Elad Inbar

Software is Feeding the World: "Closer Look" edition (no not inspired by Seth Meyers!) Starting this month, subscribers will get a monthly edition (still free) where I take a closer look at a particular topic. The inaugural edition looks at the following: 1. IoT devices and applications within food and agriculture 2. Space for high temporal resolution data to drive decision making throughout the supply chain. 3. Why connectivity is the big unlock to drive positive ROI 4. Types of emerging connectivity technologies and different types of IoT sensors and their applications 5. How low cost technologies like LoRaWAN could potentially unlock new use cases, but it does require connectivity infrastructure investments 6. What can we learn from examples of low cost data in India and how it drove many new use cases very rapidly. Do give it a read and would love to get your constructive feedback on the addition of a new format to the SFTW portfolio! Link to the write up in the first comment #agtech #IoT #infrastructure #connectivity

30

2 Comments

A nice quick overview of DSPy: https://lnkd.in/gQKj42va

1

How does the speed of SwiftData compare to Realm? What about memory usage? Size? ⚖ Do the results change if you're mutating 100 objects vs. 1,000,000? How about a simple data model vs. a complex one? 🤔 We have answers 🤓 https://lnkd.in/gpymY7wU

57

9 Comments

When building custom #LLM or #SLM models in the #GenerativeAI #Lab, it can now automatically generate tests #Bias, #Fairness, #Robustness, and other #ResponsibleAI categories, Tests can then be run, edited, shared, and reused across projects. https://lnkd.in/g5y3aqbg #ai #datascience #ethicalai #largelanguagemodels

28

1 Comment

Join me and ElectrifAi CTO Luming Wang for our latest episode of #ConsequentialAi where we explore an important topic: Driving Success at the #Ai Application Layer. With all the focus on #GPUs and #foundationmodels, we decided to share our thoughts on application layer success. As many of you know, ElectrifAi is a #software firm whose products quickly solve consequential business problems for the #Csuite. We start at solving the business problem and work backward. The #ML, #NPL, #ComputerVision and #LLMs are simply the enabling tools in out toolbox. And we put our money where our math is by guaranteeing 5x to our clients and only being paid on success. That's pretty bold. Along the way and after working with many clients, we learned a few things about what it takes to be successful at the application layer with Ai - delivering quantifiable business value to the C-suite. And in Podcast 8 of #ConsequentialAi we share four or five key points from our journey. As a bonus at the end, we also share a little bit about LLM applications. We hope you enjoy the conversation and find it helpful. #ai #bigdata #digitaltransformation #privateequity #venturecapital https://lnkd.in/eYQCMZWZ

16

Matrix multiplications (MatMul) are the most computationally expensive operations in large language models (LLM) using the Transformer architecture. As LLMs scale to larger sizes, the cost of MatMul grows significantly, increasing memory usage and latency during training and inference. Now, researchers at the University of California, Santa Cruz, Soochow University and University of California, Davis have developed a novel architecture that completely eliminates matrix multiplications from language models while maintaining strong performance at large scales. In their paper, the researchers introduce MatMul-free language models that achieve performance on par with state-of-the-art Transformers while requiring far less memory during inference. #artificialintelligence #matrix #multiplications #LLMs #efficiency #transformers https://lnkd.in/gPBMTUNw

6

💡 Why do we need so many evaluation tools for LLMs? As engineers, we build 'production-ready' LLM products using these metrics. But what happens next? How do we maintain control and ensure reliability? At Feedback Intelligence, we’ve crafted a cookbook to keep your LLMs reliable and aligned with user expectations. 🍲 📖 Give it a read and let’s chat!

16

Custom Fine-Tuned LLM With Just A Few Clicks With Abacus.AI you can fine-tune LLMs and vastly improve them for a particular task or specific knowledgebase. We have invented several fine-tuning techniques that make your custom LLMs more performant and help surpass GPT-4 performance. Check out how easy it is to create your own custom fine-tuned LLM!

126

4 Comments

In less than 24 hours ‘Meta Llama 3 Fine tuning, RAG, and Prompt Engineering for Drug Discovery’ has become a top ChemicalQDevice event seminar, second only to the March ‘Tensor Network Developer Revolution: 125 Lines of Code or Directions’ presentation with 949 views. Seminars: https://lnkd.in/g7jq3Kj8, https://lnkd.in/gfmJDZfy Here’s a look ahead: 1) RAG incorporations into larger pre-trained models; and improved fine-tuned models. 2) Better utilization of specific cancer datasets to better address Drug Discovery customer needs.  3) Tensor network and relevant quantum-inspired algorithms to ameliorate DNA/RNA/Protein data curse of dimensionality, growing LLM file sizes, controllability, and explainability issues. Slides download: https://lnkd.in/gCubePSJ, https://lnkd.in/gqR7gmMC

20

Technical report: Learn about fine tuning  Small cheaper to run models can do better than GPT-4 with a small amount of fine-tuning. "We find that 4-bit LoRA fine-tuned models outperform base models by 34 points and GPT-4 by 10 points on average" How to fine-tune? LoRa method is the most well known and can be done on any decent gaming PC for learning. Many YouTube tutorials can be found. Having your own fine-tunes gives your business something unique in the market, you build your dataset up over time, building that value. Paper: https://lnkd.in/gi6PqyYt

8

2 Comments

NVIDIA's Jensen said "ChatGPT democratized computing, Llama2 democratized generative AI" ...and now Llama3 is the next level 🔥 It's fascinating to watch Meta's strategic moves. With Llama3, they prioritized training on a massive 15T token dataset to pack all into a lean 70B param model, instead of building a massive model. This allows Llama 3 to match trillion+ parameter models like GPT-4, but at 1/10th the compute, storage and inference costs! 💰 This technique was published by Google DeepMind a few years ago https://lnkd.in/gx7VU8aA Meta is not a dark horse in AI anymore. They might be the top dog. https://lnkd.in/gRRJYuxt #llama2 #llama3 #llm #chatgpt #gpt4 #ai #ml

16

5 Comments

What I really like about Ahmed BESBES’s article is that he exactly describes what we aim to achieve with Weaviate — building end-to-end AI-native applications! https://lnkd.in/eWZJf3zS

53

2 Comments

StableMoFusion is a new framework for generating human-like motion using diffusion models. Diffusion models have shown great potential in motion generation, but current methods are inefficient and have limitations. The authors of StableMoFusion conducted a thorough analysis of existing methods and identified areas for improvement. They designed a tailored model that addresses these limitations, including: 1: Efficient network architecture 2: Effective training strategies 3: Reduced computational overhead Despite its promising performance, the model still suffered from "foot skating" (a common issue in diffusion-based motion generation). To address this, the authors introduced a correction mechanism to ensure foot-ground contact and natural foot motions. Key Takeaways 1: StableMoFusion is a robust and efficient framework for human motion generation 2: It addresses limitations of current diffusion-based methods 3: It produces high-quality motion while reducing computational overhead 4: It corrects for "foot skating" issues, resulting in more natural and realistic motion https://lnkd.in/gtv7qnnA

1

Learn about the main training strategies in #MachineLearning that #NeuralNetworks use to learn. #AI

3

Hybrid search, which has enjoyed a resurgence due to the rise of vector databases and Generative AI, relies heavily on Reciprocal Rank Fusion (RFF). By using RRF to merge results from keyword search, dense vector similarity, and other retrieval methods, hybrid search systems can leverage the strengths of each approach to provide more comprehensive and relevant results to users. For the non-technical like me, I put together an article that breaks down Reciprocal Rank Fusion in layman's terms 👉 https://lnkd.in/gZ9rYqZZ

30

2 Comments

A great insight from the #CEO of #NVIDIA. In the age of #AI disruption and opportunity how do we handle #rejection when your industry or customers does not see the full value of your idea or product? "Sometimes you have to ignore your customers(stakeholders)" - Jensen Huang. As an aspiring #tech #entrepreneur that has gotten more than his fair share of rejection over the years pitching a wide array of product ideas, many of those nos were disheartening, I am reminded that getting a no does not mean you are always wrong. I've learned so much from every no, far more than I've learned from every yes! My big #takeaway from these experiences is if you believe in an idea , enough to move ahead with great risk there is potential for a great pay off. Know the #problem you are solving for and know your #clients, know them better than anyone else, execute and fail quickly and learn and improve eventually the industry will catch up and that risk may pay off . #genai #venturecapital #entrepreneurship #startup

1

Coding with Claude 3.5 Sonnet is ~10x more efficient than GPT-4o or any other LLMs out there. The new Artifacts feature enhances user experience by generating and running code next to your chat. Wild - Try it for free: claude.ai #ai #anthropic #claudeai #llms #coding

5

1 Comment