Simon W.

Building a reliable RAG system isn’t just about performance—it’s about trust. Matt Turk's blog post breaks down the importance of this for both technical and non-technical folks—highlighting how Cleanlab's TLM, paired with Pinecone’s vector database, delivers high-quality data retrieval, accurate, context-aware responses, and scalable performance to meet growing demands. 👏

7

Stop the madness around modifying hundreds of components manually in your Cost Segregation studies. Use SegStream's "bulk select and modify" instead and save yourself tons of time when depreciating, labeling, or reclassifying building components. Find out more at https://segstream.com/

1

Just like the big boys, bookend.ai now supports Llama3, alongside our menagerie of open source LLMs. While Llama3 is great, it's also just the tip of the iceberg. What will truly separate the herd from the herd-wranglers is going to take a lot more work than house training a llama. Real value lies in automating the real-life, workhorse apps that power the enterprise and that requires maximizing both safety and simplicity without compromising accuracy and flexibility.

16

Chasing the perfect Software QA plan is WRECKING your speed—and your product. Unfolding (thanks, Rob Snyder, for introducing me to the concept!) is the idea that things gradually take shape over time, and you make adjustments along the way. Instead of trying to plan every detail upfront, unfolding allows room for flexibility, learning, and growth as you progress. The key is to keep moving forward, knowing that your path will become clearer as you take steps and respond to new information or challenges. In QA, that means starting testing with WHAT you have, testing early, and refining your strategy as new information surfaces. Start testing NOW, even if your strategy isn’t flawless. Test and adjust/improve your strategy as you learn. That’s how you catch critical issues faster and keep momentum. Perfect QA plans don’t exist. Unfold your testing strategy in real time .... or fall behind.

20

1 Comment

Sharing for reach: Dave Snowden is holding a 2-day "Cynefin & Sense-making" workshop in Seattle in November. https://lnkd.in/g-Ye4xde Along with Simon Wardley's value chain mapping, Dave Snowden's sense-making tools have impacted my thinking tremendously over the last several years. (I imagine both of them might respond to that sentiment with, "You need to get out more," which may be true.) While I recognize there are many valuable techniques out there, I have to emphasize how deeply these approaches have shaped my thinking. They've consistently brought clarity and results in ways that have resonated with me on a personal and professional level. If you have the opportunity to attend, I think you'll learn something fantastic. #cynefin #sense-making #sensemaking #complexity #complexadaptivesystems

3

Having trouble moving the needle on AI implementations? You aren't alone. I highlight the top challenges to adoption in my latest Substack (linked below). Want to read it first, or see it in your inbox? Be sure to subscribe! https://lnkd.in/dhKAGwvU

26

2 Comments

New Overture Maps Foundation release for September in Fused (data hosted by Source Cooperative!) It's very easy to explore this data yourself at https://lnkd.in/g5ptFA6v -- just click "Add to UDF builder" to start. In this video I'm comparing this release and the past release's address data.

149

6 Comments

I'm having enlightening convos with folks who deliver infra for others to run. It sounds like "runbooks/wiki docs" are a real problem from a handoff/support perspective. We have an elegant solution, when you deliver runbooks as "click to play/pre-configured" Notebooks, and a command fails, the user gets an `escalation button` that shares a link to the output/metadata into the shared Slack room. It's ideal for bridging the gap when there's a support relationship. 🤓 This sounds like a case study in the making.. anyone else struggling with this?

14

A little while ago Cursor AI was super popular and I ended up checking it out to see what all of the fuss was about. I realized they were doing pretty much the same thing as Github Copilot was doing except they allowed 2 new features: - Allowing the AI to automatically inject the script in the places they belong - Letting the AI view multiple files at once. Cursor AI was even just straight up using a reskinned VS Code. So, I figured that Microsoft would eventually implement the same technique in probably the same way. The way the Cursor team was doing it was a simple string replace, so there was like 0 tech moat. Well today I needed to lookup an error I had not seen before, and lo and behold, Co-pilot allows for "Apply in Editor". My example just changed the whole empty file, but if you already have code, it will insert new code at different locations in different files and allow you to accept or discard changes. So... might as well drop your Cursor subscription which is like 20 bucks for copilot which is 9.

22

3 Comments

Great video Boston Dynamics that includes a nice demonstration of recovery from error (e.g. engine cover getting snagged). The text describes ML vision for SLAM, but curious how Atlas is using proprioception, full-body haptics and touch to detect the physical interactions. As humans we know when to let go of something based upon touch and haptics. Vision enables planning, touch is critical for reacting. Great to see Atlas doing some “hands-on” work! BTW: Kudos for making it clear whether this is autonomous, teleop, real-time etc. Too many videos of humanoids don’t make it clear.

5

Enums enhance code readability and clarity. They define named values for model attributes, making them ideal for representing statuses, stages, and states. An article from the team at reinteractive explores the powerful features of enums, particularly how they make code readable and self-explanatory … https://buff.ly/4cKGWiu #RubyOnRails #Programming #Enums #reinteractive

1

Did you miss my final #RAGHack stream, "Evaluating your RAG chat app"? Catch the recording here: 👇 https://lnkd.in/gSd32wXm I covered: 🤔 What does it mean for a RAG answer to be high quality? 🔍 Which factors affect RAG answer quality, and which affect it the most? 💻 Using promptflow-evals SDK to run GPT evaluators in Python 📊 Using my ai-rag-chat-evaluator CLI tools to generate synthetic ground truth data, run batch evaluations, and compare metrics across runs 🤷🏻‍♀️ Strategies to make sure your app can say "I don't know" at the appropriate times Grab slides from https://lnkd.in/g-9xSQvV #rag #openai #python #evals

20

Prebunking???? 💡 Technologists are putting prebunking training in our feeds? Good for them! Check this out, an entire department in Google called Jigsaw. Take the quiz, how did you do? Here are some examples: 👉 Emotional Manipulation: “The judge issued a DISGUSTING ruling today. It will make your blood boil.” 👉 False Dichotomy: “If you vote for X, you won't be able to pay your bills." 👉 Cherry Picking: “Hot air balloons prove gravity is not real.” 👉 Fake Experts: Nutrition advice from a fitness influencer who lost 50 pounds 👉 Red Herring: “Why are you upset? Think of all the starving children around the world who have it worse than you.” 👉 Scapegoating: “Unemployment is high because the economy is controlled by one ethnic group.” 👉 Ad Hominem: A headline calling attention to a candidate’s "screechy" voice during a rally instead of addressing the person’s arguments. 👉 Polarization: “Don’t believe the other party’s lies. They said they care about workers but more people are out of work than ever.” 👉 Impersonation: Impersonation attacks include, but are not limited to, setting up fake social media accounts associated with a target, accessing a target’s online accounts to send messages, and spoofing the sender email address or phone number of a target to make it appear as if the target authored the message. 👉 Slippery Slope: “If we implement even a modest change in policy, we’ll lose all of our freedoms.” 👉 Decontextualization: A photo of damage from a natural disaster posted with a misleading commentasserting the damage was caused by a recent protest Let's begin labeling our feeds with these labels! https://lnkd.in/dgEaTY9V

2

1 Comment

I added bigint bindings for h3-duckdb https://lnkd.in/gjcJnvaN this past weekend. This makes it easier to work with the performant, 64-bit integer, representation of H3 indexes, regardless of how you choose to store them in Parquet. Since H3 doesn't use the sign bit, I've seen some people use unsigned and some use signed. I usually use unsigned - which h3-duckdb has always supported - but I want to ensure people who chose signed get the same experience.

68

2 Comments

How did you meet your co-founder at LendAPI? It’s one of the most frequently asked questions, and I truly believe the stars were aligned at that time. I feel blessed to have Tim as a partner on this journey. Here is the story: https://lnkd.in/gDeFYQG8 #fintech #startup

22

Been prepping for next week's panel discussion with Eva M., Thomas Soederholm, Nick Sinas, and Dileepa Prabhakar. I'm curious... New research shows that traditional hardware projects are now predominantly software projects, with 60% of total spend going towards software development. Have you experienced this? Our discussion will tackle questions like: Are teams set up and resourced correctly to deal with this shift? How will this transition impact development cycles and product ship schedules? Will it drive more hardware products to deliver new features using ‘as-a-service’ business models? What q’s would you add to the discussion? Join the conversation - RSVP here: https://lnkd.in/eSzAX7ai

44

1 Comment

Yesterday I thought, "Oh wow, what a lack luster release from OpenAI, surely people will realize in droves that OpenAI is slippin". But I was very wrong. All I have seen is people starry eyed drooling all over it. I'm semi skeptical that it will hold up. Like with previous releases there was initial excitement but then "all of a sudden GPT sucks" and people complained about everything under the sun about it. But, in general, from all of these videos all I can see is an AI explaining things to people like they are five-year-olds and then an unbearable amount of yes man platitudes. Don't get me wrong, audio, video, and text processing will be the foundation of all AI moving forward, but please calm down. Please tell me how your life is going to change with an AI looking at stuff with you at all times and why you are so giddy over it. Maybe there is a disconnect between my vision of what is possible and how long it will take, and what the general public's vision is.

42

21 Comments

The Weapon: Targeted Discipline Tech debt grows in a codebase the same way we gain unwanted weight—it’s all about how our brains prioritize short-term over long-term goals. When hunger strikes, health goals vanish, and the nearest donut meets an untimely death. Similarly, under pressure to ship features, good coding practices are replaced by poorly tested spaghetti code. The issue? Our brains are wired to solve immediate problems, often at the expense of long-term outcomes. The solution? Targeted discipline. Discipline is tough to instill and even harder to sustain, but it’s the key to keeping tech debt under control. Elite athletes like Floyd Mayweather know this well—he’d jog home from the club at 3 AM in jeans. Fun? No. Hard? Absolutely. Effective? You bet. At CodeIntegrity, we’re building developer tools to embed this discipline into software development, empowering capable, motivated teams to reduce technical debt and sustain long-term success. Because in the end, targeted discipline isn’t just a weapon—it’s a superpower. —Abi

9