Phillip Rhodes

Exploring #A2A protocol? Here's a repo of examples in Java and Python, including examples showing interop across languages. More to come over time. github.com/mindcrime/A2AS…

I've been using Firefox since before it was called Firefox, and I'm slowly growing to hate it. If for no other reason than its inability to do something as simple as play a Youtube video without erroring out constantly. *sigh*

Another great interview with @bengoertzel - youtu.be/b6q3ppo4D4w Ben is a fascinating thinker. Not sure I agree with him on everything, but it's definitely worth listening/reading/watching what he has to say and pondering it.

This is getting ridiculous. Folks, there is NO justification for violent acts like this. sfstandard.com/2026/04/12/sam…

Great commentary here by @bengoertzel - pointing out that the world militaries and other powerful groups are going to get access to advanced AI no matter what. So we should build Open Source AI to ensure that *everybody else* has access to advanced AI as well. youtu.be/omvXj8lLKcs

Oh hell yes. #Kreator youtu.be/Fn-nPKJBONY

Yikes. Working structure fire for North Chatham Fire, right next door to their Station 3. With Carrboro, White Cross, Pittsboro, and [??] responding mutual aid. Engine 312 with initial attack, Engine 212 with water supply from the hydrant at Station 3.

🚀Speed is the MOAT!

@StevenCheung47 You are one special kind of stupid if you believe that shit.

There has never been a President who has worked harder for the American people than President Trump. On this Easter weekend, he has been working nonstop in the White House and Oval Office. God Bless him.

I've been hacking for 15 years, and what I've seen @tryaether_ai do to companies defences in the last 60 days is something I have never seen before. I'm talking... point it at a company with over $500 million (USD) in annual cyber security budget, with WAF, SIEM and God knows what else - and it pokes a whole straight through the front door.

Sometimes all you can do is shake your head and laugh. Actually, these days that's becoming more and more "all the time" not just "sometimes".

Ha! Watching Elementary and hear this exchange: Sherlock: "These leads are terrible. We need better leads." Watson: "The Glengarry leads?" ROFL. Did not expect a Glengarry Glen Ross callback to appear in Elementary. #PutThatCoffeeDown #CoffeeIsForClosers

Tickets acquired for the #AGIConf 2026! Now to decide if I want to fly to San Fran or attend virtually. I haven't been to the West Coast in a few years, so it would be kinda fun to get out that way again. #AGI #AGIConference #AGI2026

@gvanrossum It would be better to refer to the "a prompt (or several), skills, and tools" kind of Agents as "LLM Agents", but too many people think "AI" is something that was invented in 2024 and basically have zero clue about the big picture. *shrug*

@gvanrossum That's one kind of Agent. It's certainly not the only kind. I might equally say that "An Agent is a BDI loop, an Environment, and a bundle of sensors and actuators". or "An Agent is a forward-backward chaining logical inference engine combined with a neural network", or ...etc.

I think I finally understand what an agent is. It's a prompt (or several), skills, and tools. Did I get this right?

Tonight's project: working on #APISIX integration for the #AgentRegistry, to support having the auto-registration module create Ingress routes so that external clients can talk to the #Agents. "It's not much, but it's honest work". #SwarmModeIsSickFun #Agents #AgenticAI #A2A

It's my favorite kind of work: linear algebra insight + fast kernels. When playing w Muon a while ago, we were thinking why not speed it up by operating on the small square matrix X X^T instead of the large rectangular matrix X. Jack, Noah, and Berlin spent many months understanding eigenvalues/vectors of the intermediate matrices in Muon, and finally came up with a simple and elegant algo to make this work.

OK, excellent. I have @karpathy's #AutoResearch running on my #AMD #Radeon RX 7900XTX card now. Granted this is pretty limited compared to an Instinct card or even a Radeon Pro. But you use what you've got. And credit to andyluo7 for the initial port to #ROCm.