Your curated collection of saved posts and media
Life Feels Fragile. Love Feels Strong. Last Month in Tokyo reminded me of one truth: Nothing matters more than the people who love you. I prayed for my dad’s strength at a shrine - quiet, emotional, grounding. Grateful for family, friends, and the kindness that keeps us going. Happy Thanksgiving. 🧡 If you feel gratitude today, tell someone in your words might be the thing that saves them.
A highlight from Web Summit is seeing countries aggressively pitching startups to set up shop in their country. Portugal, Poland, Moldova, Georgia, Qatar, Ireland all blasting music, celebrating culture, pitching tax breaks. Epcot for engineers 🤓 https://t.co/gYRTqqrxhg
Finally, some tech reporting I can get behind… https://t.co/NchWzpqpgl
If your company is consistently cutting 7-figure commission checks, that is your culture… No need to obsess over 996 or create quirky office rituals. Create the right environment and incentive structure where ambitious people make real money. https://t.co/mHlDADYcA1
Cosplaying. Y'all are crazy for this lifestyle… https://t.co/PCIgruoU9G
Dual income no kids vs. family of four https://t.co/1BFOOYkz2v
Gemini 3 is the best model at creating animations. It's not even close. Here are the prompts I used for creating these UI animations. You don't need animations everywhere, but there are key places where it can enhance your design: the hero intro, hover interactions, slow content reveal, background effects and navigation transitions. Gemini 3 relies heavily on good prompting, ideally with image or code. Intro + animation on scroll: "Animate when in view: fade in, slide in, blur in, element by element. Use 'both' instead of 'forwards'. Don't use opacity 0. Add a clip animation to the background, column by column using clip-path." Buttons: "Add a 1px border beam animation around the pill-shaped button on hover. " Text animation: "Add a vertical text clip slide down animation letter by letter" Logos looping: "Add a marquis infinite loop slow animation to the logos using alpha mask" Content switching: "Animate the big card to rotate between 3 cards in a loop. Add prev/next arrows to switch between cards." Flashlight on hover: "Add a subtle flashlight effect on hover/mouse position to both background and border of the cards." Testimonials looping: "Make the cards animate marquis in an infinite loop with alpha mask slowly."
What if one training run could produce multiple high-quality LLMs for free? 🤔Turns out it can. ❗️We’re releasing Nemotron-Elastic-12B - a single model that gives you 6B/9B/12B variants without extra training cost. ✨ Highlights: - Many-in-one model: Zero-shot slicing gives you 6B, 9B, 12B from the same checkpoint. No retraining. No extra runs. - Constant training cost: Traditional pipelines pay linearly for each size. Elastic cuts this to ~constant — 7.2× token savings for the 6B/9B/12B family. - Constant deployment memory: All variants fit in 24GB (just the 12B footprint). 2.25× reduction vs storing separate checkpoints. - Great reasoning: Hybrid Mamba-2 + Transformer architecture, competitive with same-size models on MATH-500, AIME, GPQA, LCB, etc. - Perfect for edge: Pick the right model size on-device without juggling multiple checkpoints or retraining. Elastic models = less compute, less memory, higher accuracy — and all from a single model. 📖 Read the full technical paper: https://t.co/DrxZCyvvjX 🤗 Explore the model: https://t.co/3PnZudn5PW
Advice for AI engineers 💡 A small Audio Language Model transcribes audio in real-time on your device, without internet connection or cloud costs. For example, LFM2-Audio-1.5B by @LiquidAI_ ↓ https://t.co/MoaduXlg2h
Self-hosted long-term memory for AI https://t.co/gtNpz1wwcc https://t.co/Bz9uHFj2zm
Self-hosted long-term memory for AI https://t.co/gtNpz1wwcc https://t.co/Bz9uHFj2zm
I built osgrep It’s a local code search tool that understands natural language. Works as a standalone CLI or a plugin for Claude Code. No API keys or subscription. I wanted the power of "semantic search" without the latency, price, or privacy trade-offs. Video in realtime. https://t.co/ARBvLi3KPl
Wrote a guide on how to get the most out of the context window in Amp: https://t.co/6HMjN7KbaX https://t.co/59lcykIuu4
Wrote a guide on how to get the most out of the context window in Amp: https://t.co/6HMjN7KbaX https://t.co/59lcykIuu4
The future of finance isn’t closed. It’s open source. Meet Dexter, a deep research agent built in public. It’s crushing evals, improving fast, and every line of code is yours. Finance belongs to everyone.
Wild. By far the most complete Claude Skills repo yet 🤯 @Composio’s Awesome-Claude-Skills packs 100`s of ready-to-use workflows: ↳ PDF tools, changelog generation ↳ Playwright automation ↳ AWS/CDK tools, MCP builders ... and much more! Free and open-source. Repo in 🧵↓ https://t.co/MWZmwQiqwt
@DataChaz @composio Not even close to this https://t.co/qPFE1Wik9K
@DataChaz @composio Not even close to this https://t.co/qPFE1Wik9K
Super interesting post from @karpathy. I wanted to dive deeper, so I created a @NotebookLM notebook based on this tweet, and then did a Deep Research run in-app to gather related sources. Then generated one of our new slide decks to explore further. Instant knowledge base. https://t.co/0UAGnPPdCA
Something I think people continue to have poor intuition for: The space of intelligences is large and animal intelligence (the only kind we've ever known) is only a single point, arising from a very specific kind of optimization that is fundamentally distinct from that of our tec
Meta’s SAM 3D is killing Auto CAD simply click on any segment of your engineering drawing, it generates a 3D model with correct dimensions it’s completely free, link in comments https://t.co/PIyZK94xEB
i use chatgpt to learn almost everything, but i always wish i could fork chats and go down rabbitholes on things without clogging context so i built it, powered by @tan_stack and @convex https://t.co/AeoDWtO3c2
Implemented Olmo 3 from scratch (in a standalone notebook) this weekend! If you are a coder, probably the best way to read the architecture details at a glance: https://t.co/wF8PkoDuBe https://t.co/rvXJJ5FZ3r
Olmo models are always a highlight due to them being fully transparent and their nice, detailed technical reports. I am sure I'll talk more about the interesting training-related aspects from that 100-pager in the upcoming days and weeks. In the meantime, here's the side-by-side
Releasing a new "Agentic Reviewer" for research papers. I started coding this as a weekend project, and @jyx_su made it much better. I was inspired by a student who had a paper rejected 6 times over 3 years. Their feedback loop -- waiting ~6 months for feedback each time -- was painfully slow. We wanted to see if an agentic workflow can help researchers iterate faster. When we trained the system on ICLR 2025 reviews and measured Spearman correlation (higher is better) on the test set: - Correlation between two human reviewers: 0.41 - Correlation between AI and a human reviewer: 0.42 This suggests agentic reviewing is approaching human-level performance. The agent grounds its feedback by searching arXiv, so it works best in fields like AI where research is freely published there. It’s an experimental tool, but I hope it helps you with your research. Check it out here: https://t.co/n7ctnDilJJ
As a fun Saturday vibe code project and following up on this tweet earlier, I hacked up an **llm-council** web app. It looks exactly like ChatGPT except each user query is 1) dispatched to multiple models on your council using OpenRouter, e.g. currently: "openai/gpt-5.1", "google/gemini-3-pro-preview", "anthropic/claude-sonnet-4.5", "x-ai/grok-4", Then 2) all models get to see each other's (anonymized) responses and they review and rank them, and then 3) a "Chairman LLM" gets all of that as context and produces the final response. It's interesting to see the results from multiple models side by side on the same query, and even more amusingly, to read through their evaluation and ranking of each other's responses. Quite often, the models are surprisingly willing to select another LLM's response as superior to their own, making this an interesting model evaluation strategy more generally. For example, reading book chapters together with my LLM Council today, the models consistently praise GPT 5.1 as the best and most insightful model, and consistently select Claude as the worst model, with the other models floating in between. But I'm not 100% convinced this aligns with my own qualitative assessment. For example, qualitatively I find GPT 5.1 a little too wordy and sprawled and Gemini 3 a bit more condensed and processed. Claude is too terse in this domain. That said, there's probably a whole design space of the data flow of your LLM council. The construction of LLM ensembles seems under-explored. I pushed the vibe coded app to https://t.co/EZyOqwXd2k if others would like to play. ty nano banana pro for fun header image for the repo
I’m starting to get into a habit of reading everything (blogs, articles, book chapters,…) with LLMs. Usually pass 1 is manual, then pass 2 “explain/summarize”, pass 3 Q&A. I usually end up with a better/deeper understanding than if I moved on. Growing to among top use cases. On
Excited to share a System Instructions for Gemini 3 Pro that improved performance on several agentic benchmarks by around 5%. 🚀 We collaborated with the @GoogleDeepMind post-training research team to include some best practices in our docs. 🤝 https://t.co/XxDzdQzyP7
Dia2 is OUT! a streaming text-to-speech model that can generate voice in real time 🤯 > Generates audio on the fly, no full text input needed > Real-time conversational streaming (perfect for agents + AI voice apps) > Prefix conditioning for smoother, contextual voice responses > Comes in 1B & 2B model sizes > Fully open-source with Apache 2.0 license
My notes on Google's new Antigravity IDE - at first glance it looks like yet another VS Code fork but there are some interesting new ideas in there https://t.co/KhrRzAObl6
Hey @arcprize @fchollet you should look into to DeepMind PushWorld. Some good ideas for an interactive benchmark there. It’s open source, and the puzzles can be played in the browser. Link below. https://t.co/PFaojsYYRk
oh my.. hows this even possible Meta just dropped SAM 3D, you can auto select any object in still image and.. turn them into high quality 3D model the model is free to download and test in playground link in comments https://t.co/qEmTWcz3k3
To say Nano Banana Pro is good at text is an understatement. Here's the Gemini 3 blog post, as a glossy magazine article. > Put this whole text, verbatim, into a photo of a glossy magazine article on a desk, with photos, beautiful typography design, pull quotes and brave formatting. The text: [...the unformatted article]
He and his wife used to play chess together for fun until their lives got too busy. He tried to fix it by making a chess board that he stuck to the fridge so that they could play passively while going about their day. https://t.co/QiuQhmd0ZK
BREAKING: Nvidia, $NVDA, founder Jensen Huang says “the whole world would've fallen apart” if Nvidia missed earnings expectations. https://t.co/dBhkKAVLGE
