Mike McCandless

Considering the rapid advancements in Large Language Models (LLMs), it's becoming clear that for 95% of companies and domains, the underlying LLMs will soon be a commodity provided by major players. The real play for these companies will be at a higher level, focusing on building neural agents and advanced neural planning. Imagine this: planning what needs to be done by neural agents will itself leverage LLMs, becoming your unique edge. For most, fine-tuning an LLM might be necessary, but the true focus should be on higher layers. Think about this: If ChatGPT, Gemini, and Claude are all trained on relatively the same data from the internet, why do they understand and respond to your queries differently? It's their planning skills that give them superpowers. Neural planning will be the research focus for years to come, until it too becomes a commodity offered by the big players.

34

2 Comments

Very insightful work by Anthropic’s interpretability team. And an amazing paper, with outstanding writing and figures. The idea is very simple: interpret LLMs by leveraging sparse autoencoders as surrogate models of the MLP of transformer blocks, which allow one to disambiguate the superposition of features captured by a single neuron. A simple idea, but a very careful and complex execution, as it is often the case in our line of work. The paper goes into many details and provide a large array of insights, although the gist of the implementation remains obfuscated due to the closed source nature of Claude. Too bad, because this is the kind of work that we need to better understand and eventually trust LLMs. This is demonstrated by the authors in the section ‘Influence on Behavior’, where they show that clamping some features to either high or low value during inference is “remarkably effective at modifying model outputs in specific, interpretable ways”. Hopefully this kind of work is going to be replicated and generalized to open-weights models, such that we have new ways to steer their behavior. https://lnkd.in/eVym7f_f #interpretability #xai #explainableai #steerableai #anthropic #claude #anthropic

3

Sharing my experiment on creating a Streamlit based chatbot, that combines inputs from Agentic RAG using llamaindex (with openai and MistralAI), and langchain google genai (with Gemini-1.5-Flash), to extract information from images embedded within the PDF files. Was a nice experience to build this chatbot. Still many open questions, ways to improve the solution, in a production scenario..but nice to combine features from multiple libraries and come up with a solution. Please check it out! Thanks. #chatbot #gemini #openai #langchain #llamaindex https://lnkd.in/gPdtxhQF

19

The latest advances in artificial intelligence (particularly large language models) continue to reverberate. Even for an “old school” AI person like myself (who cut his teeth with Prolog) it is clear that there has been a step change in our ability to create computer systems that can interact with humans using natural language. GPT-4 et al are exhibiting early signs of “common sense” and have encoded useful conceptual representations of the world. The debate rages on as to whether this is “intelligence”, but to an engineer like me, it sure seems useful! #LLM #AI #Intelligence #Programming https://lnkd.in/ebAgNapK

14

4 Comments

Abstractions leak in the worst case. So it's always good the low level details behind the abstraction. Despite this, abstractions are super important: What if there was no abstraction in computer networks. Imagine every backend developer implementing TCP in their web server. It would be total chaos!

8

2 Comments

Nice post on nature talking a bit about what we are doing here at Osmo. I am excited to share more soon about what we have done since the early research. https://lnkd.in/d5W-d98y

51

1 Comment

With NVIDIA AI Foundry, enterprises and nations can supercharge #generativeAI with a service spanning curation, synthetic data generation, fine-tuning, retrieval, guardrails, and evaluation to deploy custom AI at Meta Llama 3.1 NVIDIA NIM Microservices.

1

Since January, I've been up to my eyes in retrieval-augmented generation and large language models. It's the wild west of technology 🤖🤠 With so little guidance publicly available, the learning curve is steep! Our team frequently crawls ✨ Arxiv ✨ for inspiration because academia is leading the way 📚 My mistakes don't have to be yours! Here are 5 lessons I've learned the hard way: in production 😅🚀

3

Sorry for the delay, but Joachim and I are back with, dare I say it, my favorite episode of Learn, Make, Learn yet! It's focused on the intersection of generative AI and product creation. This is a thorny topic and, fortunately, we're joined by Anmol Anubhai, who brings deep expertise in the field to our conversation. Her human-centered approach to the technology—one rooted in amplifying humans instead of replacing us—actually converted me from an AI-sceptic to a (cautious) AI-optimist. I hope you'll join us for a refreshingly grounded conversation on a topic that’s typically full of hype. You can find “Learn Make Learn” at any major podcast directory or listen at https://lnkd.in/gGv-tE6F And don’t hesitate to share your questions & perspectives here! #LearnMakeLearn #ProductMarketing #ProductManagement #ProductInnovation #GenAI #GenerativeAI

60

5 Comments

Large Language Models (#LLMs) are evolving rapidly, and the demand for running them at scale is skyrocketing. #Ollama is stepping up to meet this need with its official container image on Docker Hub, excelling in local #deployments. However, there’s a catch for containerized environments: downloading #models requires the ollama serve command to be running within the #container. This extra step complicates automated deployments via #CI #CD #pipelines, making seamless model access challenging. To take your #model to production, it needs to be #container-ready—a feature Ollama has yet to fully support. The gap in #AIOps for smooth deployments gets blurred with the inference services available as an easy way to take it to production. Goal is to overcome obstacles and deliver solutions 😉 so here is to get you started on having your (O)#llama #production ready in a private #deployment from your #local to #cloud all in #oneshot . Enjoy Medium: https://lnkd.in/g9jqBQ8d dev.to: https://lnkd.in/gi_gGwam #AI #LLM #Ollama #llama3 #phi3 #mistral #S6overlay #AIOps #CI #CD #oneshot #container #docker #podmanEnjoy

11

An interesting hybrid approach for conditional generation that leans into the strengths of both diffusion and autoregressive models - to produce text with a set of desirable properties: https://lnkd.in/eeiXR_MC It would be interesting to see this explored/evaluated further, across more use cases (and domains).

15

If you're seeking guidance in responsible machine learning, here is a comprehensive and valuable list of resources. #DataScience #MachineLearning #AI #BigData #Analytics #DataAnalytics #DataMining #DeepLearning #Python #RStats #Statistics #DataVisualization #BusinessIntelligence #ArtificialIntelligence #PredictiveAnalytics #DataDriven #DataScientist #DataEngineering #NaturalLanguageProcessing #ComputerVision

3

Microsoft’s new GraphRAG tool is changing the game for data discovery. Using a graph-based approach, it surpasses traditional RAG in finding and retrieving complex information. Amazing news is that, Microsoft has made it public now. Perfect for tackling big questions and deep data dives. Check it out on GitHub and see how it can boost your insights! https://lnkd.in/d7Gu-H_E

30

1 Comment

From Black Box to Blueprint: Inside the World of Retrieval-Augmented Generation Ever wondered how large language models like me seem to magically access and process information beyond their training data? The answer lies in a fascinating technique called Retrieval-Augmented Generation (RAG). Today, we'll peek under the hood and explore the inner workings of RAG, understanding how it empowers language models to generate more informative and accurate text. #RAG #Retrieval_Augmented_Generation #Under_the_Hood #LLM #AI #DeepLearning

14

1 Comment

There's been a growing focus on effective benchmarking methods for large language models. Came across an interesting study introducing a new approach that utilizes LLM debates to assess model performance. The method involves creating an LLM debate framework and using an LLM judge that analyzes debate arguments based on criteria like clarity, factual accuracy, counter-arguments, and persuasiveness. This eliminates the need for human evaluation, making it more scalable for LLM development and evaluation. Could this be the step towards LLMs acting as judges in the future? Are we ready for AI referees? 🙂 #AI #machinelearning #LLMs #naturallanguageprocessing https://lnkd.in/g7A8pEUW

14

2 Comments

Interesting paper showing that LoRA doesn't learn your data that well which raises questions on how effective it is at fine tuning your Gen AI model. LoRA Learns Less and Forgets Less - https://lnkd.in/eG4tEuSe

19

You know about Tome's awesome AI slide generation tool. Now we’re building a platform for sales teams, powered by our own custom LLMs. We need Applied Researchers and ML Engineers to help us make this happen! Want to join us? Over the last few months, we’ve sharpened the focus of our company to enterprise customers, and the focus of our highly successful slides tool to the most valued presentations in the world: the ones that sales teams use to win new business. We’re building a sales assistant for the work that goes on before that presentation can be created. The deep research to understand your customer, what they care about most, and which aspects of your product can help them succeed. Our domain-specific AI approach outperforms generic LLMs. It’s better at business intelligence, buyer research, and value articulation than general purpose models. And we’re growing our ML team. If you’re inspired by the challenge of building innovative new AI products for people doing serious work, check out these roles and let us know! - Machine Learning Engineer - MLE, Applied Research & Model Development - MLE, GenAI Platform Want to learn more about the next chapter of Tome? Our new site is now up here:

46

3 Comments

When looking for a vector database, choosing the right vector search algorithms to establish the proper trade-off between accuracy and speed of vector retrieval is challenging. This article demystifies the foundational concepts of vector search so you can set up your vector database successfully. https://lnkd.in/e-9tfSfh #VectorDB

5

Overcoming the Spark Challenge: My Journey into PySpark I've just completed the "Introduction to PySpark" course on DataCamp, and I'm thrilled with the progress I've made. For the longest time, Apache Spark has been this intimidating behemoth in the world of big data processing that I've been hesitant to approach. However, taking this course has been a significant breakthrough for me. Key Takeaways: 1. Demystifying Spark: The course helped me understand what Spark actually is - a powerful, distributed data processing engine designed for big data applications. 2. PySpark's Role: I now grasp how PySpark serves as a Python API for Spark, making it accessible to Python developers like myself. 3. Practical Applications: I've learned about the various use cases for Spark, from data transformation and analysis to machine learning on large datasets. 4. Core Concepts: The course introduced me to fundamental concepts like RDDs (Resilient Distributed Datasets), DataFrames, and SparkSQL. 5. Performance Benefits: I now appreciate how Spark's in-memory processing and distributed computing can significantly speed up data operations. 6. Machine Learning Pipelines: While transformers and estimators still feel like a mystery to me, I am glad I got an introduction to them and feel comfortable exploring them further. Moving Forward: This introduction has given me the confidence to further explore PySpark and its capabilities. I'm excited about the possibilities it opens up for handling large-scale data processing tasks efficiently. Some areas I'm keen to delve into next include: - Advanced data manipulation techniques in PySpark - Optimizing Spark jobs for better performance - Setting up and managing a Spark cluster Overall, I'm glad I took this first step in conquering my apprehension towards Spark. It's transformed from an intimidating technology into an exciting tool that I'm eager to master and apply in real-world scenarios.

14

2 Comments