How do we scale evaluations for agentic #AI? 🤖 @besanushi (Senior AI Research Manager @nvidia) joins #PyTorchCon Europe to discuss "The Unbearable Lightness of (Agentic) Evaluations" and the road to consistent measurement pipelines. 📍 Paris | 7-8 April Learn more: https://t.co/4ZTGohRi22 Register: https://t.co/prWHolBpFf

+1 more

Using AI in writing just crossed a line with real consequences. A literary critic was dismissed after relying on AI in a way that led to plagiarism concerns, triggering a broader debate across publishing. The message is clear. In creative work, how you use AI matters as much as whether you use it. https://t.co/X4cmN7jOd0 @ObserverUK

Something subtle is changing in the classroom. As more students use AI, writing styles are starting to converge, making different voices sound increasingly similar. What used to reflect individual thinking now risks becoming standardized. The challenge is not just using AI, but preserving originality while doing so. https://t.co/NXPynFf5c5

More people are turning to AI for health advice, but it comes with trade-offs. Uploading medical data can deliver useful insights, yet it also raises serious privacy concerns and risks around how that data is stored and used. The lack of clear standards makes it hard to know where that information ends up. AI can be helpful, but when it comes to sensitive data, convenience can come at a cost. https://t.co/p0JAOx6unN @wsj @nicnguyen

Insane I’m running Gemma 4 on my iPhone 16 pro max Vibe coded the app in under 1h Singularity is here https://t.co/hgJvWn6VHR

Between imaging something and creating something lies resistance. In storytelling, the limiting factor is often not a lack of ideas or talent, but access to technology. There are far more people who want to say something, tell stories, and create worlds than the industry has historically enabled. That resistance has kept the number of creators artificially small. Every time you make creation simpler, faster, and more accessible, you expand the population of people who can participate. More people make more stories. More stories create more culture.

seedance 2.0 is insane for 3D animation real human faces still get flagged a lot… but animated characters works very well https://t.co/95v8jEGVGX

Andy Weir describing how he wrote The Martian and film got made is very entertaining. He said, “I wanted to make sure dorks like me would enjoy it, so made it as scientifically accurate as possible” Here the play-by-play: ▫️started writing full time in 1999 after getting fired from AOL (which had just merged with Netscape) ▫️he was a programmer and had a ton of stock options…because he was fired, had 6 months to excercise them and sold at AOL’s peak before bubble popped ▫️spent 3 years on book ideas that got “no traction” (and no agent wanted to sign him) ▫️went back to work as software engineer and wrote online as a hobby (webcomics, short stories, serials) ▫️The Martian was one of three serials he was working on 2009 (other one was about alien invading earth, another about a mermaid in 19th century New England) ▫️sent stories to a mailing list of 3,000 readers he had built up over a decade ▫️took 3 years to write The Martian, “posting a chapter at a time [every] 2 months” or so ▫️ “spent more time researching than writing, because researching was more fun than writing” (very relatable) ▫️readers corrected him along the way and when he was done, a bunch of them asked if he could make a Kindle version because reading it on his website sucked ▫️Weir didn’t want to charge because was making solid living as programmer…but Kindle has minimum of $0.99 ▫️all his fans started buying the Kindle book and leaving positive reviews ▫️this “sales spike” pushed book up charts and hit bestseller lists and just snowballed after that on Amazon ▫️within a short span, major book publisher reached out…then Matt Damon expressed interest on a film project…then Ridley Scott threw his hat in the ring ▫️Weir negotiated these deals while “debugging code” on his day job Described the process as “one of those things where everything just snuck up, bit by bit” before boooooooooom!! Film came out in 2015: made $630m on $105m budget (and Project Hail Mary might pass that now). *** Link to vid (2017): https://t.co/V9urNgxahL

I just called my Mom from my smart glasses. Introducing video calls on Mentra Live smart glasses. Join Google Meet, Teams, or Zoom call directly from your glasses. Imagine field workers like HVAC techs, electricians, or mechanics calling into an expert remotely. The expert can see what they see and have a live voice call with them, getting help exactly when they need it. You might have seen a demo like this before... so what's different here? It's all about wearability. Classical enterprise smart glasses are heavy, uncomfortable, and never make it past pilot pergatory. Mentra Live are super lightweight and shaped like regular glasses. Workers can wear them all day, and the battery lasts for a full shift. Also, since MentraOS is open-source, you can build your own apps, deploy on-prem, and stay fully compliant. Video calling is rolling out in general availability to Mentra Live users later this month.

Gemma 4 is the #1 trending model on @huggingface 🤗 https://t.co/wBIQWsjX6H

Gemma 4 is the #1 trending model on @huggingface 🤗 https://t.co/wBIQWsjX6H

Just spent 10 minutes playing the ARC-AGI-3 games and i genuinely cannot get over it. You figure out the rules yourself in like 2-3 minutes. no instructions. just vibes. GPT-5, Gemini 3 and Claude score below 1% on these. Try it yourself: https://t.co/2uuSED6EYs

@pomogranet1 @JoshuahTouyz I would probably start with 1. my Build A Large Language Model (From Scratch) book to understand the basic architecture and basic pipeline. Then maybe 2. Build A Reasoning Model (From Scratch) for inference scaling and reinforcement learning 3. Maybe one of the "production" PyTorch code bases to adapt. E.g., - Allen AI’s OLMo 3 (32B): https://t.co/stZqgmwO9N - Hugging Face’s SmolLM3 (3B): https://t.co/KkocdKVzJ4 - Intellect-3 (106B MoE): https://t.co/fZ2GbAvb5d - Nemotron 3: https://t.co/pznZ0la8f7

+2 more

Single-prompt AI workflows often hit a performance plateau. Multi-agent systems can push past it, but they usually require a massive amount of setup. Squad, an open source project built on GitHub Copilot, initializes a preconfigured AI team directly inside your repo. Learn how to run multi-agent workflows that stay inspectable, predictable, and collaborative. https://t.co/1ewya9yPpC

Gavin Newsom's State Medicaid program ‘Medi-Cal’ is losing $146 BILLION per year to fraud. The level of fraud in California is insane. https://t.co/k8A3J2wAS0

🚨“WE GOT HIM! My fellow Americans, over the past several hours, the United States Military pulled off one of the most daring Search and Rescue Operations in U.S. History, for one of our incredible Crew Office Members, who also happens to be a highly respected Colonel, and who I am thrilled to let you know is SAFE and SOUND!” - President Donald J. Trump 🇺🇸

There are no words. https://t.co/W7JRAN8JeJ

SpaceX now has over 10,000 Starlink satellites in orbit, which is honestly an absurd engineering achievement. And no, they are not just up there freelancing and hoping for the best. They stay separated because they are placed in organized orbital lanes, constantly tracked, and able to maneuver when needed. Starlink also uses automated collision-avoidance systems, which is how a constellation this large can operate without turning low Earth orbit into a scrapyard. It’s already the closest thing in the world to a true work-anywhere network and they are just getting started.

Global sales of gas cars peaked the year the Tesla Model 3 entered volume production Guess what happens next? https://t.co/ZsNfxHmUxo

What a beautiful day 🐰 Grok Imagine https://t.co/iHPmff6G4y

Another Grok Imagine upgrade. Update now. Everything’s tighter. Subtle movement lands. It just flows. Details lock in. xAI engineers showing off. @Grok https://t.co/rGF1Y91jME

El Salvador President Bukele: “They are worried about the human rights of the k*llers. What about the human rights of the women who don't want to be r*ped? Or the kids who want to safely play in the park?” https://t.co/FaB1bVkg98

Met a founding engineer today from @Replit. Jen Li. We were both judging the @Pokee_AI hackathon. They have me some credits and I built two apps in 20 minutes using the X API:a weather one which mapped storms being reported in by my climate scientist list and another monitoring my three news lists for information about the Iran war. Pokee includes X API for its customers automatically. While doing that my other AI read all your posts: https://t.co/8L5xphk0qQ He tells me about why Replit is hugely important in the AI industry. In other words how you can use it in your life and business.

Grok Imagine has gotten real good. These are Hollywood level quality of video. https://t.co/5cNezLczSG

The videos feel so incredibly real and are filled with amazing details and genuine emotion https://t.co/lNtwL8fIon

Wisprflow Steer Queue Codex https://t.co/IqRHEUg4WI

Five major supply chain attacks in two weeks, including LiteLLM and axios. We install most of these packages without thinking twice. @kstonekuan and I built YOINK, an AI agent that removes complex dependencies you only use for a handful of functions, by reimplementing only what you need, so you don't need to worry about supply chain attacks anymore. @karpathy recently called for re-evaluating the belief that "dependencies are good". OpenAI's @_lopopolo echoed this in his harness engineering article: agents reason better from reimplemented functionality they have full visibility into, over opaque third-party libraries. YOINK makes this capability accessible to anyone.

667+ InsurTech ventures. 18 countries. $10.8 billion in funding. That is what the #SilkRoad insurance corridor looks like today. I spent last week in Istanbul at the first Silk Road Insurance Forum, mapping this ecosystem, delivering a keynote on the #agenticfrontier, and leaving with one conviction reinforced: The #protectiongap is not a future risk. It is a present failure of ambition. $9 trillion. 4.5 billion people without adequate coverage. $83 trillion mortality protection gap in Asia alone. The frontier insurer does not wait for the market to mature. It architects the infrastructure that closes this gap, through parametric products, embedded distribution, mobile microinsurance, and AI-augmented underwriting for the previously uninsurable. 82% of senior leaders say 2026 is the pivotal year to rethink their operating model. Only 20–30% of digital transformations succeed. Frontier transformation is not digital transformation with better branding. It is a redesign of the operating model itself. Three phases. 1️⃣ One intelligence layer. 2️⃣ A fundamentally different human-agent ratio. 3️⃣ Everyone of us becoming agent managers. I've published my full conclusions from Istanbul in this morning's newsletter. Enjoy 👇🏽 https://t.co/OSgdEDcJeZ What does your 2026 operating model redesign look like? Read the full commentary...