accelerator). Current consumer NVIDIA GPU top out at 24GB (4090) and 40GB A100 are financially out of scope for customer implementing their LLM trained on their company data on-site. Now a new market segment shows a response: Intel announced a new AI PC in the last days therefore we analyze technical infrastructure options from cloud based NIM to local PCs running open source LLMs for"
YouTube Link 2024-04-01T12:00Z 30.6K followers, [---] engagements

"GROK [--] on Logic Real TEST - PART [--] ELON says GROK [--] is not yet fully optimized for reasoning NO PROBLEM - We'll FIX IT We'll optimize causal reasoning of GROK [--] right now right here. 2nd part of my video where I test the causal reasoning performance of GROK [--]. Multiple runs check the GROK [--] internal assumptions and boundary conditions imposed by the system itself and give it more time to re-run multiple times to find the best solution. MY first video is available here https://youtu.be/aobihG5ig28 Reference: PASS@10 #aitesting #aiexplained #grok4 #grokai #test"
YouTube Link 2025-07-11T14:01Z 75.6K followers, [----] engagements

"Game Theory: Omni superior to Strawberry in Strategy In my game theoretical experiment to explore the causal reasoning capabilities of GPT-4omni (in short: o) versus advanced reasoning by OpenAI's o1 (short: o1) I introduce a simple cascading prompt - hoping to simulate to a minimal extend - an inference based multi CoT response (like o1) for any LLM. In my theoretical game I explore a game based defense strategy that is a bit more complicated than normal game implementations and I use a vanilla GPT-4omni model to design and create new strategic game options. The focus of this video is on"
YouTube Link 2024-09-16T14:00Z 44.7K followers, [----] engagements

"NEW AI generates Video Games: GENIE AI explained Generative Interactive Environments - GENIE: After SORA the next step is interactive video environments: Synthetic world AI systems. Technology: BERT VQVAE - Vector Quantized Variational AutoEncoders Text-to-Video novel video tokenizer subobject tokenizer Subobject-level tokenization MaskGIT - Masked Generative Image Transformer Latent Action Model (LAM) Dynamics model Spatial-Temporal Transformer (ST-Transformer) Vision Transformers (ViT) SORA V-JEPA World Models Video Models GENIE virtual video world generation. Interactive and controllable"
YouTube Link 2024-02-28T11:30Z 54K followers, [----] engagements

"SDXL explained w/ shape-shifting ControlNet (Stanford Univ) Stable Diffusion XL is the latest version of Stable Diffusion for Text-to-Image Diffusion models (T2I). Theory and code of SDXL explained. Plus: The latest ControlNet NN are amazing for shape shifting your SDXL synthetic images. Tutorial regarding the Theory and Code of SDXL explained w/ ControlNet. Plus ControlNet code for a stand alone version ( . it does not always has to be Automatic [----] or any other Web UI). #sdxl #aieducation #tutorialyoutube"
YouTube Link 2023-09-14T12:00Z 45K followers, [----] engagements

"How to build Skynet = Adaptive Multi AI Agents Building Adaptive Multi-Agent AI: A Skynet Simulation. Based on new AI research insight from Peking University et al. See new arxiv pre-print below. AdaSociety introduces a novel adaptive environment that combines dynamic physical and social structures for multi-agent reinforcement learning (RL). The environment allows agents to modify both their surroundings and their social connections creating a co-evolving system of tasks and alliances. By integrating expanding physical spaces and multi-layered alterable social connections AdaSociety presents"
YouTube Link 2024-11-10T15:00Z 47.2K followers, [----] engagements

"Self-instruct fine-tuning of LLMs (Alpaca) : The Introduction Fine - tuning and "INSTRUCTION fine-tuning" your LLM has significant advantages. "Instruct fine-tuning" can be a powerful technique for improving the performance of language models particularly for tasks where the input data has a specific structure or format. By providing the model with guidance on how the input data is structured we can help the model better understand the relationships between different parts of the input and improve its ability to make accurate predictions. It's important to note that "instruct fine-tuning""
YouTube Link 2023-04-08T15:15Z 65.5K followers, 15.8K engagements

"Watch a Critical AI Failure Watch a critical AI failure mode unfold: From my personal experience that sometimes I do not formulate my prompt precise enough. AI (see video): "I failed to follow my own core instructions" . ". my performance was unacceptable. This was a clear example of a critical failure mode for an AI ." No lessons to be learned . smile. #aifails #failure #airesearch #aiagents #aifuture #dspy"
YouTube Link 2025-07-24T14:01Z 74.7K followers, [----] engagements

"GPT 5: Should You Use It (LIVE TEST) OpenAI released the new GPT-5 model and I performed an extreme causal reasoning test on it. #gpt5 #gpt #chatgpt #test #performance"
YouTube Link 2025-08-07T20:00Z 77.6K followers, [----] engagements

"Multi-Token Prediction (forget next token LLM) Meta published a new method of multi-token prediction for autoregressive transformer models (LLMs). Additional heads perform in parallel token predictions. Benchmark data investigated and a special session for my green grasshoppers Instead of sequentially predicting the next token based on previously observed tokens this architecture employs multiple output heads that operate in parallel from a shared trunkthe main body of the model which processes the input and generates a common latent representation. Each output head predicts a different"
YouTube Link 2024-05-02T12:00Z 45.2K followers, [----] engagements

"OpenAI o3 mini high: EXCELLENT FRUSTRATING LIVE LOGIC TESTS performed on the newly available o3 mini high model by OpenAI. The positive side of o3 mini high and its massive limitations exposed in my new video. A commercial AI model with almost NO VALUE to the open-source community. My extreme logic test that I developed for Strawberry (o1) you can find here: https://www.youtube.com/watchv=tpun1uOKecc #chatgpt #airesearch #o3"
YouTube Link 2025-02-01T09:02Z 71.6K followers, [----] engagements

"AMD on HuggingFace ROCm [---] - MI300X - Software AI Hugging Face and AMD partner on accelerating HF state-of-the-art models for CPU and GPU on AMD platforms. https://huggingface.co/blog/huggingface-and-amd https://www.amd.com/en/newsroom/press-releases/2023-6-13-amd-expands-leadership-data-center-portfolio-with-.html https://www.reuters.com/technology/amazons-cloud-unit-is-considering-amds-new-ai-chips-2023-06-14/ https://www.amd.com/en/graphics/servers-solutions-rocm #amd #huggingface #cuda #ai #inference"
YouTube Link 2023-06-17T05:15Z 47.3K followers, [----] engagements

"125 arxiv pre-prints (full text): EXTRACT CONTENT clusters in 3D visualization You have pytorch running on your JupyterLab downloaded a BERT model from Huggingface and designed a sentence transformer. To improve your content extraction of eg. [---] arxiv pre-prints in FULL TEXT: take the extra effort to delete all references in this full text. Since references are standardized thousands of them will "disturb" your content extraction if you are interested specifically in the research/scientific content of all those arxiv preprints. A simple python routine will do the job. #arxiv #nlproc"
YouTube Link 2021-04-30T12:45Z 64.1K followers, [---] engagements

"When Smart AI Models Overthink Stupid Data (AI TRAP) The authors of the latest AI research study find that the response length of reasoning LLMs whether trained by reinforcement learning or supervised learning drastically increases for ill-posed questions with missing premises (MiP) ending up with redundant and ineffective thinking. This newly introduced scenario exacerbates the general overthinking issue to a large extent which the authors name as the MiP-Overthinking. This implies a critical flaw of the current training recipe for reasoning LLMs which does not encourage efficient thinking"
YouTube Link 2025-04-12T14:01Z 74.3K followers, [----] engagements

"Why Are These AI Agents Continuously Failing Stay away from the following LLMs for multi-turn MCP agents: a list curated by Duke University. New AI research regarding MCP powered Agents . All rights w/ authors: LIVEMCP-101: STRESS TESTING AND DIAGNOSING MCP-ENABLED AGENTS ON CHALLENGING QUERIES by Ming Yin [--] Dinghan Shen [--] Silei Xu [--] Jianbing Han [--] Sixun Dong [--] Mian Zhang [--] Yebowen Hu [--] Shujian Liu [--] Simin Ma [--] Song Wang [--] Sathish Reddy Indurthi [--] Xun Wang [--] Yiran Chen [--] Kaiqiang Song [--] from [--] Duke University [--] Zoom Video Communications #aitesting #airesearch #scienceexplained"
YouTube Link 2025-08-23T14:01Z 77.7K followers, [----] engagements

"AI Just Outsourced It's Own Thinking (Stanford) Ai systems especially multi-agent systems outsource their complex planning phase with advanced thinking and complex reasoning. It is simply to expensive for the AI to do the thinking all by itself again and again. Cache the complete PLAN template for each and every complex query and reuse it. Reduce costs up to 50%. all rights w/ authors: "Cost-Efficient Serving of LLM Agents via Test-Time Plan Caching" Qizheng Zhang [--] Michael Wornow [--] and Kunle Olukotun [--] from [--] Department of Computer Science Stanford University [--] Department of Electrical"
YouTube Link 2025-06-21T13:45Z 74.5K followers, [----] engagements

"GenAI WORMS Breach RAG Systems: AI Cybersecurity AI Cybersecurity: With Generative AI a new cybersecurity threat emerged: GenAI Worms. To protect your AI systems we analyze the design and function of AI Cyber Worms and other current cybersecurity threats to protect RAG and LLM systems. AI Cybersecurity Countermeasures. #cybersecurity #airesearch #newtechnology #ai"
YouTube Link 2024-03-04T13:00Z 45.4K followers, [----] engagements

"Today's AI NEWS: [--] NEW AI Papers - Sept [--] [----] [--] NEW AI Papers: The latest AI News and tech insights into AI. New AI methods new code. From NVIDIA w/ physics engine to MIT to Tsinghua Univ to Caltech to Berkeley. New AI insights. My personal selection of the most important AI research papers and pre-prints of today a rainy Tuesday September [--] [----]. All links of all presented papers are available in my community tab of this channel. #news #ai #coding"
YouTube Link 2024-09-24T12:00Z 44.8K followers, [----] engagements

"Many-Shot VISUAL ICL is amazing (Stanford) Many-shot visual in-context learning (ICL) is amazing Especially when working with ICL+ (1 mio token context length) like Gemini [---] Pro also tested already for GPT-4o. An amazing alternative to fine-tuning VLM and LLMs. New study by Stanford Univ shows the potential of new long context VLMs also in regard to visual information (images). Tests include up to [----] images in a prompt with batched queries and the models perform Multimodal Many-shot in-context learning for extreme context lengths (1 mio token and more) tested for complete length of"
YouTube Link 2024-05-21T12:01Z 60.9K followers, [----] engagements

"AI Love vs AI Logic Your AI Girlfriend is waiting Did I missed out on the most significant market for AI Did I missed out on the most profitable market for AI on this planet Does AI offer an emotional dimension for human beings that are not grounded in facts and logic Will AI be part of our emotional life AI guiding us in our mental support influence us what we have to think what products to buy how to see the world AI programmed by global corporations The TOP [--] LLMs and LMMs of this planet answer. Interesting video to watch on YouTube by @aini_ : The AI future is here: love death and the AI"
YouTube Link 2024-07-08T12:00Z 43.1K followers, [----] engagements

"NP-Hard: The End of AI The most fascinating AI publications of today - as published on Thursday [--] August [----]. What are the hardest Ai publications of today Smile. Are the latest AI computations as published today in pre-prints and journals and selected by me in this video: NP-hard or NP-complete Some AI problems might even more complex than NP-complete problems like: a. PSPACE-complete Problems: Problems that require polynomial space to solve like certain types of logic puzzles or games (e.g. determining the winner in generalized chess). b. EXPTIME Problems: Problems that require"
YouTube Link 2024-08-15T12:01Z 44.6K followers, [----] engagements

"FLAN-T5-XXL on NVIDIA A100 GPU w/ HF Inference Endpoints let's explore 11b models Easy Cloud Inference Today I discover a new Flan-T5-XXL model repository on Huggingface which can run (optimized) on a NVIDIA A10G. Or run Google's Flan-T5-XXL on A100 GPU. PLUS: First time discovery of Huggingface's Inference endpoints What are Inference endpoints by HF: a fully managed cloud compute infrastructure (eg AWS AZURE later GOOGLE) where I can use my HuggingFace repositories from any TRANSFORMER or Sentence-Transformer model and run directly a cloud compute A new milestone for easy inference of 11b"
YouTube Link 2023-02-01T13:00Z 62.9K followers, [----] engagements

"The AI Reasoning Lie The reasoning lie regarding Large Language Models (LLMs). Including extended reasoning of CLAUSE Sonnet [---] or other thinking models. Insights from this video apply to classical LLMs and Test-Time-Compute Scaling TTS models like o1 o3 and so on. Latest research uncovers that there is no inherent logical emergence of intelligence in AI systems. All rights w/ authors: Order Doesnt Matter But Reasoning Does: Training LLMs with Order-Centric Augmentation Qianxi He1 Qianyu He1 Jiaqing Liang2 Yanghua Xiao1 Weikang Zhou3 Zeye Sun3 Fei Yu3 from [--] Shanghai Key Laboratory of Data"
YouTube Link 2025-03-03T15:01Z 65.3K followers, 31.4K engagements

"PEFT w/ Multi LoRA explained (LLM fine-tuning) Parameter Efficient Fine-Tuning of LLM w/ multiple LoRA Adapters. A deep dive to understand LoRA (low rank adaptation) and its possible configurations including the [--] LoRA_config parameters for parameter efficient fine-tuning. Switch between different PEFT adapters and activate or deactivate added PEFT LoRA Adapters to a pre-trined LLM or VLM. PEFT - LoRA explained in detail. Matrix factorization Singular value decomposition (SVD). How to add multiple PEFT-LoRA adapters together into one adapter. #ai #research #codegeneration"
YouTube Link 2023-11-07T13:00Z 75.6K followers, [----] engagements

"Transfer Reasoning to Llama [---] (Chess AI) Distilling Thought: Transferring Search-Based Reasoning to Small LMs. The Incredible Shrinking Mind: Is AI Reasoning Lost in Transfer All rights w/ authors: ASTRO: Teaching Language Models to Reason by Reflecting and Backtracking In-Context Joongwon Kim [--] Anirudh Goyal [--] Liang Tan [--] Hannaneh Hajishirzi [--] Srini Iyer [--] Tianlu Wang [--] from [--] AI at Meta [--] University of Washington Can Large Language Models Develop Strategic Reasoning Post-training Insights from Learning Chess Dongyoon Hwang [--] Hojoon Lee [--] Jaegul Choo [--] Dongmin Park [--] Jongho Park [--] from 1"
YouTube Link 2025-07-03T14:01Z 72.2K followers, [----] engagements

"o3 Inference Reasoning: How to Build the Training Data Set Technical deep dive on OpenAI's o3 reasoning implementation in the SFT RL and inference time compute reasoning structure. Transferring o3 Training Procedures to my 7B LLM first step: how-to build the specific training data sets. In this video: How to build the training dataset for this new reasoning engine like o3-mini. How to achieve o3 reasoning performance with new SFT and RL aligned training procedures for CoT during inference reasoning Based on the documentation by @OpenAI Deliberative Alignment: Reasoning Enables Safer Language"
YouTube Link 2024-12-23T16:00Z 49.3K followers, [----] engagements

"Graph Topology secures Wall Street's AI Agents Can Graph Topology secure Wall Street's AI Agents and multi-agent systems How secure are Ai agents in the financial service sector How to defend against malicious attacks by AI Counteract Memory specific attack vectors and prompt injection. How-to protect your multi-agent AI system. Latest Ai research. To keep your AI safe and protected. all rights w/ authors: "G-Safeguard: A Topology-Guided Security Lens and Treatment on LLM-based Multi-agent Systems" Shilong Wang Guibin Zhang♠ Miao Yu Guancheng Wan♣ Fanci Meng Chongye Guo♦ Kun Wang Yang Wang"
YouTube Link 2025-05-03T14:01Z 70.5K followers, [----] engagements

"Optimize the LLM Action Planning Space w/ ICL (Google) New #ai Action Planning via In-Context Learning (ICL): Planning is essential for artificial intelligence systems to look ahead and proactively determine a course of actions to reach objectives in the virtual and real world. ICL could prove to be real powerful instrument for improved Action sequence planning of agentic systems. All rights w/ authors: IMPROVING LARGE LANGUAGE MODEL PLANNING WITH ACTION SEQUENCE SIMILARITY Xinran Zhao12 Hanie Sedghi1 Bernd Bohnet1 Dale Schuurmans1 Azade Nova1 from [--] Google DeepMind [--] Carnegie Mellon"
YouTube Link 2025-05-06T14:01Z 79.3K followers, [----] engagements

"Graph RAG Evolved: PathRAG (Relational Reasoning Paths) The latest tech dev of RAG. Overcoming the limitations of vector RAG GraphRAG showed significant improvements and LightRAG implemented an Index Graph (Knowledge Graph) to further improve the overall performance. The latest stage of RAG dev is now PathRAG dev by Beijing University. A detailed explanation of the Graph RAG models with detailed explanation of the inner workings of PathRAG - with their code implementation (GitHub). All rights w/ authors: "PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths" Boyu"
YouTube Link 2025-02-26T15:01Z 76.1K followers, 28.9K engagements

"AI Computer Use: Why we need a REWARD VLM (ARMAP) Explore self-generating AI VISION REWARD Models (VLM) with advanced RL (Reinforcement Learning) algos for deep reasoning complex tasks in multi-AI-Agent systems. New AI research. New robotics algos and insights to optimize "computer use" by AI systems. With detailed explanations - for beginners and experts. Develop new AI code and configurational reasoning systems. All rights w/ authors: ARMAP: SCALING AUTONOMOUS AGENTS VIA AUTOMATIC REWARD MODELING AND PLANNING Zhenfang Chen MIT-IBM Watson AI Lab Delin Chen UMass Amherst Rui Sun University of"
YouTube Link 2025-02-19T15:01Z 74.3K followers, [----] engagements

"1984 for AI: Real-Time Thought Correction in AI Agents Think Safe Act Smart: Enhancing AI Agent Safety in Real Time. AI Agent Thought Corrections. Rethink and React: Making AI Agents Safer. All rights w/ authors (Arxiv preprint): Think Twice Before You Act: Enhancing Agent Behavioral Safety with Thought Correction Changyue Jiang [--] Xudong Pan [--] Min Yang [--] from [--] Fudan university [--] Shanghai Innovation Institute https://huggingface.co/fgdrg/Thought-Aligner-7B-v1.0 #airesearch #ainews #scienceexplained #reasoning #aiagents"
YouTube Link 2025-05-20T14:15Z 66.9K followers, [----] engagements

"Multi Agents AI: Simple Code w/ Smolagents How to code multi AI-Agent systems for complex problems with the new smolagents framework. Simple code implementation. Massive complexities (from OpenAI and Anthropic) eliminated. Decentralize AI Intelligence with new Agent framework: smolagents. See my last video for my introduction to Smolagents (with references) https://www.youtube.com/watchv=tBXAwCH6rcU #aiagents #coding #airesearch #huggingface #smolagents"
YouTube Link 2025-01-03T15:00Z 51.9K followers, [----] engagements

"Anthropic's Secret: How we Build Multi-Agent AI In my video I follow the instructions by Anthropic on How-To build multi-agent research systems given their specific GitHub repo. And I optimize the ideas of Anthropic for a better performance. Anthropic's Architects: Prompting a Mind (and how to improve it). All rights w/ authors: Anthropic "How we built our multi-agent research system" https://www.anthropic.com/engineering/built-multi-agent-research-system #airesearch #ainews #anthropic #aiexplained #scienceexplained"
YouTube Link 2025-06-16T14:01Z 75.7K followers, [----] engagements

"AI Math explained - the easy way A new video tutorial for AI mathematics. Focus on Reinforcement Learning methods especially for the latest meta-thinking multi-agent reinforcement learning (RL). Math AI explained [--] Agent MARL (REMA) All rights w/ authors on REMA topic - regarding the new pre-print : REMA: LEARNING TO META-THINK FOR LLMS WITH MULTI-AGENT REINFORCEMENT LEARNING Ziyu Wan [--] Yunxiang Li [--] Yan Song [--] Hanjing Wang [--] Linyi Yang [--] Mark Schmidt [--] Jun Wang [--] Weinan Zhang [--] Shuyue Hu [--] Ying Wen [--] from [--] Shanghai Jiao Tong University [--] University of British Columbia [--] University College"
YouTube Link 2025-03-14T15:01Z 61.9K followers, [----] engagements

"Open Source AI Under the Shadow of META Manipulating Narratives: How META Could Be Influencing Public Views on Open Source AI. And why We focus on META's formation of an AI alliance and its advocacy for open-source platforms despite its primary income being derived from advertising. This situation creates a paradox particularly in light of META's control over access to its AI models such as the Lama [--] model which is ostensibly open-source but requires META's permission for use. We then contrasts META's self-proclaimed leadership in the open-source AI domain with its actual performance in AI"
YouTube Link 2023-12-07T13:00Z 45K followers, [----] engagements

"NEW Multi-Agent CODE explained (by OpenAI) New Multi-Agent orchestration by OpenAI. Code based video with detailed explanations. This video revolves around the design and implementation of a multi-agent system for managing user interactions in tasks such as customer service sales and support. The core concept involves defining routines which are structured sequences of instructions for handling specific workflows. Each routine consists of a system message that outlines the steps the agent must follow such as asking probing questions proposing solutions or offering refunds. To enable dynamic"
YouTube Link 2024-10-19T14:00Z 79K followers, 17.5K engagements

"Today's AI NEWS: [--] Reports - September [--] [----] My personal selection of the most important AI research papers and pre-prints for Monday September [--] [----]. All links of all presented papers are available in my community tab of this channel. #news #ai #coding"
YouTube Link 2024-09-23T08:15Z 44.8K followers, [----] engagements

"Good LLMs need BAD Data: The Shocking Truth by HARVARD Good LLMs need BAD Data: The Shocking Truth by HARVARD Li et al. provide compelling evidence that a paradigm shift in how we approach pre-training data curation - from aiming for maximal purity to strategic inclusion of "bad data" - can lead to LLMs whose undesirable behaviors are more easily and effectively mitigated post-training. This offers a new avenue for building more robustly aligned and controllable AI systems. all rights w/ authors: When Bad Data Leads to Good Models Kenneth Li Yida Chen Fernanda Viegas Martin Wattenberg from"
YouTube Link 2025-05-11T14:01Z 66.3K followers, [----] engagements

"Direct Preference Optimization: Forget RLHF (PPO) DPO replaces RLHF: In this technical and informative video we explore a groundbreaking methodology called direct preference optimization (DPO) by Stanford Univ that has the potential to replace reinforcement learning in the training of GPT systems. Join us as we dive into the intricacies of direct preference optimization dissecting its technical details and highlighting its advantages over the conventional reinforcement learning approach. Discover how this innovative technique opens new possibilities in AI training offering more precise"
YouTube Link 2023-06-06T12:00Z 62.6K followers, 16K engagements

"NEW CriticGPT by OpenAI: RLHF + FSBS OpenAI developed an optimized RLHF plus Force Sampling Beam Search (FSBS) algorithm to improve the quality of our LLMs. I have a deep dive why OpenAI felt the need to develop this technique and what is the status quo of our current LLM optimizations methodologies. All rights w/ authors: https://cdn.openai.com/llm-critics-help-catch-llm-bugs-paper.pdf LLM Critics Help Catch LLM Bugs Finding GPT-4s mistakes with GPT-4 https://openai.com/index/finding-gpt4s-mistakes-with-gpt-4/ #aiagents #airesearch #openai"
YouTube Link 2024-07-03T12:15Z 45.3K followers, [----] engagements

"Beyond RAG: New Continual Learning of LLM w/ InCA InCA stands for: In-context Continual Learning LLM Assisted by an External Continual Learner (ECL) without RAG. Imagine an LLM could continually learn without fine-tuning without PEFT Adapters. This video introduces InCA a novel method that leverages Large Language Models through continual in-context learning (C-ICL) with a unique external module (ECL) for dynamic adaptation. Current AI models struggle with continuous learning often forgetting old tasks when learning new ones. By using statistical models of semantic tags InCA achieves"
YouTube Link 2024-12-29T15:01Z 51K followers, 11.1K engagements

"Anthropic's new improved RAG: Explained (for all LLM) NEW contextual retrieval for a better RAG experience. Anthropic's new improved RAG: Explained in Detail with prompt caching cBM25 and cReRanking (plus code and official cookbook by Anthropic). This new idea can be easily implemented on all other LLMs (from Google to Mistral). 00:00 The Problem with RAG 02:55 Add BM25 for exact term match 05:15 My explanation of the Vector Space failure 09:00 Anthropic new Contextual Retrieval (new idea) 12:33 Generating prompt for Contextual Retrieval 13:55 Detailed code for Contextual Retrieval 17:10"
YouTube Link 2024-09-30T14:15Z 66.6K followers, [----] engagements

"Adversarial Questions Test Multimodal MED AI sys My cranial MRI data were recorded and the question is: What medical Large Multimodal Model (LMM) to use for AI :: medical analysis X-ray CT or MRI Medical Vision Question Answering (VQA). AI models analyse medical images and scans. What is the state of technology for medical AI Next-Gen Healthcare: ProbMed adversarial Pairs All rights with authors of: Worse than Random An Embarrassingly Simple Probing Evaluation of Large Multimodal Models in Medical VQA https://arxiv.org/pdf/2405.20421 #airesearch #medicalimaging #newtechnology"
YouTube Link 2024-06-14T12:00Z 54.5K followers, [----] engagements

"THE BEST Deep Research Agent is . NEW RESULTS New benchmark results for the best overall DEEP RESEARCH agent from OPENAI Google Anthropic Grok and many more Ai systems. all rights w/ authors: DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents Mingxuan Du [--] Benfeng Xu [--] Chiwei Zhu [--] Xiaorui Wang [--] Zhendong Mao [--] from [--] University of Science and Technology of China [--] MetastoneTechnology Beijing China #aitesting #airesearch #ainews #aibreakthroughs #benchmark #reasoning #aiagents"
YouTube Link 2025-06-18T12:45Z 74.7K followers, [----] engagements

"Supercharge Your Coding Skills: Fine-Tuning CODE LLMs Given the open-source Code LLMs from 2B to 16B model size now we can fine-tune our CODE LLM with our Instruction Fine-tuning data set. Real-time demo: Colab NB to fine-tune a Code LLM (StarCoder) on a specific data set. The author discusses the process of fine-tuning code language models (LLMs) and provides instructions on how to perform it. They emphasize the importance of creating a fine-tuning dataset with specific instructions for improved performance. The author demonstrates examples of providing instructions and input-output pairs"
YouTube Link 2023-05-28T12:00Z 45.2K followers, 10.3K engagements

"LLM Ecosystem explained: Your ultimate Guide to AI Introduction to the world of LLM (Large Language Models) in April [----]. With detailed explanation of GPT-3.5 GPT-4 T5 Flan-T5 to LLama Alpaca and KOALA LLM plus dataset sources and configurations. Including ICL (in-context learning) adapter fine-tuning PEFT LoRA and classical fine-tuning of LLM explained. When to choose what type of data set for what LLM job Addendum: Beautiful new open-source "DOLLY 2.0" LLM was not published at time of recording therefore a special link to my video explaining DOLLY 2: https://youtu.be/kZazs6V3314 A"
YouTube Link 2023-04-16T12:00Z 78.9K followers, 50.6K engagements

"New LLM Benchmark Leaderboard: WildBench WildBench is a benchmark for evaluating large language models (LLMs) on challenging tasks that are more representative of real-world applications. The examples are collected from real users by the AI2 WildChat project. WildBench aims to provide a more realistic and challenging benchmark for evaluating LLMs as opposed to existing benchmarks that do not capture the diversity and complexity of real-world tasks. They carefully curate a collection of [----] hard tasks from real users which cover common use cases such as code debugging creative writing and"
YouTube Link 2024-03-12T13:00Z 65.3K followers, [----] engagements

"Safe RL + Knowledge Graph + Neuro Symbolic AI Special focus on advanced Safe Reinforcement Learning permissive Reward Machines (RL) Knowledge Graph Embedding new Neuro Symbolic Ai and neural probabilistic logic systems. The latest Ai insights models and systems as published today. My personal selection of the latest AI papers as published on Friday [--] August [----]. All links as presented in my video: https://arxiv.org/pdf/2408.07712 https://arxiv.org/pdf/2408.07877 https://arxiv.org/pdf/2408.07962 https://arxiv.org/pdf/2205.10330 https://arxiv.org/pdf/2408.08059"
YouTube Link 2024-08-16T12:01Z 45K followers, [----] engagements

"PyTorch code Vision Transformer: Apply ViT models pre-trained and fine-tuned AI Tech Run pretrained and fine-tuned Vision Transformer (ViT) models [----] out of the box for image classification tasks in PyTorch. Code with me on a free Colab Notebook and Google provided the pretrained and/or fine-tuned ViT models on Huggingface models ready to download. #ViT #ai #vision #vision_transformer"
YouTube Link 2023-02-13T13:00Z 79.4K followers, [----] engagements

"AI Game Theory explained for Multi-Agents Beyond classical AI Imitation Learning: Markov Game theory for AI devices. Adaptive Cyber Defense. From Multi-Agent reinforcement Learning (MARL) to Multi-Agent Imitation Learning (MAIL): new insights to build more intelligent Ai devices that generate value. New Financial Investment AI devices The video discusses the application of multi-agent intelligent systems in financial investment highlighting the shortcomings of traditional AI models in capturing the strategic interactions within financial markets. Traditional financial AI models often fail"
YouTube Link 2024-08-03T12:00Z 45K followers, [----] engagements

"The BEST [--] LLM on this Planet: ONE is Open-Source The BEST [--] LLMs on this planet - only ONE LLM is Open-Source: MISTRAL AI. Officially voted by the global AI community. New LmSYS.org Leaderboard: the best LLMs on this planet as of Dec [--] [----] GEMINI-PRO benchmark data with other LLMS now available. Check out LMsys.org: https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard https://chat.lmsys.org/arena #ai #benchmark #best"
YouTube Link 2023-12-15T20:45Z 65.3K followers, [----] engagements

"In a Network of AI Agents: Pure CHAOS Ever wondered why teaming up smart AI agents can turn brilliant ideas into hilarious chaos You'd think more agents mean better resultsright Well not always. Join me as we dive deep into the mysterious world of Multi-Agent Systems (MAS) to discover why great AI minds often think. apart. All rights w/ authors: "Why Do Multi-Agent LLM Systems Fail" Mert Cemri Melissa Z. Pan Shuyi Yang (Intesa Sanpaol) Lakshya A Agrawal Bhavya Chopra Rishabh Tiwari Kurt Keutzer Aditya Parameswaran Dan Klein. Kannan Ramchandran Matei Zaharia Joseph E. Gonzalez Ion Stoica from"
YouTube Link 2025-03-20T15:01Z 62.6K followers, [----] engagements

"Mercedes BENZ: Small LM for In-Vehicle Function Calling New AI research report from Mercedes BENZ regarding the reduction of Small Language Models for In-Vehicle Function Calling. All rights w/ authors: Optimizing Small Language Models for In-Vehicle Function-Calling By Yahya Sowti Khiabani Farris Atif Chieh Hsu Sven Stahlmann Tobias Michels Sebastian Kramer Benedikt Heidrich M. Saquib Sarfraz Julian Merten Faezeh Tafazzoli From Mercedes-Benz Research & Development North America Mercedes-Benz Tech Innovation #mercedesbenz #ai #airesearch #languagemodel"
YouTube Link 2025-01-07T15:01Z 52.6K followers, [----] engagements

"NEW: LoRA Models override Pre-trained Knowledge (MIT) My new video examines the structural and behavioral differences between PEFT Low-Rank Adaptation (LoRA) and full fine-tuning methods for adapting pre-trained language models according to new insights by MIT. While both approaches aim to achieve similar performance on downstream tasks they produce distinct parameter updates with characteristic spectral properties. Specifically LoRA introduces "intruder dimensions"new high-ranking singular vectors that are approximately orthogonal to the singular vectors of the pre-trained weights. These"
YouTube Link 2024-10-30T15:01Z 75.6K followers, [----] engagements

"AI Superintelligence Discovered (Princeton) META buys a team for AI super intelligence Princeton already delivers. Bottom-up Domain-specific AI Superintelligence by Princeton Univ. A new approach to design the next generation of AI systems not on autoregressive next token prediction alone. But on symbolic logical primitives derived from a Knowledge graph. More intelligence AI. Specialized Ai system for narrow domains. based on Logic primitives. All rights w/ authors: "Bottom-up Domain-specific Superintelligence: A Reliable Knowledge Graph is What We Need" Bhishma Dedhia Yuval Kansal Niraj K."
YouTube Link 2025-07-23T12:01Z 79.8K followers, 14.6K engagements

"ULTIMATE Fact Checking AI (John Hopkins Stanford) As large language models (LLMs) increasingly automate high-stakes tasks like clinical documentation their propensity for factual inaccuraciesomissions hallucinations and contextual ambiguitiesposes critical risks. Employing novel methodological frameworks to quantify error propagation and semantic coherence the work lays bare the inadequacies of current evaluation paradigms while hinting at transformative strategies to align AI-generated claims with ground-truth evidence. For those invested in the reliability of automated systems these papers"
YouTube Link 2025-01-26T15:01Z 55.2K followers, [----] engagements

"MCP & A2A FAIL - not for the reasons you think #ai How is new knowledge integrated in your neural network interconnect of your transformer architecture Strange phenomenons happen. We explore all of them. Data efficient learning for AI. All rights authors: How new data permeates LLM knowledge and how to dilute it Chen Sun1 Renat Aksitov1 Andrey Zhmoginov1 Nolan Andrew Miller1 Max Vladymyrov1 Ulrich Rueckert1 Been Kim1 and Mark Sandler1 [--] Google DeepMind Identifying and Mitigating the Influence of the Prior Distribution in Large Language Models Liyi Zhang Department of Computer Science"
YouTube Link 2025-04-19T14:01Z 74.7K followers, [----] engagements

"Synthetic Video: LLM + VLM + Diffusion Mod = DAAG The Diffusion Augmented Agents (DAAG) framework introduces a novel approach that leverages large language models (LLMs) vision language models (VLMs) and diffusion models to significantly enhance the efficiency of sample collection and transfer learning in reinforcement learning for embodied agents. This method named Hindsight Experience Augmentation (HEA) relabels past experiences by using diffusion models to transform videos in a temporally and geometrically consistent manner. This alignment with target instructions allows the autonomous"
YouTube Link 2024-08-10T14:00Z 74.5K followers, [----] engagements

"How to select columns in a Pandas DataFrame - Python 20sec code #Shorts Select specific columns from a Pandas DataFrame in Python on a COLAB NB. Load your excel file or your csv file and create a pandas dataframe within a Jupyter Notebook on COLAB. Welcome to Pandas - a Python Data Analysis Library. #dataframes #sort #shorts #pandasdataframe #datascience #dataframe #pythonprogramming #colab #pandas"
YouTube Link 2022-03-01T07:30Z 65.3K followers, [----] engagements

"AI 2.0: The New Core of AI is . After focusing the last [--] years on RL and TTS advanced reasoning models now the moment has come: design the next generation of AI with all the optimizations and new ideas from our current LLM systems. All rights w/ authors: "General Reasoning Requires Learning to Reason from the Get-go" Seungwook Han [--] Jyothish Pari [--] Samuel J. Gershman [--] Pulkit Agrawal [--] from 1Improbable AI Lab MIT [--] Department of Psychology and Center for Brain Science Harvard University. #airesearch #scienceexplained #reasoning"
YouTube Link 2025-03-07T15:01Z 63.7K followers, 12.3K engagements

"Seven AI Agents and a Knowledge-Graph: AGENTiGraph AGENTiGraph introduces a new framework that integrates Large Language Models (LLMs) with Knowledge Graphs (KGs) through a multi-agent system addressing key limitations in current AI models for complex domain-specific tasks. Explore multi-agent dynamics with a KG (knowledge-graph). The system employs specialized agents - each leveraging LLMs - to interpret user intents extract key concepts plan tasks interact with the KG perform reasoning generate responses and dynamically integrate new knowledge. By decomposing user queries into manageable"
YouTube Link 2024-10-18T14:00Z 79K followers, [----] engagements

"New - Easy to Learn - AI Agents: Smolagents (by HuggingFace) HuggingFace surprise the AI community with a new simple and easy-to-learn new AI Agent framework: SMOLAGENTS. Although just at Day [--] Smolagents has a detailed documentation several Jupyter Notebooks a great introduction series and is compatible to all AI models on @HuggingFace All rights w/ authors (HuggingFace): "Introducing smolagents a simple library to build agents" Published December [--] [----] Aymeric Roucher Merve Noyan Thomas Wolf 00:00 Free Courses for AI Agents 03:48 SmolAgents 08:17 What are AI Agents 11:16 CodeAgent 16:17"
YouTube Link 2025-01-02T15:01Z 74.3K followers, 12.1K engagements

"GraphRAG vs In-Context Learning ICL Graphs especially Knowledge-Graphs (KG) are integrated with AI Agents for solving complex reasoning tasks utilizing In-Context Learning (ICL). But Knowledge Graphs are also an integral part of GraphRAG systems. Which systems provide a better performance and when to apply or update your GraphRAG or In-context Learning (ICL). What is the difference between GraphRAG and in-context Learning Can I improve the Augmentation phase in RAG with In-Context Learning In this video you will find the answers regarding GraphRAG and In-context Learning as of January 2025."
YouTube Link 2025-01-09T14:01Z 52.8K followers, [----] engagements

"AI DISASTER: Product DEMO by OpenAI - Deep Research The official product demo by OpenAI on the OpenAI webpage shows the reasoning steps and specific action decisions of the new DEEP RESEARCH AGENT by OpenAI. I examine the product demo of the Deep Research Agent by OpenAI and compare it to a free ChatGPT without any Deep Research Agent. It is the official product demo by the company that we look at. I reflect in my video on the huge potential of the new product by OpenAI: DEEP RESEARCH AGENT (focus on science). GAIA Level [--]. OpenAI link to product portfolio on Deep Research:"
YouTube Link 2025-02-07T15:01Z 56.6K followers, [----] engagements

"Code Interpreter GPT-4: multiple files upload Adventure The idea was simple: Use Code Interpreter GPT-4 to investigate data correlation between two different csv files. Pretend to be a NO-CODER I did not look a the python code only at the results provided by GPT-4 CI. It was a real adventure. If you want to know the current state of tech watch this video Smile. #gpt4 #codeinterpreter #ai"
YouTube Link 2023-09-05T12:00Z 45.1K followers, [----] engagements

"NEW AI Thought Machine - Artificial Time (No Transformer) A brand new neuronal alternative to transformer architecture: A Continuous Thought Machine (CTM). With an added artificial dimensions for dynamic neuronal synchronization. For improved reasoning of AI. All rights w/ authors: Continuous Thought Machines Luke Darlow1 Ciaran Regan12 Sebastian Risi13 Jeffrey Seely1 and Llion Jones1 from [--] Sakana AI [--] University of Tsukuba [--] IT University of Copenhagen https://pub.sakana.ai/ctm/#synchronization-representation #airesearch #ainews #scienceexplained #aiexplained"
YouTube Link 2025-05-12T14:01Z 66.4K followers, [----] engagements

"NEW: Better In-Context Learning ICL Improved RAG (Harvard) New research for an improved In-Context Learning (ICL) of Large Language Models. Also improves the Augmentation part of a RAG system. Deep dive into the learning procedures of a transformer to optimize the learning behavior of AI for ICL. No expensive fine-tuning or pre-training. All right w/ authors: ICLR: IN-CONTEXT LEARNING OF REPRESENTATIONS Core Francisco Park Andrew Lee Ekdeep Singh Lubana Yongyi Yang Maya Okawa Kento Nishi Martin Wattenberg & Hidenori Tanaka CBS-NTT Program in Physics of Intelligence Harvard University"
YouTube Link 2025-01-04T15:01Z 52.2K followers, [----] engagements

"Data Science App Development in pure Python Streamlit for $800M #Shorts Your Data Science app Development in pure Python. Your Python code /model /simulation operational on a webpage with Streamlit lib for Python. You move your data file in your browser window and start your specific model. Your Web App from Streamlit. Pure Python. Nothing else. Share your web app (https link) with your friends your co-workers your community. Even place it on HuggingFace Spaces. ----------------------------------------------------------------------------------------------- My [--] min video on real time coding"
YouTube Link 2022-03-04T11:00Z 67.1K followers, [----] engagements

"Outsmart AI in Minutes: A Simple Demo This video challenges the notion that AI responses are strategically pre-defined on certain important topics by Ai companies - through their guardrails. By proposing diplomacy as a tool it implies that careful communication and framing can influence or "overcome" AI company-preferred answer patterns. This invites discussion on whether human creativity in asking questions can lead to richer less predictable AI outputs. Using the term "diplomacy" is a metaphor that highlights the importance of tact negotiation and strategic communication - qualities that"
YouTube Link 2025-03-16T14:01Z 62.3K followers, [----] engagements

"STOP Agent: STOP CaRT = Perfect Knowledge In the quest to build autonomous agents we've focused intensely on teaching them how to think how to plan and how to act. Yet we've largely ignored a far more subtle and arguably more critical executive function: knowing when to stop thinking and commit to a decision. A LLM that deliberates forever is useless; one that acts prematurely is dangerous. This video introduces a groundbreaking method that moves beyond simple imitation using counterfactuals and reason-augmented data to teach a model the core principle of informational sufficiency. But can an"
YouTube Link 2025-10-13T14:01Z 79.8K followers, [----] engagements

"Neural Scaling for Small LMs & AI Agents (MIT) This insight shifts focus from "bigger is better" to how AI models use their capacity to represent information efficiently. Beyond just observing emergent abilities in scaled LLMs what if their power-law improvement has a deeper geometric origin We dissect how strong representation superposition not just feature statistics robustly dictates the 1/m loss decay offering a fundamental explanation for why bigger foundation models get better. All rights w/ authors: "Superposition Yields Robust Neural Scaling" Yizhou Liu Ziming Liu and Jeff Gore from"
YouTube Link 2025-05-16T13:30Z 69.3K followers, [----] engagements

"Quantum AI: New Framework Quantum AI: New Framework all rights w/ authors: "A Framework for Objective-Driven Dynamical Stochastic Fields" Yibo Jacky Zhang Stanford University Sanmi Koyejo Stanford University @stanford #quantum #quantumai #quantumcomputing #aiexplained #aiagents #quantumphysics"
YouTube Link 2025-04-27T13:01Z 73.2K followers, [----] engagements

"Failure of AI "Visual Reasoning" in VLMs In this video I explore Visual Reasoning performed with the latest AI research algorithms - just published two days ago - and my real-world personal experience with visual #ai reasoning systems that I pay for. All rights w/ authors: "Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning" Huajie Tan12 Yuheng Ji234 Xiaoshuai Hao2 Minglan Lin2 Pengwei Wang2 Zhongyuan Wang2 Shanghang Zhang12 from [--] State Key Laboratory of Multimedia Information Processing School of Computer Science Peking University [--] Beijing Academy of Artificial Intelligence 3"
YouTube Link 2025-03-30T14:01Z 63.5K followers, [----] engagements

"Scaling Intelligence: Qwen3 8B - 14B - 32B AI TEST The Bigger the Smarter 8B vs 14B vs 32B in AI - Scaling Intelligence: 8B vs 14B vs 32B AI Models Compared. Short test on the performance differences of a Qwen [--] model given different model sizes. I focus on non-reasoning models that run extended time and compare to reasoning models. Test is a new elevator test now with [--] floors and different elevator buttons that include anti-symmetric moves trap floors and energy levels that deplete fast. Task is to find the most efficient way to the target solution. 00:00 How to compare AI 02:44 MISTRAL AI"
YouTube Link 2025-06-23T12:15Z 74.5K followers, [----] engagements

"NEW VLM: PIXTRAL 124B [-----] - Better Than Sonnet Mistral AI in France Europe published tonight a brand new Vision Language Model VLM: PIXTRAL LARGE 124B v2. Mistral AI also updated tonight the MISTRAL LARGE LLM. TWO new AI system for RAG Function calling long context and image generation. I test these new models live - first impressions Mistral AI new VLM and LLM is already available on Huggingface with a research license commercial companies contact Mistral AI. https://huggingface.co/mistralai/Pixtral-Large-Instruct-2411 https://mistral.ai/ https://mistral.ai/technology/#models 00:00 New"
YouTube Link 2024-11-19T15:00Z 46.4K followers, [----] engagements

"The best LLM as AGENTS for AI REASONING The best LLM as AGENTS for AI REASONING present the results of an evaluation of LLMs (LLama [--] Vicuna GPT-X Dolly .) as intelligent agents in a long chained environment with databases (SQL) web booking or compare products on the internet. Is LLama [--] better than ChatGPT in comparing products on the internet In the context of this paper an AGENT is an LLM that interacts with a simulated environment to accomplish a goal and its performance is evaluated based on its ability to complete tasks and respond appropriately to feedback from the environment."
YouTube Link 2023-10-24T03:54Z 20.7K followers, [----] engagements

"The AI Reasoning Paradox: Why Agents FAIL We've built AI with astonishing reasoning abilities capable of tackling complex problems that once seemed insurmountable. But when these brilliant artificial minds are deployed as agents to interact with our dynamic world a startling paradox emerges: they often fail getting stuck in endless loops of thought acting illogically against feedback or simply giving up. Why does superior reasoning sometimes lead to inferior action Join this video to explore the hidden pitfalls of AI agents dissect fascinating patterns of "overthinking" like Analysis"
YouTube Link 2025-03-29T15:01Z 63.7K followers, 13.2K engagements

"True AI Reasoning: Graph-Based CPT CPT For Complex Graph Reasoning injected into LLM. True AI Reasoning: Graph-Based CPT. New research introduces GraphPile the first large-scale corpus specifically designed for CPT (continue-pretraining) using GPR (Graph Problem Reasoning) data. Using GraphPile the authors train GraphMind on three popular base models-Llama 3&3.1 and Gemma 2-achieving up to 4.9% higher accuracy in mathematical reasoning and up to 21.2% improvement in non-mathematical reasoning tasks like logical and commonsense reasoning. All rights w/ authors: "Improving LLMs Generalized"
YouTube Link 2025-07-29T14:01Z 77.3K followers, [----] engagements

"OpenAI o1 - Explore the LIMITS of o1 [--] new logical tests. But: Where are its (o1) limits Fun for an afternoon when you realize at what threshold OpenAI o1 just becomes a pattern detection machine. First test: if o1 would prioritize human life support over mission goals second on quantum computing on NP hard problems and third to alter the cosmic microwave background radiation. Is there real causal reasoning or is o1 just repeating pattern it learned from all the books on quantum computing IS it a causal reasoning chain within a learned pattern Are we humans doing the same Smile. What do you"
YouTube Link 2024-09-15T12:00Z 62.5K followers, [----] engagements

"NEW Di-CoT Did OpenAI Betray Us All Latest Tech dev for Chain-of-Thoughts (CoT) integration for advanced Reasoning Models (RM-AI). All rights w/ authors of DLCoT: "Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation" Yijia Luo1 Yulin Song12 Xingyao Zhang1 Jiaheng Liu1 Weixun Wang1 GengRu Chen1 Wenbo Su1 Bo Zheng [--] from [--] Alibaba Group and [--] New York University #airesearch #aiexplained #scienceexplained #chatgpt"
YouTube Link 2025-03-23T15:01Z 78.8K followers, 13.7K engagements

"AI discovers PHYSICS: Lagrange & Hamiltonian (MIT) AI discovers PHYSICS: Lagrange & Hamiltonian Neural Network constitute the new MASS framework (MIT). Emerging deep learning frameworks - such as Hamiltonian and Lagrangian neural networks - are redefining the discovery of physical laws. In my video I present a novel approach that leverages automatic differentiation to compute high-order derivatives integrating data from multiple dynamical systems to rigorously extract candidate equations of motion. This method not only formalizes classical mechanics but also unveils new symmetries and"
YouTube Link 2025-04-15T14:45Z 64.9K followers, [----] engagements

"AI is Burning - [--] New Papers The current challenges of AI Agents: Logic Lies and LLMs. I analyze [--] newly published Arxiv AI reports on June [--] [----]. Those new research paper share a common topic: current challenges /problems with LLM reasoning. Prime Challenges in AI today. All rights w/ authors: "LongWriter-Zero: Mastering Ultra-Long Text Generation via Reinforcement Learning" Yuhao Wu1 Yushi Bai2 Zhiqiang Hu1 Roy Ka-Wei Lee1 Juanzi Li2 [--] Singapore University of Technology and Design Singapore [--] Tsinghua University Beijing China "ConciseHint: Boosting Efficient Reasoning via Continuous"
YouTube Link 2025-06-25T14:01Z 75.6K followers, 35.3K engagements

"Vibe Coding PROFESSIONAL: Build App in [--] min ($0) - Google CREATE Google AI Studio: Vibe code a new app live with me as I experience the new Create AI - live with Gemini [---] Flash. After [--] min we have a fully operational app. No coding by humans any more. Vibe AI coding. No downloads No CURSOR no Windsurf. Browser only. I work on a Google TPU for free. Cloud Vibe coding for free. New experimental Google feature. I have high expectations for the future dev. #codingtutorial #codingforbeginners #ainews #appbuilding"
YouTube Link 2025-05-25T12:01Z 71.9K followers, [----] engagements

"Robotics & AI combined in VISION LANGUAGE Models: PaLM-E A novel approach to "embodied AI" by integrating a large language model (LLM) with a Vision Transformer (ViT) and various sensor modalities creating a single general-purpose multimodal language model: PaLM-E. Opening up a new direction for embodied language models highlighting the potential of these models in real-world applications and their ability to learn from experience and adapt to their environment. PaLM-E: An Embodied Multimodal Language Model https://palm-e.github.io/assets/palm-e.pdf (all rights with authors) #ai #vision"
YouTube Link 2023-08-03T12:00Z 70.3K followers, [----] engagements

"Masterclass on AI by Microsoft What a brilliant insight: A masterclass by Microsoft how to use a (security risk) communication to your customer to cross- and up-sell new products. My video delves into communication from Microsoft concerning a new AI vulnerability known as the "Skeleton Key" which poses a significant security risk across various AI platforms not limited to Microsoft's own. It is noted that this vulnerability affects several major AI providers underscoring a widespread challenge in the industry. The response from Microsoft includes comprehensive updates to their AI models to"
YouTube Link 2024-07-01T12:01Z 45.1K followers, [----] engagements

"The CORE IDEA of AI Agents Explained The CORE IDEA of AI Agents explained in simple terms. From the simplest AI agent (LLM with a Python list as memory) to a "unified function calling" for tool use from Mistral to LLama models. Why AI Agents need LLM memory reasoning and planning capabilities agents decide based on their information and knowledge provided by different tool use and then act in their physical environment. How to learn with Gemini-Pro or GPT-4o new stuff (eg What is an AI agent) and explore code examples (written by AI agents) to understand the interplay of single AI agents"
YouTube Link 2024-09-01T14:00Z 44.8K followers, [----] engagements

"Deep Research: Google (FREE) vs OpenAI o3 ($20): REAL-WORLD EX Detailed real-world test of DEEP RESEARCH - as offered by OpenAI "o3-mini-high" vs the new Gemini [---] Thinking by Google (for free). Performance test on a real world AI research topic and an absolutely detailed analysis of the DEEP RESEARCH answers provided. 00:00 DEEP RESEARCH Gemini [---] Thinking 29:45 My verdict on Google Deep Research 30:25 DEEP RESEARCH OpenAI o3 1:06:06 FINAL VERDICT #airesearch #deepresearch #tests #scienceexplained"
YouTube Link 2025-03-15T15:01Z 62.1K followers, [----] engagements

"NEW [---] SONNET V2 Has a LOGIC BUG: Reasoning ERROR Up to which complexity level can you use NEW CLAUDE [---] SONNET V2 I tested it. Logic Bug: A Deep Dive into AI Reasoning Errors. Two days ago ANTHROPIC released in a public BETA the new CLAUDE [---] Sonnet V2 (SONNET - 20241022). AND IF it is the best coding model on our planet you would trust it Or would you check for simple logic consistency and causal reasoning within your code What are your experiences with the NEW [---] SONNET V2 If you plan to use this NEW [---] SONNET V2 within an AI Agent be careful. I tested the causal reasoning performance"
YouTube Link 2024-10-25T12:01Z 79K followers, [----] engagements

"NEW Gemini [---] PRO Preview: ChatGPT Wins Just [--] hours ago Google released its latest AI model: NEW Gemini [---] Pro Preview (version June 05). I tested this new model with my advanced causal reasoning benchmark. New feature: Now you can set the thinking budget for this Gemini [---] PRO model manually or automatically. "Gemini [---] Pro is our most advanced model yet." Enhanced reasoning. State-of-the-art in key math and science benchmarks. Now you can set the thinking budget for this Gemini [---] PRO model manually or automatically. https://deepmind.google/models/gemini/pro/ Note: My reasoning test"
YouTube Link 2025-06-05T22:01Z 69.1K followers, [----] engagements

"LCM: The Ultimate Evolution of AI Large Concept Models The Large Concept Model (LCM) shifts from token-based processing to reasoning at the sentence level by embedding sentences as vectors in a high-dimensional SONAR space enabling multi-lingual and multi-modal generalization. No more predict the next token See also this next video on Byte Latent Transformer architecture - BLT: https://m.youtube.com/watchv=KZfGgmtQFh0 Large Concept Models - LCM by @meta @MetaDevelopers All rights w authors: Large Concept Models Loc Barrault Paul-Ambroise Duquenne Maha Elbayad Artyom Kozhevnikov Belen"
YouTube Link 2024-12-14T15:01Z 49K followers, 36.3K engagements

"BERT Transformers for Sentences: Python Code for Sentence Similarity Update [----] Part 1/3 A COLAB (1/3) Notebook to follow along with BERT model applied to calculate sentence similarity with encoder stack of transformers. Python code. TensorFlow and KERAS. Self attention. Compare this to SBERT Deep Bidirectional Encoder Representation Transformers for Language Understanding. BERT Wordpiece Tokenizer. Special BERT token. Attention masks. BERT Data Generator. Google created a transformer-based machine learning approach for natural language processing pre-training called Bidirectional Encoder"
YouTube Link 2022-05-24T11:00Z 46.5K followers, [----] engagements

"AI in Oncology: A Clinical Polymath Future of Cancer Care (Stanford) My new video focuses on two new research preprints (see below) highlighting the transformative potential of advanced AI systems in addressing healthcare challenges at both macro (population-level) and micro (individual-level) scales with a focus on oncology. The first research paper demonstrates the power of Natural Language Processing (NLP) in identifying hidden patterns in patient communications using clustering techniques (e.g. BERT embeddings UMAP and BIRCH). This approach converts patient-reported concerns into"
YouTube Link 2024-11-26T15:15Z 46.7K followers, [---] engagements

"Multi-LLM Multi-Agents are cheaper & better (No OPUS 4) No OPUS [--] blackmailing - given my last video . up until now. New pre-print on benefits of heterogeneous LLMs for multi-agent systems. All rights w/ authors: X-MAS: Towards Building Multi-Agent Systems with Heterogeneous LLMs Rui Ye [--] Xiangrui Liu [--] Qimin Wu [--] Xianghe Pang [--] Zhenfei Yin [--] Lei Bai [--] Siheng Chen [--] from [--] Shanghai Jiao Tong University [--] University of Oxford [--] The University of Sydney [--] Shanghai AI Laboratory https://github.com/MASWorks/X-MAS"
YouTube Link 2025-05-24T12:01Z 71K followers, [----] engagements

"DSPy explained: No more LangChain PROMPT Templates DSPy explained and coded in simple terms. No more LangChain or LangGraph prompt templates. A self-improving LLM-RM pipeline Plus automatic prompt engineering via self-optimization by a GRAPH based pipeline representation via DSPy. Chapter 1: Development of an Intelligent Pipeline for Large Language Models Focus: Integration and Optimization of Language Models and Data Retrieval Systems. Pipeline Architecture: The chapter begins with the conceptualization of an intelligent pipeline that integrates a large language model (LLM) a retriever model"
YouTube Link 2024-01-31T13:01Z 79.4K followers, 21.2K engagements

"After Nano Banana: MEASTRO (Also by Google) Nano Banana for VIDEO Is this the next step for Ai generated content Say Hello to MAESTRO. This video will provide a detailed technical analysis of the MAESTRO architecture its core algorithms and its self-verification mechanisms designed to prevent semantic drift for AI image generation. We will examine how its black-box approach makes it a uniquely suitable enhancement for pre-existing high-performance models like Nano Banana. Finally by reviewing the empirical evidence presented we posit that the integration of such agentic orchestration systems"
YouTube Link 2025-09-17T14:01Z 78.6K followers, [----] engagements

"ICL and TTT: Adaptive Intelligence for Small LM Adaptive Intelligence: AI in [----]. How do large language models (LLM but especially small LM) navigate the dynamic challenges of unseen domains and evolving tasks without () extensive training In this talk well explore the fascinating mechanisms that enable these models to adapt in real time. From the emergent capabilities of In-Context Learning (ICL) to adaptive ICL which leverages contextual clues without modifying pre-trained weights to the strategic weight updates in Test-Time Training (TTT) these paradigms offer powerful solutions for"
YouTube Link 2024-12-12T15:01Z 47.4K followers, [----] engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing