Todd Fast

This article provides a helpful, straightforward explanation of the options for improving the output of LLMs for specific types of use cases - worth a read.

4

1 Comment

I am excited to share the latest blog, co-authored by myself, Junhwi Kim and Vyas Sekar, detailing how we managed to successfully localized and improved LLM by combining Fine-Tuning and RAG. https://lnkd.in/gKDkgGPq

70

3 Comments

The most clear, fact-based, logically reasoned (and well-written) view on the the positive places that AI can take us from someone who really knows what he's talking about, Dario Amodei (Founder and CEO of Anthropic). This is the article I would have aspired to write if someone asked me to write about this topic. https://lnkd.in/g-xN7zRW

7

2 Comments

Your code is important, today AI can generate many generic components of code yet each application will have certain intrinsic elements that needs to be protected. How to secure your code when there are multiple parties involved? Introducing Castler Code, a product we recently developed that solves this problem by building code Escrows. Whether you are a big tech company outsourcing development or a freelance developer with multiple clients, you get to define the terms and conditions that work for you, keep your code in an Escrow, make sure it's running without actually sharing the code, define security layers and release it only when the terms are met. Like, follow our new page for more! Website: https://code.castler.com/

38

🌐 There’s been growing concern lately about AI "training itself" into oblivion by feeding on AI-generated content, as if we’re on the verge of a digital "mad cow disease." But this fear feels exaggerated. Think about how we, as humans, learn: we are, in essence, trained by other humans who learned from yet other humans. This recursive nature of knowledge doesn’t lead to our mental degradation; instead, it helps refine and expand our understanding. We build on prior ideas, just like AI models do with their training data. The difference is that AI processes far more data, far faster, than we ever could. So, just as humans don’t lose coherence by learning from those who’ve been taught before them, AI won’t suddenly collapse into chaos because it trains on its own output. We have safeguards, just like human education systems have frameworks, curriculums, and critical thought to prevent echo chambers. In both cases, structure and checks keep things in line. Of course, it’s wise to remain cautious. Just like how language evolves over time, symbols and their meanings can shift. For instance, the tilde (~)—often overlooked as a simple, almost trivial character—has a much deeper history. It was originally devised in the late 19th century by Matilde von Rothschild to represent an intricate cross-stitch pattern. Isn't it funny how something that once stood for precision and complexity is now used as a symbol for approximation? But we shouldn’t let the Matilde Mark’s origin be forgotten. In the same way that AI needs to understand its lineage to avoid losing context, we should pay homage to the origins of the symbols we use. Because sometimes, the smallest marks on a page carry the richest histories. #AI #MatildeVonRothschild #HistoryOfSymbols #CrossStitching #TechEvolution

6

2 Comments

🚨 Happening tomorrow at 10 AM PT! Discover how to develop, manage, and integrate workflows using your code and systems to optimize your agreement processes with the ✨ new ✨ #Maestro API (beta). Joining this webinar will provide you with the opportunity to: 👉 Learn Integration Techniques: Gain insights on how to seamlessly integrate the new Maestro API with your existing systems. 👉 Explore Use Cases: Understand practical applications and use cases to effectively use the Workflow Designer. 👉 Interactive Learning: Participate in a live Q&A and watch demonstration coding solutions in real-time. 👉 Early Access: Be among the first to join the beta program and start utilizing the new features. This session is perfect for developers and technical professionals looking to streamline their agreement processes with advanced API integration.

1

Gitlab makes their strategy document public. It provides a great inside view of what's driving their product development efforts. Here are my reflections:

18

3 Comments

This is a great video about how our personal ideology impacts the way we process data and information. For me, it shows the importance of fighting against populism and "framing" in the media and teaching our children in critical thinking. https://lnkd.in/eXQdB4fR

2

Was playing around with v0.dev before last weeks Next.js conf, wanted to stress test how much it could help someone build a complex feature in Catalyst. One of our merchants was looking to add a Gift Certificates section like Stencil has and we just haven't got to it yet within the out-of-the-box version. Was a perfect scenario! High level take? If you are building in Next.js you should 100% add v0 to your toolbox. It'll help you slice away at features much faster than any other AI solution in market. I include more notes in the PR along with a demo of what was built on GitHub: https://lnkd.in/eBm6wVdP

47

AgentEval, from Microsoft: A Game-Changing Developer Tool for Assessing LLM-Powered Applications AgentEval, is a groundbreaking framework that empowers developers to comprehensively assess the utility and effectiveness of LLM-powered applications. AgentEval leverages recent advancements in LLMs to provide a scalable and cost-effective alternative to traditional human evaluations. It consists of three main agents: CriticAgent, QuantifierAgent, and VerifierAgent, each playing a vital role in evaluating an application's task utility across multiple dimensions. Key Takeaways: 🎯 CriticAgent suggests evaluation criteria based on task description and execution examples 📈 QuantifierAgent quantifies application performance against each criterion ✅ VerifierAgent ensures criteria robustness and relevance through stability and discriminative power tests 🔧 AgentEval offers flexibility, scalability, and the ability to incorporate human expertise 🧪 Empirical validation on math problem solving and household task simulation demonstrates AgentEval's effectiveness AgentEval represents a significant leap forward in evaluating LLM-powered applications, providing developers with valuable insights to drive improvements and ensure applications meet users' dynamic needs. Excited about the future possibilities as AgentEval continues to evolve! #llm #gpt #evaluation #taskutility #developertool #AgentEval https://lnkd.in/gkYMXT22

4

In this episode of the Mob Mentality Show, join us as we delve into "The Discovery of Ignorance in Product Development: Time for a Smarter Way to Deliver Customer Value" with our special guest, Gary Cohen 🟦: https://lnkd.in/ggGMHyfu https://lnkd.in/gx2tXEu6 Drawing on his rich experience across diverse sectors—from government agencies to startups—Gary brings a wealth of insight into optimizing product discovery and delivery for true customer impact. ### **Key Topics Explored: The Product Discovery and Delivery Intersection** Even if a team is familiar with the 80/20 rule (e.g., where 80% of features go unused), how does it determine the right 20% to deliver? Gary shares strategies to tackle this challenge, advocating for a shift from traditional roadmap fixation to embracing dynamic, continuous discovery that keeps customer value at the forefront. Together, we examine the effectiveness of "strict" roadmap discipline (the practice of adhering to set timelines and objectives) and if it becomes a barrier to true innovation and customer value. We also briefly touch on the FAST framework and its relevance to product discovery, emphasizing how collective discovery through mobbing allows teams to stay closer to evolving customer needs. ### **The Role of GenAI in Modern Product Development** Is AI is reshaping product discovery? We explore how GenAI can be a collaborative asset for mobs, not only for code development but also for bringing value by uncovering insights into customer behavior and streamlining feature prioritization. We compare "introverted" AI and "extroverted" AI in mob environments and discusses how developing refined AI inquiry skills is key to leveraging AI’s full potential. From training mobs in nuanced prompt engineering to using AI to help drive a more lean delivery, we share ways to maximize AI's impact while maintaining expert human oversight. ### **Unpacking Behavioral Expectations: A Smarter Approach to Product Decisions** Beyond the technical, we look at the behavioral side of product monitoring, asking, “What responses do we expect from our users?” Gary emphasizes the importance of behavioral prediction in discovery, cautioning against information bias and providing ways to gauge responses to feature experiments in real-time. We highlight the importance of monitoring usage data and what it reveals about customer engagement—whether a feature elicits a “wow” or a mere “meh.” Chris Lucian

6

9 Comments

A good read on building with LLMs https://applied-llms.org/

11

★ Keeping your A-Team together with developer AI When DHH spoke about being able to scale from the small by reigning in complexity at his RailsWorld keynote it resonated a lot. I have seen the struggles as teams grow and coordination headwinds slow progress to a crawl. This is actually one of the areas that I am seeing AI help with. As I talk with Augment Code customers, a repeating pattern is how they are able to keep teams smaller, for longer, and I'm so here for it! I wrote more on what we are seeing: https://lnkd.in/gj8366-n #developerai

34

1 Comment

If you're a data person and spend more of your time worrying about how to get value from your data and less about how to make it scale, join us at Small Data SF, Sept 23-24. We've got absolutely badass speakers & sponsors, a who's who in data and AI. We've got one day of workshops, so folks can get hands-on with Small Data technologies and solve real problems, and one day of short-format talks that cover topics ranging from local-first AI models to data visualization at 60 frames per second.

79

Our latest paper, "Alignment For Performance Improvement in Conversation Bots," delves into cutting-edge alignment methods that significantly enhance the adherence of conversational agents to guardrails. Raghav Garg and team demonstrates how innovative techniques like IPO, KTO delivers superior performance for customer service agents while maintaining natural, coherent interactions. Read more about it here: https://lnkd.in/d2e8Rk6G

106

A successful systems design and delivery strategy begins by assuming that change is likely the only constant. Be adaptable, or fail.

2

1 Comment