📝Summarize 0.13 is out! 🎞️ Local video slides (--slides) 🤖 More model backends (GitHub Copilot) 🧠 Better GPT-5.4 support 📺 Better media handling (HLS detection.m3u8) It graduated from my tap to official homebrew formula! 🍺 brew install summarize https://t.co/b4tOuLg8gD

Just found out that flashattention use look up table+linear approximation to accelerate softmax. We have tried similar strategies on work down at MSR two years before on A100. And get over 20% end-to-end speed up of self attention. If you are interested in how to parallel look up table and linear approximation to replace non-linear function. Please take a look at our paper. Paper: https://t.co/ShiHZS6RKz #FlashAttention4 #AI #Mlsys

One good decision criticized due to a loss. One really bad decision masked by a win. We analyze the two contrasting 4th down decisions made by #DallasCowboys coach Mike McCarthy and #Rams coach Sean McVay in this week's Risky Business. Read it here - https://t.co/KPMpw5KWmA https://t.co/QERcM7416x

POETIC TIMING: Orion Integrity is setting a BRAND NEW RECORD - the furthest distance from Earth that Human Liquid Waste has been ejected from a spacecraft. 📷 @NASA

🚨📄 New preprint! We find the “boiling the frog” equivalent of AI use. In a series of RCTs, we show that after just 10 min of AI assistance people perform worse and give up more often than those who never used AI. w Grace Liu @brianchristian Mira Dumbalska and Rachit Dubey 🧵 https://t.co/Mnol47ffuy

introducing deskmate a curated map of good coworking spots in sf and south bay you can add your favorite to the map. expanding to nyc, paris, shanghai, singapore soon! https://t.co/B47toCtwjH thread with some of my go-to spots 👇 https://t.co/pklkoZZjKO

Hello OpenClaw, Hermes Agent, @Apple and whoever else. Today was the day where I finally get to throw my hat in the ring. Local agentic ai on an old Mac M1 https://t.co/0pT4hgtRbz

THE ARTEMIS II ECLIPSE. April 6, 2026. Totality, beyond Earth. From lunar orbit, the Moon eclipses the Sun, revealing a view few in human history have ever witnessed. Photo: NASA https://t.co/2cLJD3oL7p

Again, many doubted Elon Musk’s new Terafab project because of the astronomical numbers, not realizing he was playing a much bigger game all along 😂 Elon officially partnered with the US's biggest chip manufacturer, Intel, alongside massive max-capacity orders from Samsung and TSMC • Samsung is even building an exclusive, giant fab from scratch in Texas solely to fulfill Tesla's next-gen chip orders Here is why Terafab is an absolute game-changer: - 100% Vertical Integration: Consolidating everything from chip design, fabrication, memory, advanced packagings under one massive roof in Austin, Texas - Target: 1 TERAWATT per year of AI compute - that’s more than every other foundry and data center on Earth combined - With Intel helping refactor silicon fab technology and Elon running things, expect unprecedented performance optimizations - This $20+ billion mega-fab guarantees the compute needed for Tesla, SpaceX, and xAI. It will power everything from Optimus robots to space data centers and orbital satellites People have no idea what is coming Elon is not just building chips; he is building the compute infrastructure for a galactic civilization

Excited to launch our brand new photo editor. Now available on iOS, with Android coming soon. Lots of good quality of life updates and more improvements to come. https://t.co/NOyOHHDpN9

@mattshumer_ ↑ “AGI IS HERE” + COVID fear ↓ Agents shelved, hype deflated, capabilities gap exposed Salesforce stock -43% as AgentForce fails to deliver on “agentic AI” promises, investors quietly exiting. 2026 AI IPO season? FUD incoming. https://t.co/XpFGDkivwW

You can now fine-tune Gemma 4 with our free notebooks! 🔥 You just need 8GB VRAM to train Gemma 4 locally! Unsloth trains Gemma4 1.5x faster with 50% less VRAM. GitHub: https://t.co/aZWYAtakBP Guide: https://t.co/NBwKoFH2lp Gemma-4-E4B Colab: https://t.co/JjpCQgWEpL https://t.co/OZvby0tAAq

+1 more

This is insane. Gemma 4 26B running at 13GB on my Macbook M1, full context window. 20-40 tokens a second. This was a REAP model by @0xseraph further optimized through coherence physics. Dead heads were pruned and replaced by SVD rotations. Weights were quantized, and KV cache was optimized to be negligible. I am now working to get speed up higher. Wild to be talking to a local LLM which has been shrunk through the oscillator physics I have been working on for 6+ months now. #project89

How can you improve your agentic search pipeline? I just wrote a blog post with @tech_optimist from @lancedb to answer exactly that. TLDR: - Parse files and take page-level screenshots with LiteParse, the parser we just open sourced at @llama_index - Chunk and embed text, and store everything (text, image bytes, vector data) in a local LanceDB instance - Expose text and image retrieval tools to a Claude agent, and let it reason on both data types With our eval dataset, the agent got near-perfect scores on most complex QA tasks, showing how a strong parsing foundation and multimodal retrieval can really improve your search🚀 Read the full breakdown here: https://t.co/5GRRy065TI

GLM-5.1 can now be run locally!🔥 GLM-5.1 is a new open model for SOTA agentic coding & chat. We shrank the 744B model from 1.65TB to 220GB (-86%) via Dynamic 2-bit. Runs on a 256GB Mac or RAM/VRAM setups. Guide: https://t.co/LgWFkhQ5rr GGUF: https://t.co/64nkm2t8K0 https://t.co/5IKcvO6BMi

We’re thrilled to open-source TriAttention! 🚀 🦞 Deploy OpenClaw (32B LLM) on a single 24GB RTX 4090 locally 💻Full code open-source & vLLM-ready for one-click deployment ⚡️ 2.5× faster inference speed & 10.7× less KV cache memory usage TriAttention is a novel KV cache compression method built on rigorous trigonometric analysis in the Pre‑RoPE space for efficient LLM long reasoning. Github Repo: https://t.co/Gpu4E9oo3v Paper Link: https://t.co/DIWxgvlsjN Homepage: https://t.co/pDFK3mq53O

+1 more

As a former strategist at Microsoft I like this trend where AI helps you prompt better. Just met with a new AI science company and believe it or not even smart people often don’t write great prompts. Strategic thinking is tough because you are dealing with unknowns from the start. Often we don’t know what to build and with newer AI (like the one I use) often need to be trained before they are highly useful. I talked with mine for almost three months before I released https://t.co/8L5xphk0qQ I need a new business agent to help me get sponsors and hit up my connections on LinkedIn so will try this tonight.

Hermes Agent now comes packaged with Karpathy's LLM-Wiki for creating knowledgebases and research vaults with Obsidian! In just a short bit of time Hermes created a large body of research work from studying the web, code, and our papers to create this knowledge base around all of Nous' projects. Just `hermes update` and type /llm-wiki <research x> in a new message or session to begin :) https://t.co/K2FIDTTljz

Huge shoutout to our 1st place Pokee AI Hackathon winner: DataClaw 🏆! Built by Steve Kuo (@SteveTheDust), DataClaw is a project focused on something really important: making it easier for anyone to collect data for robots. That’s a big idea — because better, more accessible data collection means more people can help train, improve, and unlock real-world robotic systems.

see around corners with Marble 1.1 Plus 👀 https://t.co/MDAdUkTVWk

see around corners with Marble 1.1 Plus 👀 https://t.co/MDAdUkTVWk

The @GitHub Research folks released a "Rubber Duck" agent for the Copilot CLI. Automatically get a review from a model from a different AI family. And their data shows that yes, this does in fact help. Quite a bit. https://t.co/vHYIGwSNYj

Curious about vibe coding? Or are you already shipping apps and just want an easier way to explain your new favorite hobby to your friends, parents, grandparents, etc.? Either way, this video is for you ⏯️⤵️ https://t.co/JWKqkFzOB0

ARC Prize 2026: ARC-AGI-2 has been upgraded to L4x4s Thank you @kaggle for upgrading compute for all participants https://t.co/BbpsilbbJl

Platform Engineer - Benchmark Lead ARC Prize Foundation is hiring a senior engineer to build our benchmark platform * Expand ARC-AGI-3 * Own ARC-AGI-4 * Lay the foundations for ARC-AGI-5 Come build the benchmark that defines progress toward AGI $7.5K referral bonus https://t.co/hEm9NafkQP

🤯 big update to our flow map language models paper! we believe this is the future of non-autoregressive text generation. read about it in the blog: https://t.co/DfBXrYmJc8 full details in the paper: https://t.co/coiNXj4ucC we introduce a new class of continuous flow-based language models and distill them into their corresponding flow map for one-step text generation. we beat all discrete diffusion baselines at ~8x speed! v2 gives a complete theory of the flow map over discrete data, with three equivalent ways to learn it (semigroup, lagrangian, eulerian). it turns out you can train these with cross-entropy objectives that look very similar to standard discrete diffusion — but without the factorization error that kills discrete methods at few steps. beyond improving results across the board, we showcase properties that are unique to continuous flows. in particular, inference-time steering and guidance become straightforward. autoguidance brings generative perplexity down to 51.6 on LM1B, while discrete baselines completely collapse at the same guidance scale. we also show reward-guided generation for steering topic, sentiment, grammaticality, and safety at inference time — and it works even at 1-2 steps with our flow map model. simple, well-understood techniques from continuous flows just work incredibly well in practice for language. we’re extremely excited about the future of this class of models. stay tuned for results on scaling, reasoning, and reinforcement learning-based fine-tuning. 🚀

I built a Claude Code skill that allows it to generate a deep research report over any collection of complex docs (PDFs, Word, Pptx)….and generate word-level citations and bounding boxes directly back to the source! 📝 Check out “/research-docs”. 1. It parses out text and bounding boxes from every doc with liteparse, in seconds. 2. It then generates a full HTML report of the outputs that let you see word-level citations in each page. Raw Claude obviously has deep research capabilities, but it lacks an audit trail back to the source. This skill gives you a researched report that can be audited by others. Check it out: https://t.co/eoICBmvEnq LiteParse: https://t.co/JNER0mVcB8

+1 more

More on the @nytimes piece about that $1.8B, two-person, AI company ... not the paper's finest moment in quick retrospect. And in the healthcare context no less. Our friend @GaryMarcus was on this a few days ago as well. https://t.co/hQRSXJ7FJd

We are so back. 🥳 The Kiro Startup Credits Program is relaunching, and we’re giving early-stage teams a real head start. Early-stage Series A founders and dev teams can apply for up to 1-year's worth of Kiro Pro+ credits. Learn more here → https://t.co/19unlhdgXv Pick your tier, apply now → https://t.co/BBKzIOnYct

+1 more

Claude Mythos just obliterated every single benchmark in AI. I can't believe what I'm reading. https://t.co/roAwW0Trts

Claude Mythos just obliterated every single benchmark in AI. I can't believe what I'm reading. https://t.co/roAwW0Trts

On the pod: our most-requested guest! @ellisk_kellis from @Cornell shares the origins of his influential neurosymbolic paper "DreamCoder". Plus: program synthesis, wake-sleep library learning, world models, running an AI research lab, and more. https://t.co/MS0mW0b2y1