Guy Royse

@bheimann My favorite was when he made a joke about it himself in one of the Expendables movie. "It is true that I was once bitten by a cobra. And after seven agonizing days of pain, the cobra died."

@guyroyse Even in death his memes will live on.

Chuck Norris doesn't rest in peace. Peace rests in him.

@dsawyer It's just half.

Half and half: half milk, half cream (i.e. butterfat) Fat free half and half: half milk, half... whatever. Oh hey, look at the birdie!

It's not every day that I get to quote Dave Thomas—the founder of Wendy's, not the author of The Pragmatic Programmer—but I do in this podcast with Roberto Perez. Dave Thomas said that the most important bites of a burger are the first bite and the last bite. Those are the bites you remember. Those are the bites that influence your reaction to eating that burger—in other words, will you get another in the future. LLMs, it turns out, do something similar. They tend to remember stuff at the beginning of the context and the stuff at the end of the context. And sometimes struggle with the in the middle stuff. So keeping context small matters. The less middle you have, the less chance the middle is missed. And, of course, a smaller context is cheaper and faster too. Fewer tokens. Less time spent processing them. So smaller contexts are, generally speaking, better. Of course, just dropping messages off the back of the conversation to keep a small context doesn't solve the problem. The context needs to be tight, not truncated. To do this, you need Agentic Memory. In the podcast, I demo a simple chatbot that uses Agent Memory Server to provide just that. Watch it for all the details: youtube.com/watch?v=fkdqwm…

@ssslomp Pretty sure this is always the reason why we can't have nice things.

ok community notes was fun but we essentially gave redditors a meta social network on here and they've gone too far once again

The first meeting for the Central Ohio Radio Enthusiasts is April 7. The topic is WiFi HaLow. Don't know what that is? I know how you can find out! RSVP. Join our Discord. Get on our mailing list for future events. How? It's all at core.radio. And shout out to our awesome host—@LeadingEDJE!

@ddankenst @peterevoss @davepl1968 I mean, if you're not smelting your own copper, did you really build it. 😉

@peterevoss @davepl1968 My dad wire-wrapped an entire digital clock with magnetic flip digits using only logic chips and discrete components. The whole thing was a sea of brown 30AWG wire. It was amazing. Kids today are so soft.

If you used a compiler, you did not, in fact, "personally build" it. Heck, if you used an assembler, you're benefiting from the labor of the assembler author. Get busy flipping switches!

@saltnburnem I just turn 54. You get used to it.

Today is my last day of my 30s! Shits hittin hard.

International Women's Day is a fantastic thing to celebrate. May all the women in my life—and those not in my life for that matter—have a wonderful day. Of course, this year it landed on my birthday. So you'll forgive me if I also call it International Guy Day. Just this once.

@DavidGiard Thank you sir!

Happy Birthday, @guyroyse

@housecor And ice cream!

When I travel alone, I sometimes prefer the grocery store over a restaurant. My $10 feast from the Walmart next to my hotel. High protein, healthy fats, simple ingredients.

@aditiitwt Older.

You may be old But are you this old

When you first get started with LangGraph.js, you work with message history using the MessagesAnnotation type. For simple workflows like chatbots, this works pretty well. But when things start to get more involved and your mind turns to ideas like context engineering, it begins to break down. You have to go rummaging through that message history to select just the right bits to toss into the LLM—all in an effort to save tokens and focus the model. The code quickly turns into the kind of spaghetti that will give you flashbacks to jQuery. The solution here is custom annotations. Custom annotations let you design and manage state deliberately instead of just tossing it in a pile, like the unpaid stack of bills on my desk. And that's what context engineering is really about—designing state. Want to learn how to do this? The next installment of Learning LangGraph.js is all about custom annotations. You can read it here. medium.com/the-guy-wire/l…

I find it funny that people keep saying that AI can't be thinking—that it is just a stochastic parrot that repeats what it was told (i.e., its training data). It's not that this is or isn't a valid argument. I don't know if AI is thinking or not. I don't think anybody does. And I'm not making a claim about that. But I do know that lots of humans repeat what they hear and call it thinking, too. And some of them are repeating what they've heard about AI being a stochastic parrot.

I've been exploring patterns for building apps with Redis + AI for a bit now. Some of this is by reading what others have done, but a lot is just by trying things. I think I found one for using Agent Memory Server with LangGraph.js. To that exploratory end, I've been building a workshop called "Build Your Own AI News Agent with Redis + LangGraph.js". The goal for this workshop is probably rather self-evident given the title—teach LangGraph.js and how it can work with Redis. It features: - a workflow that ingests and enriches news from RSS feeds - a search interface that finds news using structured and semantic search - a chatbot that lets you discuss and analyze the news - a workflow that generates recent news summaries And I learned something. When you're using Agent Memory Server with LangGraph.js, you really don't need a checkpointer, at least for chatbots. AMS handles the message history for you. You just need to take a user's input, run it through AMS to extract memories and build a summarized message history, and then submit it to an LLM. After you get a response, you just update AMS with the user's request and the AI's response. Then loop. While the workshop is still being developed, I just wrapped up the full solution today. So, if you want to try it out and look at the pattern, the code is out there ready to be run and ready to be examined. Check it out at github.com/guyroyse/ai-ne…

Welp. What may have been obvious to others but not me is now official. I have ADHD. Now, gimme some Adderall!

The best data comes from cherry trees.

We've all seen it happen. You're chatting with an LLM and you just want a quick answer to a question and it just spouts out all sorts of stuff. Way more detail than you need, often as a bunch of bullet lists for maximum something or other. It's annoying and you tend to just skim it at best. ai;dr This is, in many ways, the latest version of first-world problems. Oh, woe is me. My magic brain in a box is talking too much. Frankly, I'm kinda sympathetic to the LLM. I'm a rather talkative fellow myself. I've got lots to say. I get it. And when I give talks, this works out pretty well. But when I talk to my wife and kids, not so much. I'm pretty sure they tune out. tl;dr But, when an LLM does it as part of a LangGraph.js workflow, it actually breaks stuff. Let's say you have an LLM deciding how to route an email. It needs to return a single word so the workflow can programmatically decide if it needs to go to Sales, Billing, Support, or wherever. But, instead of returning just that word, it returns a little extra commentary. It explains itself. This breaks the routing and, best case, plops it somewhere for a human to review. A common solution is to fine tune the prompt or use regular expressions to match substrings for the desired word. But these are brittle for the same fundamental reason—the LLM can, and sometimes will, generate anything. The real solution is forcing it to return structured output. How? Define a schema with Zod. Bind it to the LLM. Use that LLM with LangGraph.js. Easy, right? Actually, it's not that hard. But, LinkedIn is a terrible place to share code. I wrote up all the details as part 3 in my Learning LangGraph.js series over on Medium. Go and check it out: medium.com/the-guy-wire/l…

@Kor54E There's an enshitification joke hiding in here somewhere.

@guyroyse An app where you enter your salary, and any time you opened the app, all you would see was how much money you earned in the time the app has been opened. So when you used the restroom in the office you could see how much money you made while peeing.

Share your dumbest SaaS product ideas! Code is cheap now. And so I'd bet dollars to donuts lots of you are thinking "It's time to build that SaaS product I've been dreaming about for years." Let's share some of those ideas. But... ...don't tell me your good ideas. Build those. Make some dollars and buy some donuts. But ideas rarely live in isolation, and you've probably had some real lemons as well. Those are the ones I wanna hear!