@akoratana A small caveat: agents rely on models that don’t reason or deliberate. Their outputs come from sampling probability distributions learned during training and conditioned on context during inference. The mechanism is statistical density, not logical inference or causation. Traces don’t create new signal, the probability landscape is already fixed. Reinforcement learning simply sharpens the same distribution. Without new external signal, the system is just iterating over its own statistics.

I'm pleased to announce that a new 2026 edition of my New York Times bestselling book, Rise of the Robots: Technology and the Threat of a Jobless Future, will be available on June 2. I have extensively updated the book to cover the latest advances in generative #AI and robotics and to examine the future economic and job market implications of the unfolding AI disruption. The book focuses on what we can do as individuals, and as a society, to successfully navigate the looming transition into the age of AI. You can pre-order from the link in the reply. @BasicBooks #RiseoftheRobots

@GibsonHimself Depends what we see as net positive. If it's shorter wars with less casualitites, I'll take it!

@mattshumer_ That sounds like an ad...

@nikillinit speedrun or yc?

@ASychov @Rivian The auto industry has known for years that vehicles with straight lines are highly polarizing, but highly masculine, and that vehicles with curves and circles are seen as nice, but feminine. Translation: we'll argue about the design endlessly. :-)

@Midnight_Captl Yeah, NVIDIA has me busy that morning, but will try to stop by. You broadcasting this from GTC?

What's the SotA for AI music generation these days? Any AI generated bangers you've listened to lately?

@barkmeta AI isn’t electricity. It’s software. Tokens aren’t a natural resource, they’re a billing unit. Calling intelligence a “utility” is marketing meant to justify the burn rate while the underlying economics fall apart.

@BoringBiz_ AI isn’t electricity. It’s software. Tokens aren’t a natural resource, they’re a billing unit. Calling intelligence a “utility” is marketing meant to justify the burn rate while the underlying economics fall apart.

I get dozens of pitches each day from ghostwriters (thoughtful, well-written) but my reply will always be don’t delegate the things that bring you the most joy.

Ukraine didn't just release drone footage. It built the world's first open training ground for military AI. Real war is now a source of data. Every autonomous system hits the same wall: it needs reality, not simulations. Self-driving cars solved this with billions of miles on actual roads. Military AI never had that option. No country had millions of labeled combat video frames until now. Thousands of drone missions worth of data are open to partners. Partners get irreplaceable training data. Ukraine gets back improved AI for its frontline. Claude was the first major AI model deployed on the Pentagon's classified networks through Palantir. It's now embedded in the Maven Smart System, generating real-time targeting for military operations in Iran. OpenAI, Google, and xAI received parallel $200 million Pentagon contracts under identical terms. The frontier labs aren't selling chatbots to soldiers. They're becoming infrastructure for how wars get fought. Ukraine's dataset changes what all of them can build: - Drones that identify targets in near-zero visibility without human input - Navigation trained on real terrain destruction, not clean satellite maps - Autonomous decisions tested against thousands of actual combat scenarios Two forces are converging. Silicon Valley provides the brains. War provides the only classroom where those brains can learn. Ukraine is training small, efficient AI models designed to run on cheap chips inside drones Center for Strategic and International Studies, not massive cloud servers. The arms race isn't about who builds the best model. It's about who controls the best data and who's willing to plug AI directly into the loop where decisions turn lethal.

The top news of the morning was @PalantirTech's announcement of an AI operating system. People overindex on Palantir's military/government surveillance business and forget that many of the largest enterprises use it to run their businesses. An "operating system" that runs businesses will be welcomed into those that already use it, and will gather new customers who are looking for ways to use AI to automate theirs. In Silicon Valley it has long been known that enterprise is boring, but where the money is. @arrington who started @TechCrunch once told me he refused to cover enterprises in the early days because they are so boring, and so few people care about them. But that is where the money is and everyone is racing to get enterprises to use their AIs for that reason. Will they buy Palantir? From what I'm hearing from enterprise folks: yes. That's why this was the big news of the morning. The market understood this which is why PLTR is one of the few green stocks on my list today.

@JoesInvestments @vulnerablecodes Maybe a container issue. I know Nvidia recommends their docker containers, but I am usually just running it on the Spark directly.

ngl i feel so cool working out of a terminal and pretending to be an engineer with claude code😎

@karrisaarinen looks layoffs impacted ex-Microsoft, Seattle based big tech engineers the most very different skills than where the world is going

Ever wondered what we mean by 'agentic' OCR? It's parsing that reasons about documents instead of just reading them. Agentic OCR adapts to layout changes by treating document processing as a goal-oriented task rather than simple text extraction. 🧠 Uses multimodal language models to understand document structure and context, not just convert pixels to text 📍 Provides visual grounding with bounding boxes so every extracted field traces back to its source location 🔄 Runs self-correction loops to catch inconsistencies before they reach your downstream systems ⚡ Achieves 90-95%+ straight-through processing rates on new document formats without template setup This matters for legal teams processing M&A due diligence, healthcare admins handling medical forms, and finance teams reconciling reports across subsidiaries. The agent doesn't just extract data - it completes document workflows with built-in validation and business logic. LlamaParse is our implementation of agentic OCR. Get 10,000 free credits to test it against your actual documents: Read the full breakdown: https://t.co/FRoyXKGUia

The Gemini Embedding 2 baseline here is.. 2 days old. Was just being celebrated and is now outperformed by a median of 14% and up to 91 points. If I didn't kind of know how powerful scaling ColBERTs and ColPalis can be compared to a single-vector model, I'd be in disbelief! https://t.co/Cz4lIpJuta

The Gemini Embedding 2 baseline here is.. 2 days old. Was just being celebrated and is now outperformed by a median of 14% and up to 91 points. If I didn't kind of know how powerful scaling ColBERTs and ColPalis can be compared to a single-vector model, I'd be in disbelief! https://t.co/Cz4lIpJuta

@JoesInvestments @vulnerablecodes Weird, what issue are you having. Works for me both in ollama and native llama.cpp I am using unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:MXFP4_MOE

@JamiluBasi35329 Sad

On the first day of the attacks on Iran, a school was bombed, killing more than 175. Nobody knows for sure what happened, and nobody wants to take responsibility. @GaryMarcus thinks there's a "good chance" that the incident stems from the use of AI. https://t.co/TZm3jvKiZd

@BEBischof Nothing is random “graphDBs are all you need” fresh rage bait on the timeline today

humans dont need formal graphs to traverse information space - "links" are enough and the same is true for agents