Your curated collection of saved posts and media
New course: Build AI agents that generate images and videos -- an under-explored frontier. A key to performance is having the agent evaluate its own output, and iterate to improve quality. This short course is built together with @googlecloudtech and taught by Katie Nguyen and Wafae Bakkali. You'll learn three evaluation techniques and combine them in an agent: image-text similarity scoring to check the output matches the prompt, an LLM judge that scores against custom criteria like brand consistency, and structured rubrics that break a prompt into verifiable yes/no questions like "is the subject in the frame?" and "does the camera motion match?" Skills you'll gain: - Learn image and video prompt engineering - Build an image agent that turns brand guidelines into UI mockups - Build a video agent that plans multi-scene explainers and animates reference frames with synchronized audio Join and build agents that create images and video! https://t.co/bjuSjIxcIG
New course on serving LLMs efficiently -- how do you serve models to many concurrent users at low latency and reasonable cost? This short course is built with @RedHat and taught by @cedricclyburn. Efficient LLM serving requires efficient memory management. A 70B-parameter model takes ~140 GB just to load the weights. On top of that, every active request needs its own chunk of GPU memory, the KV cache, to store the token context it has built up so far. In this course, you'll learn to reduce a model's memory footprint with quantization and serve it using vLLM, which handles many concurrent requests efficiently through smart memory management. Skills you'll gain: - Quantize a model and measure the accuracy tradeoff - Serve a model with vLLM and watch it handle concurrent requests efficiently - Benchmark your deployment and make informed tradeoffs between speed, cost, and accuracy Join and learn to serve LLMs efficiently: https://t.co/x04xMbFlkO
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision. The longer and more complex the task, the larger Fable 5’s lead over our other models. https://t.co/DxgSu0KUxh
@SarthakPar16 Ikr. There is a small fraction of good, constructive, “high effort” takes mixed with *a lot* of dunking, piling on, jumping to conclusions, wild extrapolations and conspiracies. Always happy to see/read the former. The reality is that there are genuinely well intentioned people trying their best to navigate all of this.
@ChainZenit https://t.co/Wvn4jh59Qx
Always back to the basics: LatentMoE was probably inspired by MLA, which was inspired by LoRA, which was inspired by SVD, which was inspired by eigendecomposition. https://t.co/bWqo5iOPbP
mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community also the fact that this is un purpose not visible to the user is crazy https://t.co/n3p4niUKJ2
@IanBaer @graphtheory Try this (originally found somewhere here and was able to reproduce) https://t.co/DakkhOoifK
Just wrapping a fantastic NY Tech Week, packed with startup and VC events. My 6 takeaways on AI, startups and relationships— and a prediction for each. 1. IRL IS KING The most valuable thing this week wasn't on a screen. The more our days run through feeds and inboxes, the more a room of real people becomes something you can't fake or scale. People are increasingly craving "real" moments– and this is a big opportunity for building community AND growing business. ➡️ Prediction: the more we automate, the more we'll seek each other out. The gatherings that bring people together for real become the moments worth protecting. 2. AI AGENTS ARE TRUE TEAMMATES Agents went from demo to default. A year ago, a party trick; this week— every vertical, every workflow, someone's shipping one. The shift I'm watching: from agents you operate to agents you delegate to. And as they take on long-horizon work, the bottleneck stops being capability. It's joint teamwork. ➡️ Prediction: winners won't have the smartest agent — they'll have figured out how to work with it as a trusted teammate. 3. PARTNERING UP Partnering up is 1+1=100. We hosted a Google DeepMind + Google for Startups + Google Cloud day for founders. 3,000 applied. By pooling across teams, we brought the best of Google into one room for the best startups. Alone, none of us could have pulled that off. ➡️ Prediction: the organizations that show up as one — rather than shipping their org chart to the world — will win. (One Google FTW!) 4. AI FOR REAL USE CASES The best founders I met aren't chasing the tech — they're bridging real business needs with the edge of what's now possible. Companies like InstaLILY AI, Simular and Rehearsals start with the problem, then pull emerging AI forward to close the gap. Not technology in search of a use case. The use case, pulling technology toward it. ➡️ Prediction: the breakout companies won't be the ones with the flashiest tech — they'll be the ones pointed at problems worth solving. 5. STORYTELLING STILL WINS When everyone's slides look great, taste becomes the last moat. Gamma hosted one of the best events of the week — Malcolm Gladwell, Mike Birbiglia, Bobbi Brown — pairing gorgeous visuals with live spoken-word storytelling. When great production gets effortless, the story is the diffentiator. ➡️ Prediction: as content becomes infinitely remixable, the premium on crafted story and delivery goes up. 6. RELATIONSHIPS WIN New connections expand you; old ones ground you. The magic is doing both. Tech Week throws you into rooms full of new people — and that energy is real. But some of the best moments I had were the stolen coffees and late dinners with founders, partners & friends I've loved for years. ➡️ Prediction: those who continue to expand in both breadth and depth of relationships will be the ones living the good life. Thanks to our partners at a16z. And… wild to see the Knicks fan party at MSG!! 🏀
what if you could just talk to a 1987 nes cartridge i just turned the @cursor_ai sdk into a nes rom hacker you describe a mod in plain english → agent searches the binary, proposes a byte patch, emulator hot-reloads in <1s "infinite lives" "give me the laser" "keep weapons after death" all just work s/o @poteto pstack is goated cc @ericzakariasson @leerob
Introducing Gemini 3.5 Flash Live Translate, our real time speech to speech translation model which supports more than 70 languages (both in and out), and is so natural. It is available in the Gemini API, AI Studio, & Google Translate right now + coming soon to Google Meet!! https://t.co/nDiaqrbHgF
Google Translate has come a long way since 2006. Now, each month our translation tools support: 🗣️ Nearly 250 languages 🌍 1 billion+ users asking Google for translation help 💬 1 trillion+ words translated across our ecosystem https://t.co/G1nYjmWdKF
For over 20 years, we've dedicated ourselves to removing language barriers so people can learn, speak and connect more deeply than ever before. Today, we’re taking our next step with the release of Gemini 3.5 Live Translate — our latest audio model for live, speech-to-speech translation across 70+ languages. 🧵
Gemini 3.5 Live Translate is a truly magical experience, try it out for yourselves today on https://t.co/XD7w6QNdQZ at @GoogleAIStudio
For over 20 years, we've dedicated ourselves to removing language barriers so people can learn, speak and connect more deeply than ever before. Today, we’re taking our next step with the release of Gemini 3.5 Live Translate — our latest audio model for live, speech-to-speech tra
Gemini 3.5 Live Translate! We just shipped a real-time babel fish. - 70+ languages, 2,000+ language pairs. - Natural translated speech, works in noisy environments. - Stays in sync with the speaker, no lag, no awkward pauses. - Auto-detects the language being spoken. Available today in Google Translate (Android & iOS), the Gemini API (Public Preview), and Google Meet (Private Preview). I genuinely think this is the beginning of the end of language barriers. Anyone can now speak and understand anyone.
Finally reading the swe-bench paper and it's kinda cool to see GPT-3.5 again haha. That was my first foray into language models and the first time I felt the sparks of AGI :) https://t.co/Jll4KlFNiO
Introducing DiffusionGemma, our first exploration with open diffusion text generation models 🔥Generate blocks of text at a time 🤏26B MoE built on top of Gemma 4 ⚡️Up to 4x faster in popular consumer GPUs 🤗Apache 2.0 Excited to see what the community builds with it! https://t.co/zros8uvBsi
i hooked my whoop to my work calendar to find which coworker gives me the most stress 🚨 thanks to fable, I reverse engineered whoop to pull per minute heart rate. nd matched spikes with cal events and attendees I now have a leaderboard and I think about it daily. few info masked for obvious reasons ;)
Gemini Omni Flash is SOTA at image to video, text to video, and video editing : ) Excited to get this to developers in the API soon! https://t.co/u0fzmJwBb4
To test an AI agent, you run it in a sandbox: a throwaway cloud machine where it does the real work. Harbor, the framework most top labs use to run those tests, just added CoreWeave Sandboxes as an option. Switching takes one line of code. https://t.co/Oyv2ibsmUE
Learn more about the Harbor Framework: https://t.co/nqKKbF4OBX
at the @wandb weekend hacakthon!! building agi https://t.co/mRnL1S9Nnw
at the @wandb weekend hacakthon!! building agi https://t.co/mRnL1S9Nnw
Biggest turnout we've had at @wandb hackathon ever (Sorry for folks who couldn't get in) The energy is insane! Looking forward to incredible hacks over the weekend, hackers are pumped! Thank you sponsors @Redisinc @OpenAIDevs @cursor_ai and @CopilotKit https://t.co/mwJTZP1FnC
🙌 Stellar vibes at the @wandb @cursor @cerebral_valley hackathon this afternoon!! https://t.co/2qlthzJWls
🙌 Stellar vibes at the @wandb @cursor @cerebral_valley hackathon this afternoon!! https://t.co/2qlthzJWls
Peter and Valeria (who runs a daycare!) told me they used over 2BN tokens (most of it cached) over the two @wandb hackathon days, with a bunch of subagents. His existing swarm created a mini swarm! Video of their demo soon! https://t.co/nviifMJOKj
🤯 Highly impressed by the caliber of projects, especially the winners! 1️⃣ A bug bounty multi-agent system that found an exposed @googlemaps key on the @priceline website 2️⃣ A multi-agent system for observing training runs and reacting to loss spikes, including making PRs Mus
"There are no unsolved problems left in physics." Devoid of things to solve, @SGRodriques moved to the one field nobody has figured out yet: biology. The reason we haven't cracked it yet? Not enough talent. You simply can't scale brilliant scientists. So he built AI scientists to solve aging, cure every disease, and finally understand the human brain. New Gradient Dissent episode linked below.
YouTube: https://t.co/piERHDWipy Apple Podcasts: https://t.co/N46SAjglUa Spotify: https://t.co/FaVumeLylI
Benchmarking a new approach against a baseline that lives in another project? W&B Workspaces now let you pin runs from any project onto the same line plots as your local runs. No moving, no copying and your originals stay put. Live on SaaS today. 🫡 https://t.co/v2JuvEULMd
A single eval score isn't an approval record. Here's how to defend an AI release with evidence instead of screenshots. The open-source AI Governance Toolkit on @wb_weave, start to finish: 0:00 The problem: scattered evidence and review gates 1:08 W&B Weave and the open-source AI Governance Toolkit 2:25 Intake: submitting an application profile 3:07 Scope: deriving the review plan and risk tier 4:08 Assess: the failure verdict and red-team results 5:48 Probe: manual testing and pinning findings 6:56 Decide: human judgment and the final decision 8:07 Slack automation and wrap-up
Two ways to learn more. The ebook with the full framework: https://t.co/xUfCNJqnbj OR Fork the toolkit: https://t.co/1RwdNwgJ8v
