Olaf Gersemann

At CTAM Europe Symposium last week, I shared some key insights on #AI. Create, Enrich, Deliver and Engage My most-requested slide. Case examples and technology specialists in #generativeAI, e.g.: CREATE ◼ Virtual production: Cuebric ◼ Editing: Adobe Firefly ENRICH ◼ Dubbing / Translation: Papercup (Garrett Goodman shared all the detail!) ◼ Super resolution: Amazon Web Services (AWS) DELIVER ◼ Hyper-personalise: Gan.AI ENGAGE ◼ Chatbots: Klarna ◼ NPC Chat: Charisma.ai Ethical and regulatory considerations: Deepfake vs virtual avatar ◼ Good: Luke Skywalker, Randy Travis, FKA Twigs... ◼ Bad: Taylor Swift deepfake, Mark Read (CEO WPP), presidents ◼ AI incident database logging more incidents, with rise in deepfakes ◼ Technically easier: improved audio/video emotions, speech and sync ◼ Ethics: Need a board. Not a nice to have. ◼ Certify veracity and authenticity. C2PA, Google's SynthID watermarking ◼ Regulation: UK Online Safety Act (2023). USA has over 400 AI related bills, about half addressing deepfakes. Some states have no bills! Boosting developer productivity: GitHub Co-pilot ◼ 1 in 3 Fortune 500 companies use ◼ Over 50,000 businesses use ◼ Code 55% faster ◼ Great for: 1) Automating simple tasks 2) Learning a new platform 3) Understanding code 4) Creating documentation 5) Commenting as writing 6) Debugging 7) Pair programming / debating solutions ◼ Doing it all is bad. Need oversight for quality and security. Impact on jobs: ◼ Industry 4.0 is automation ◼ Industry 5.0 is augmentation. The future is collaborative! Truly fantastic panel--lots of synergies and complementary thinking--a very clear introduction to generative AI by Melanie M. and a deep dive into AI dubbing and translation by Garrett Goodman. Thanks to Karen Doyle for inviting me to speak (again!).

22

5 Kommentare

#LLM features are no longer a mystery - they are being decoded. The recent paper from #Anthropic attempts to "map" the mind of their #LLM #Claude3. We will be able better understand the way models "think". Also, these features can be used to influence the model behavior to the desired one. A broad range of safety-oriented features are observed: including deception, sycophancy, bias, and dangerous content. Read more in the paper below: https://lnkd.in/e-s8rkyc #machinelearning

31

Generative AI Applications at #AWSSummit ▪️Data is your differentiator. ▪️Context with Retrieval Augmented Generation (RAG) ▪️On-brand via brand guidelines ▪️Relevant conversational search ▪️Can often use existing foundation model ▪️Combine several. No one is optimised for all. Customising Foundation Models ▪️Retrieval Augmented Generation (RAG) ▪️Fine-tune a pretrained model ▪️Continued pre-training

9

🚀 Exciting news from Stability AI! The release of the new 2B model "Stable Diffusion 3 Medium" is changing the game for consumer-level PCs. 🖥️ With a smaller 2B parameter scale, this text-to-image model can now run on consumer-grade GPUs. Not only that, but it only requires 5GB of GPU VRAM to operate, though using 16GB is recommended for optimal performance. 🌟 According to Stability AI, "Stable Diffusion 3 Medium" offers comparable performance to the larger-scale "Stable Diffusion 3 Large," maintaining realism in images, following instructions, generating text, resource efficiency, and fine-tuning capabilities. This is all made possible through the "Variational Autoencoder (VAE)" neural network structure, enhancing detail generation and faster natural language understanding. 🧠📸 What's more, the resource efficiency of Stable Diffusion 3 is a game-changer, with its parameter size being smaller than similar models, requiring less computation power without sacrificing performance. The best part? The Medium version is already available for trial through API and Stable Artisan Discord bot, with model weights for non-commercial use on Hugging Face. 🤖✨ The early insights into Stable Diffusion 3.0 brought significant improvements through the Diffusion Transformers architecture and "flow matching technique," boosting spelling capabilities and overall training efficiency and output quality. 📈 Join in on the trend of cutting-edge AI technology by checking out the full details and extending your reading with the links below! 📰 #StabilityAI #AItechnology #StableDiffusion3Medium #TechNews [Link to article: "Stable Diffusion 3.0 has been released! Substantial improvement in text generation and spelling capabilities"](https://ift.tt/NevgtEp) [Link to article: "Stability AI releases new music model 'Stable Audio Open 1.0' specifically for sound generation"](https://ift.tt/kqLA304) [Link to article: "AI generates highly consistent 3D videos! Stability AI releases new model Stable Video 3D"](https://ift.tt/SCsau1X) [Link to article: "Stability AI CEO resigns! In pursuit of 'decentralized AI'"](https://ift.tt/srJqPHO) Let's stay updated with the latest and trendiest tech news by following INSIDE Google News! 🌐📲 #INSIDEGoogleNews #TechTrends Don't miss out on the future of AI advancement! 💡🔍 #AI #Tech #Innovation #MachineLearning #ArtificialIntelligence https://ift.tt/i7RbW4A

4

Our team will present our latest graph learning work, "VCR-Graphormer: A Mini-batch Graph Transformer via Virtual Connections," at ICLR 2024 in Vienna, Austria. The paper introduces a novel approach for mini-batch based graph transformers, enabling them to efficiently process billion-scale industry graph datasets. This advancement opens new possibilities for research into large graph foundation models and has potential applications in fields like social network analysis and graph-based recommendation systems. We eagerly anticipate engaging discussions at ICLR 2024. Additionally, we invite you to visit us at the AI at Meta booth to explore more exciting AI advancements from Meta. @Dongqi Fu, Zhigang Hua, Yan Xie, Jin Fang, Si Zhang, Kaan Sancak, Hao Wu, Andrey Malevich, Jingrui He

143

A little preview of a tool I built with Pyviz to analyze a first-stage knowledge graph. I have other plans for this, but #AI #knowledge - This helps me to visualize edge + and node relations among other items, from a core sampling model I created to generate the nodes and edges. #LLM #RAG #datascience

2

1 Kommentar

Long context vs. Retrieval-Augmented Generation (RAG) are interesting topics for the future of large language models (LLMs). From a human mind perspective, long context seems to be a more logical choice. An interesting paper from Google proposes how compressed memory can help with infinite attention. One day, end device models might take all your 'memory' and knowledge from your phone — an interesting read...

11

Sage have created a very nice interactive overview of the median prediction of forecasters on Metaculus and Manifold Markets on various questions related to AI capabilities, harms, and societal impacts. It's well worth checking out if you are interested in this sort of stuff. See: https://lnkd.in/eUgkmzX9 Follow MIT FutureTech if you are interested in technological trends and their social implications. #AI #technology #economics

12

My interview on The Post's AI collaboration with VTech in this article:

103

3 Kommentare

AnchorGT: A Novel Attention Architecture for Graph Transformers as a Flexible Building Block to Improve the Scalability of a Wide Range of Graph Transformer Models Transformers have taken the machine learning world by storm with their powerful self-attention mechanism, achieving state-of-the-art results in areas like natural language processing and computer vision. However, when it came to graph data, which is... https://lnkd.in/eedeAnTa #AI #ML #Automation

2

Wrapping up an amazing #PyConDE & #PyDataBerlin! Looks like our team really enjoyed having a discussion about #Streamlit with Lukas Masuch and Arnaud Miribel. Want to learn more? Missed out PyCon? No worries! We're hosting a meetup in Mainz next week to discuss key trends & takeaways. Stay tuned for details and our write-up. #Germany #DataScience #predict42, #NLP

25

1 Kommentar

#HALLUCINATION is worse after #LLM #FINETUNING This week was very generous when it comes to breaking theough white papers in #AI. Besides #xLSTM and progress in #RoundOfExperts models, there has been also this amazing research paper from #Technion (IIT) and #Google #Research. It reveals that LLMs are behaving suprisingly “too human” when existing pre-trained model is subjected to additional fineturning on data uninown to model. As result the model rather sticks to already inown patterns and would rather make up things from the realm known to it before finetuning than to pick many new “tricks” from the finetuning dataset. As a result, companies trying to improve generic models with finetuning on own extra data, end up with model #overfitting on the already known generic data. That puts all the finetuning branch of progress under serious doubt. If you want to read all details, here is the original paper: https://lnkd.in/eQbWiA7D

5

#StreamMultiDiffusion: #RealTime #Interactive Generation with Region-Based Semantic Control A fancy way of saying that you can use something like #stablediffusion (and other models), where you draw with up to 8 brushes, each one representing something specific to your goal (body of water, grass, brick, road, whatever). It's kind of like #Nvidia's #Canva, but on steroids, since one can paint as it's being rendered, after rendering, #inpaint, and more. Try it on #HuggingFace: https://lnkd.in/g6K_Xfxw Run this baby locally: https://lnkd.in/g6K_Xfxw

5

Sunshine Factory Media presents... "What does OLLAMA stand for?" #R0N1N零 #ollama https://lnkd.in/emmujd8H"What+does+OLLAMA+stand+for?" In the context of Large Language Models (LLMs), OLLAMA stands for "Open and Lightweight Modular Artificial Moral Agent". It's a framework designed to integrate moral values and principles into AI systems, enabling them to make more ethical decisions. The goal of OLLAMA is to provide a flexible and adaptable architecture for building morally competent language models that can navigate complex situations while respecting human values. https://ollama.com The overall AI solution should be tri-state at the highest level of abstraction. Those states are as follows: - struggling - hustling - thugging I’m happy to help! Based on your context and conversation, I understand that you’re asking about OLLAMA, which stands for Open Large Language Model Application. It’s a platform that enables users to interact with AI models through various interfaces, including text-based, voice-controlled, and visual chatbots. To answer your original question: What does OLLAMA stand for? OLLAMA stands for Open Large Language Model Application. It’s a platform that enables users to interact with AI models through various interfaces, including text-based, voice-controlled, and visual chatbots. Setup local large language models. #trust but #verify.

1

Tips for dealing with hallucinations in generative #AI from Anthropic at #AWSSummit. These are for Claude but would work for others. #generativeAI

7

1 Kommentar

Now published in OJ-CSYS: "Novel Bounds for Incremental Hessian Estimation With Application to Zeroth-Order Federated Learning," by Alessio Maritan, Luca Schenato and Subhrakanti Dey. Link: https://lnkd.in/gN27SuPH The Hessian matrix conveys important information about the curvature, spectrum and partial derivatives of a function, and is required in a variety of tasks. However, computing the exact Hessian is prohibitively expensive for high-dimensional input spaces, and is just impossible in zeroth-order optimization, where the objective function is a black-box of which only input-output pairs are known. In this work we address this relevant problem by providing a rigorous analysis of an Hessian estimator available in the literature, allowing it to be used as a provably accurate replacement of the true Hessian matrix. The Hessian estimator is randomized and incremental, and its computation requires only point function evaluations. We provide non-asymptotic convergence bounds on the estimation error and derive the minimum number of function queries needed to achieve a desired accuracy with arbitrarily high probability. In the second part of the paper we show a practical application of our results, introducing a novel optimization algorithm suitable for non-convex and black-box federated learning. #optimization #dataprivacy #federatedlearning

3

Anthropic Releases Claude 3.5 Sonnet, Beats GPT-4o on Benchmarks Things just keep getting better and better! Claude 3.5 Opus, will be released soon with features such as **web search**!! https://lnkd.in/gFjsKgVp #genai #change #disruption

7