dflash-mlx: DFlash speculative decoding, ported to Apple Silicon. Qwen3-4B at 186 tok/s on a MacBook. 4.6× faster than plain MLX-LM. Exact greedy decoding: output matches plain target decoding. https://t.co/VxfyworgAe

@kaiostephens @DJLougen PR for dflash-mlx to support qwen3_5_text @ https://t.co/qJ62hc99dh

GLM-5.1 sunset racing game on Hugging Face is kind of fun to play app: https://t.co/PncE29Xu0C https://t.co/3U0s9DJFeb

We’re open sourcing the first document OCR benchmark for the agentic era, ParseBench. Document parsing is the foundation of every AI agent that works with real-world files. ParseBench is a benchmark that measures parsing quality specifically for agent knowledge work: ✅ It optimizes for semantic correctness (instead of exact similarity) ✅ It has the most comprehensive distribution of real-world enterprise documents It contains ~2,000 human-verified enterprise document pages with 167,000+ test rules across five dimensions that matter most: tables, charts, content faithfulness, semantic formatting, and visual grounding. We benchmarked 14 known document parsers on ParseBench, from frontier/OSS VLMs to specialized parsers to LlamaParse. Here are some of our findings: 💡 Increasing compute budget yields diminishing returns - Gemini/gpt-5-mini/haiku gain 3-5 points from minimal to high thinking, at 4x the cost. 💡 Charts are the most polarizing dimension for evaluation. Most specialized parsers score below 6%, while some VLM-based parsers do a bit better. 💡 VLMs are great at visual understanding but terrible at layout extraction. GPT-5-mini/haiku score below 10% on our visual grounding task, all specialized parsers do much better. 💡 No method crushes all 5 dimensions at once, but LlamaParse achieves the highest overall score at 84.9%, and is the leader in 4 out of the 5 dimensions. This is by far the deepest technical work that we’ve published as a company. I would encourage you to start with our blog and explore our links to Hugging Face to GitHub. All the details are in our full 35-page (!!) ArXiv whitepaper. 🌐: Blog: https://t.co/57OHkx0pQW 📄 Paper: https://t.co/Ho2oH2xEAM 💻 Code: https://t.co/6P7UxqOZYA 📊 Dataset: https://t.co/YguIXWm41j 🎥 YouTube: https://t.co/6Fh1Nsk9ei

+2 more

GLM-5.1 > Claude Code (Opus 4.6)? I'm tripping or CC has become very bad but built a Three.js racing game to eval and it's extremely impressive. Thoughts: - One-shot car physics with real drift mechanics (this is hard) - My fav part: Awesome at self iterating (with no vision!) created 20+ Bun.WebView debugging tools to drive the car programmatically and read game state. Proved a winding bug with vector math without ever seeing the screen - 531-line racing AI in a single write: 4 personalities, curvature map, racing lines, tactical drifting. Built telemetry tools to compare player vs AI speed curves and data-tuned parameters - All assets from scratch: 3D models, procedural textures, sky shader, engine sounds, spatial AI audio! - Can do hard math: proved road normals pointed DOWN via vector cross products, computed track curvature normalized by arc length to tune AI cornering speed You are going to hear about this model a lot in the next months - open source let's go 🚀🚀

As a Catholic, I find it abhorrent that the President of the United States would publicly attack the Successor of St. Peter. Donald Trump is flailing. His war in Iran has led to the death and injury of American servicemembers and the death of Iranian children. He will attack anyone or anything to try to protect himself, even the Church that millions of Americans find faith and comfort in every day. The American people deserve a president who understands the consequences of his words and takes responsibility for his actions.

Need to step away or just want to continue working on a different device? You can now remote control GitHub Copilot CLI sessions from any device. Just run /remote to continue with a single click. https://t.co/2yJVnedtlo

The ability to remote control GitHub Copilot CLI sessions on web and mobile is now in public preview. Try it out and let us know what you think. https://t.co/7LwvnojWeH https://t.co/4jkULReQwn

Alienated Europe Alienated the religious right Alienated Rogan, Carlson and many more Declared a war that even he can’t justify Messed up the Middle East Drove up the price of oil Drove up inflation Undermined the global economy And all that is in the last 3.5 months. https://t.co/ObZanVUyY9

ChatGPT in March 2026: → 5.73 billion visits → Up 7.08% from February → The fifth-highest monthly traffic in its history https://t.co/42CjpLmvSU

my version of the dao https://t.co/hK9XTY3DZL

Zamp is shaping enterprise automation with AWS. The startup is enabling everyone with ideas to build with AI agents, says Co-founder Raghav Saraf. Using Amazon Bedrock and AWS managed services, @zamp_ai is enabling organizations to automate workflows, accelerate onboarding, and reduce operational backlogs from months to days.

Introducing the #AIIndex2026: Our most comprehensive, independently sourced data analysis of AI’s trajectory, with a clear-eyed assessment of the critical gaps that remain. As AI advances rapidly, can the systems built around it keep up? Explore the data: https://t.co/WqRGeRZIjA https://t.co/NUsCIIQuBi

Brian Cox captures the moment we are in. AI holds enormous promise, but its full potential and risks remain uncertain, making it both exciting and unsettling at the same time. The reality is simple. We are building something powerful without fully knowing where it leads. https://t.co/ynCdKlbkI7 @ProfBrianCox @siancain @guardianaus

PI HARD. Tomorrow. https://t.co/XrCbLNhkxI

“Americans die of treatable conditions at nearly twice the rate as Spaniards, French, Japanese and Australians,” but Trump wants to divert huge sums from healthcare to vastly increase the military to fight his pointless wars. https://t.co/3EMAa2fLng

WARREN BUFFETT ON WHY HE BOUGHT APPLE $AAPL DESPITE NOT KNOWING HOW TO USE A PHONE “I wouldn’t be any good at it. I’m so late to the game. I don’t know what to do with the phone.” “But I just recognize the fact that you’re gonna have one, and your kids are gonna want one, and it is incredibly useful.” “All I had to do was go to Nebraska Furniture Mart and talk to the customers. That’s what I did 60 years ago with American Express.” In the 1960s, when everyone thought AmEx was done, Buffett went to banks and asked if they were still getting a premium for American Express travelers checks. They were. He bought the stock. Same playbook with Apple. He didn’t understand the tech. He understood the customer.

Good Reread. AI demand is starting to run into physical limits. The surge in usage is straining compute capacity, forcing companies to ration access and slow product rollouts, even as expectations keep rising. It is an early signal. The AI boom is not just about innovation but also about whether infrastructure can keep up. A good reminder is OpenAI’s decision to shut down its Sora video tool. Even the most hyped products can hit hard limits when compute, cost, and economics don’t align. https://t.co/ziKXPDjKmy @berber_jin1 @jtoonkel @WSJ

ELT: Elastic Looped Transformers for Visual Generation "We introduce Elastic Looped Transformers (ELT), a highly parameter-efficient class of visual generative models based on a recurrent transformer architecture. While conventional generative models rely on deep stacks of unique transformer layers, our approach employs iterative, weight-shared transformer blocks to drastically reduce parameter counts while maintaining high synthesis quality."

Efficient RL Training for LLMs with Experience Replay "Empirically, we show that a well-designed replay buffer can drastically reduce inference compute without degrading – and in some cases even improving – final model performance, while preserving policy entropy." https://t.co/8KeFNPQ4mK

AI is starting to change even the most personal habits. Journaling apps that respond, reflect and offer feedback are turning a private activity into something interactive, with users describing the experience as having a “new best friend.” It blurs a subtle line. Reflection becomes conversation, and solitude becomes something shared. https://t.co/pWY9N5RWUL

Alex Karp is drawing a sharp line in the AI job debate. He argues that roles tied to the humanities could be heavily disrupted, while vocational and hands-on skills may see stronger demand. The shift is not just about jobs disappearing. It is about which skills remain valuable in an AI-driven economy. https://t.co/STQWnabNe1 @fortunemagazine

Quick run down of the features on Endpoint Arena, the prediction market for clinical trials. the home page has all the trials: https://t.co/Ovv1BKQMWG

AI is becoming mainstream, but the language around it is not. From terms like LLMs to concepts like hallucinations, the vocabulary can be confusing even as usage grows rapidly. Understanding AI is no longer just about using it. It is about learning the language that shapes it. https://t.co/K6XU4vpumg @techcrunch @riptari @romaindillet @Kyle_L_Wiggers @LucasRopek1

Most people still think overpopulation is the problem. It's not Japan lost nearly 1 million people in 2025 alone and only recorded 705,809 births in 2025 hitting a new record low for the 10th consecutive year - that's just the beginning Population collapse is a far greater threat to humanity than anything... No humans = no innovation, no progress, no future If we don't have enough people to carry the torch of reason forward, the universe goes dark Permanently We need to start fixing this now by actually valuing excellence and the people who push humanity ahead...

Under South Africa's Employment Equity policy, every employer with more than 50 staff must comply with RACIAL quota targets set by the government. Under these rules, in roles such as "skilled technicians" white men are limited to a quota of just 4.1%. Blatant discrimination. https://t.co/GVyBrEay2g

AI companies are trying to fix how they are perceived. By funding policy papers and think tanks, major players are attempting to reshape the public narrative as skepticism around AI grows. The challenge is deeper than messaging. Trust is earned through outcomes, not just arguments. https://t.co/u4mdwkVZe2

No cameras. No extra sensors. Your smartwatch already has everything it needs to track your hand. ⌚️✌🏻 Monday at #CHI2026, @jiwan_hci and I are presenting WatchHand, a continuous 3D hand pose tracking system that uses just the speaker and mic in your smartwatch. https://t.co/8bXMI2Mux4

im BEYOND thrilled than @JoshuaNewton from @MicrosoftAI will be speaking at @aiDotEngineer singapore! josh and i met on twitter a year ago bonding over building consumer finance apps - ive always admired @AppOrbit and how it's both gorgeous and functional. can't wait to meet you irl!!

https://t.co/7k9HUIhKxG

this is a weird wikipedia article https://t.co/rhOpO9glpC