Melissa (Dalis) Gadebusch

Went overboard building agents at the AGI-House hackathon this weekend in SF. The idea was to build a personal AI agent which acts as your executive assistant, has access to your calendar, can chat with your contacts on Telegram & Slack, and can book an Uber pickup for you. Didn't they say, "the next big thing will start out looking like a toy"? Hacked away a crewAI + Open Interpreter demo in a few hours with Rahul Chhabra, re-realized how broken the agent build/flow engineering toolkit is. Still! Played around with OpenInterpreter 01 and experimented with a lot of verbal prompting to build/train new skillsets. Built out a tool for the agent where natural language instructions to OpenInterpreter were fed, and those got converted to AppleScript commands on the fly. Cherry on top, bagged a podium finish at the hackathon! 🥉 Agents are the future. 🚀 Github in comments.

47

7 Comments

So much fun talking about the power of LLMs with Amazon Bedrock and Anthropic Claude. In summary, the ever-evolving landscape of GenAI, it's becoming clear that there won't be a single LLM (Large Language Model) to rule them all. The rise of application-specific LLMs heralds a new era, where various models will carve out niches to excel in. This diversity opens up a wide playing field for innovation and tailored solutions. When it comes to deploying AI models, striking the right balance in evaluation is paramount. We must weigh factors like cost, speed, and precision to ensure optimal performance and resource utilization. Finding this equilibrium ensures that our GenAI solutions are not just efficient but also effective in real-world scenarios. Anthropic Claude emerges as a frontrunner in the realm of LLMs, distinguished by its exceptional capabilities and groundbreaking advancements. Boasting unparalleled scale, efficiency, and adaptability, Anthropic Claude represents a pinnacle of GenAI innovation. Its ability to seamlessly integrate with Amazon Bedrock amplifies its prowess, providing users with an unparalleled ecosystem for knowledge integration and generation. With state-of-the-art architecture and a commitment to pushing the boundaries of what's possible, Anthropic Claude stands as a testament to the transformative potential of GenAI in shaping the future of technology and beyond. Prompt engineering holds the key to maximizing the potential of applications. Crafting prompts that consider persona, context, tags, and positioning can significantly enhance the model's ability to comprehend and generate relevant responses. This strategic approach empowers AI systems to better serve their intended purposes. In the realm of GenAI architectures, a robust knowledge base is essential. Amazon Bedrock stands out as a key component for architectures like RAG (Retrieval-Augmented Generation), providing a solid foundation to fuel intelligent decision-making and content generation. As we march towards AI integration, bridging the gap with legacy applications is crucial. Agents equipped with open APIs offer a seamless way to facilitate communication and interoperability, ensuring a smooth transition into the AI-driven future. Let's continue to explore, innovate, and collaborate as we navigate the exciting frontier of GenAI together! #aws #LanguageModels #Innovation #TechTrends #AIIntegration #webdevcon #claude #bedrock

68

Day 3/3 GenAI Summit San Francisco Causal AI, causal grounded RAG - cRAG, scaling from POC to production in AWS, a deep dive into the evolution of Llama models and what's coming next, synthetic data in the age of genAI, PII, and domain randomization. I can't count how many times I've heard the words AI, genAI, data, cloud, RAG, LLM, and multimodal in the last 3 intense days. And when I finally went to have lunch at a restaurant, at the adjacent table, a fintech AI startup founder was pitching to a VC, while the people at the table behind me were talking about data engineering. It feels so good to be in SF, where at the airport’s security check-in a big Anthropic advertisement greets you! 😊 The last 3 days were amazing, filled with inspiration and thought-provoking new tech. I can't wait to get back to building DiffuseDrive and working with our customers!

22

Last night I had the fortune of representing LaunchDarkly at Sourcegraph's CodeAI Summit NYC. An important theme I took from last night's event? Context. Models like specifics. Computers can't read our brains or know what we want to achieve. Technology is kinda useless without the right context. As we move forward in this increasingly AI-generated world (for better and for worse) let us not forget the importance of adding context to the equation. Whether that's providing context to a dataset through how it was collected, sharing context of your repo to better aid a code assist tool, or simply providing context behind the why of our decisions. It's an interesting (and important) place to keep in our minds. #AI #MachineLearning #DeveloperAdvocacy #DeveloperEvents

14

1 Comment

Had a great time at The Knowledge Graph Conference last week! Here are my takeaways: Not surprisingly, there was a ton of presentations and talk about GenAI and how LLMs and KGs can work together, primarily around retrieval-augmented generation (RAG). Some highlights for me: • Jon Stevens showing how they're implementing RAG in really innovative ways at AbbVie. • Thomas Smoker showing how they're doing RAG as a product at WhyHow.AI. Tom gave some insight on how constructing KGs with the goal of RAG can be different than constructing KGs for other uses.  • Jesús Barrasa from Neo4j talking about how you can combine vector based retrieval and prompt generation for what he called “contextual RAG”. • Juan Sequeda's presentation on data.world's work on benchmarking LLMs and how KGs significantly improve the accuracy of LLM results. • TopQuadrant, data.world, metaphacts GmbH, and Semantic Web Company (SWC) all announcing and demoing different RAG or LLM-assisted products But I couldn't help but wonder whether there is too much of an emphasis on LLMs right now. KGs can do a lot by themselves and introducing LLMs may improve them, but may also add risk and more inaccurate answers - what Tom Smoker referred to as 'context poisoning'. Some notes on this: • There are use cases that clearly align with LLMs+KGs, like report generation. As Brian Martin pointed out, the single biggest use case for KGs in life sciences is regulatory compliance. Brian mentioned that automating 3 or 4 reports saved them $10m! • But not all use cases require LLMs. If the questions are more about the retrieval than the generation part of RAG, you can just use a KG and ensure much higher accuracy than if you trust an LLM to interpret the results.  • Also, as Jesús Barrasa pointed out, you can do similarity in KGs WITHOUT vectors (Jaccard index for example). • Further emphasizing the power of reasoning and KGs, there was an entire panel on this topic called, 'Can Reasoning Rule in the Real World?'. For highly regulated industries, everything must follow strict rules and be fully explainable to regulators. Lily Wang from Barclays pointed out that the rules that her bank follows were the result of 100s of years of hard learning. • Also, loved hearing from Thilanka Munasinghe, Jennifer Wei, and Armin Mehrabian about how knowledge graphs are being used by NASA - National Aeronautics and Space Administration and Rensselaer Polytechnic Institute to combat climate change. • And, of course, loved to see a lot of love for SHACL throughout the conference! My colleagues Norm Friend and Holger Knublauch held a masterclass but it also came up in Tom Smoker's presentation on RAG - sometimes it is better to give an AI agent a graph that assumes a closed world than to let it think too much. #tqideas

46

There was a time when people paid $200-$300 for a single prompt. Now, LLMs themselves are better prompt engineers than humans. It’s interesting how quickly things change in less than a year. LLMs went from small 7-50B parameter models, all the way up to +1T parameters (parameters determine how “smart” the LLM is). With it, the bar for writing good prompts has dropped to nothing… P.S. For those shaking their heads, yes, LLMs in production still require significant work. Open-Source LLM prompting remains tough and requires thorough testing.

10

2 Comments

A most intriguing quote: "As Sutton said in the Bitter Lesson, there're only 2 techniques that scale indefinitely with compute: learning & search. It's time to shift focus to the latter." - It is indeed time to shift the focus to #search I believe Jim is quoting from this March 2019 article. https://lnkd.in/gar3MRjf It is a good confirmation that many are coming around to understand the value of search (in all it's forms) in 2024 and into the future! Everyone is a Searcher. Those who search twice are called Researchers. Those who search thrice, go a step beyond... - JTK.

4

This is f*cking hilarious: TuSimple's pivot from self-driving trucks to AI animation and gaming! 🤖🎮 Right now, everyone’s betting on humanoid and general-purpose robotics startups. But after 5 years in self-driving tech and building Kopernikus Auto, here’s the reality: most humanoid robotics companies will be dead. In the self-driving world, we’ve only seen a few real winners: Tesla, Waymo, and Comma.ai. Why?  - Humanoid robotics isn’t a hardware problem; it’s a software problem. - You either have infinite data, cash, talent, or a clear use case—or you won’t make it, no matter how much you raise. From over 100 companies, almost all in self-driving tech are dead or miles away from profitability. ✨ Don’t get distracted by the appeal of humanoids. Focus on real problems you can actually solve. What are you focusing on? 👇

13

1 Comment

At SandboxAQ, we believe that utilizing large quantitative methods could have a major impact on the future of tech. While LLMs are definitely having their moment in the sun right now, I think we'll see even more value from these LQMs in the near future. One sector in which this technology can make a major impact is the biopharma industry. Here are a few benefits of LQMs: ➡️ LQMs can accelerate drug discovery with faster, more accurate predictions for lead generation and hitting targets. ➡️ LQMs can lead to generative AI-based output of new drugs optimized for multiple objectives including synthesizability. ➡️ LQMs can produce causal reasoning over multi-omics and literature-omics data to model biology and address scientific hypotheses directly. #LargeQuantitativeMethods #LQMs #Biopharma

14

"To tackle this issue, we developed a Whole Page Ranker model which determines the optimal positions for the new content on the page. The objective of the model is to ensure that we show highly relevant content to the users while also balancing revenue objectives. By dynamically adjusting the layout based on the content type and relevance, we could also present information more intuitively." #computationalinteraction #computationaldesign

1

Today, I found a fascinating way to think about “Tokens” in LLMs. It came from a conversation between Jensen Huang and Patrick Collison at a recent Stripe event. When asked about the future of compute capacity in five years, Jensen smartly dodged that question a bit, but then gave an excellent explanation and analogy about why tokens are the new forces that will power the next decades of humanity. He explained that we're now producing something unprecedented (and at scale): floating point numbers that possess “value”, which we now call tokens. These tokens are valuable because they encapsulate “intelligence”. People are now taking these tokens and transforming them into English, French, images, videos, chemicals, proteins, robotic movements, etc. And many are working hard to expand the range of concepts and ideas we can create with these tokens. He then goes on to made a compelling comparison between tokens and electricity. In the past industrial revolution, we successfully found a way to convert “atoms” into “electrons” (by boiling water to power electricity turbines). And now, we've discovered a way to convert "electrons" into "tokens" (by using energy to power data centers that train and run LLMs). When electricity was first introduced, few people understood its value. Today, paying for kilowatts is routine. The same will happen with tokens. Right now, only early adopters and builders are paying tokens. Soon, everyone will be paying for tokens on a daily basis to supercharge productivity and power new products and services. Many new industries will be born from and built on top of tokens. When I first heard about this way of thinking, I got goosebumps. Perhaps it’s because Jensen has a good storytelling skill that helps him sell what he builds, but I have to admit this way of thinking gave me a brief surge of pride. I'm proud of humanity's collective effort and creativity, which have taken us from living under rocks to building machines that can "think." That's nothing short but a miracle.

18

5 Comments

More GenerativeAI Hype warnings coming from Wall Street... Note that my own company OfferFit is nothing like the GenerativeAI crowd that Wall Street is panicked about: OfferFit has clear ROI for every customer; OfferFit takes home humans out of the loop and finds the best action for your customers with Reinforcement Learning - we do not mess around with chatbots! https://lnkd.in/gRMTebXf

15

Been thinking about how LLMs/Transformers and teaching kids to read are related lately. I've been listening to "Sold a Story" Podcast: https://lnkd.in/e_9fbZ5t about the Reading Wars in schools. To oversimplify, it contrasts phonetic learning vs whole word/queuing methods to teach kids how to read. Highly recommend it if you have kids. As I've listened to the podcast it makes me think about NLP/GenAI concepts, like tokenization, embeddings, attention, transformers. Transformers seem like queuing methodology and tokenization seems related to phonemes. Definitely some nuance missed here. I've always thought the english language and probably others are extremely cumbersome for trying to get thoughts/emotions out of one's head and to the others (especially as an introvert!) Not sure there is a point to this post other than to wonder what is the most efficient mechanism for communicating with other people as both a knowledge transfer vehicle and an emotional one. I also need to do some research around how cognitive scientists in the reading field have related their work to the Transformer/Gen AI world.

13

2 Comments

Big thanks to Jillian Forde for inviting me to chat about my #YCombinator journey on #AWS-ON-AIR! Quick summary about our chat: 1. YC is a game-changer for first-time founders—think fewer rookie errors, faster product development, and sharper GTM strategies. Plus, it boosts your fundraising game. 2. YC isn’t just a three-month sprint; it’s a marathon with a lifetime network and friendship of fellow founders. 3. Heads up: snagging a spot in YC isn’t a golden ticket. It doesn’t guarantee a successful exit. Keep running! Check full podcast at https://lnkd.in/g2YJbSFw

31

1 Comment

Better technology won't drive the next AI wave of innovation. Your indomitable spirit will. Last week hit different. Our internal AI agent - our flagship innovation project - was showing a steady decline in weekly active users. Double digit drop since Q1. The metrics we'd celebrated at launch were now keeping me up at night. My typical playbook would scream "push harder, iterate faster." But sometimes, an indomitable spirit isn't about relentless pushing - it's about the courage to pause and face uncomfortable truths. Let me be vulnerable: After two decades leading tech teams, my biggest blind spot has been mistaking motion for progress. Especially with AI products, where the temptation to continuously iterate can mask a deeper truth: we're not solving the right problem. The hardest moment? Standing before our leadership team and admitting: "We're not ready." We need to pause, rethink our approach to building, and become better listeners. Today's market rewards speed, but it reveres wisdom. Your indomitable spirit isn't measured by how hard you can push - it's measured by your courage to pause when that's what your customers need. Thank you to the many who are helping me grow in this journey. 🙏 Special shoutout to Daina Souza Oenning, Timucin Uygungil, Phil McHugh, Matt Kiernan, Miguel Rodriguez, Sowmya S., Anne-Marie Costigan, Laurie Aquilante Faiola, Elaine Bermingham, Tony Abiera, Kelsey Balimtas, and Michael Renahan.

85

5 Comments

Meta's Llama 3: A watershed moment for Open-Source LLMs? Exciting news in the tech world as Meta's launch of Llama 3 last week (https://lnkd.in/gk7VhheQ) is redefining the open source LLM landscape! 🚀   While 2023 saw remarkable strides with Llama 2, Falcon 180B, and Mistral 7B, Llama 3 is setting new standards with its unparalleled raw performance. The larger 70B version is now only second to OpenAI’s GPT-4 Turbo on the LMSYS leaderboard for English prompts (https://lnkd.in/gce2yKhy), and the smaller 8B model outperforms GPT-3.5, which powers the free version of ChatGPT. 🏆   Users are raving about Llama 3's faster inference compared to GPT, making it a game changer in the realm of super-fast image generation. Its integration with Meta products allows for real-time modifications with prompts, vastly enhancing user experience. Since it is open source, folks on X have even shown how incredible it is when paired with Groq!🌐   What's fascinating is that Llama 3 used ~75x more data (15T compared to ~200B) than the Chinchilla paper's benchmark (https://lnkd.in/gzv5H3FN), and Meta claims it still wasn't converging. This underscores the vast potential of data in enhancing models with sub-100B parameters. Yes, the training costs are higher, but the lower parameter size reduces inference costs, paving the way for running LLMs on local devices in the future! 💡   The only downside? A context length of just 8k when 128k is becoming the norm across top models. But hey, every innovation has room for improvement, right? And they have indicated that this is bound to increase in the coming months as well!   Kudos to Meta for pushing the boundaries of open-source development with Llama 3. Its impressive performance, combined with Meta's product integration, makes this a development worth watching! Do you think the 400B+ model (summer launch expected) would beat OpenAI's GPT 4 Turbo and maybe even be as good as GPT 5?🎉   #Meta #Llama3 #OpenSource #AI #LLM #TechInnovation #MachineLearning #GenAI #GPT4 #OpenAI

25

5 Comments

Really cool chart on hallucination mitigation techniques for LLMs. The most common techniques I've heard from enterprise builders are retrieval augmented generation and supervised fine tuning - both enable companies to build on top of their proprietary data and therefore attain a moat on the performance of their specific use case. Haven't seen anyone leveraging these other techniques, but if you are, would love to connect and learn more! Check out the paper here: https://lnkd.in/e8k5Accc

25

1 Comment

If you are like me and have to send multiple Venmo payments and requests each month for things like streaming subscriptions, rent, etc, the fact that Venmo doesn't have a feature to set up recurring payments can be quite the disappointment. To get around this, I've created a solution that leverages AWS resources such as Lambda and EventBridge Schedules, along with the Venmo API, to automate recurring payments and requests. If you have an AWS account and a passing familiarity with Python, Terraform, and working in the terminal, you will find this solution incredibly easy to deploy. If you’re not familiar with the tooling, this project is simple and straightforward enough to act as a great introduction to something like Terraform. And, assuming you aren't automating 100s or 1000s of payments a month, it's essentially free. This was a fun personal project to work on. Let me know if you end up finding it useful! I currently have about 8 automated recurring payments/requests which I no longer have to keep track of, so it's definitely made my life just a little bit easier.

46

4 Comments