Adam is similar to many algorithms, but cannot be effectively replaced by any simpler variant in LMs. The community is starting to get the recipe right, but what is the secret sauce? @gowerrobert and I found that it has to do with the beta parameters and variational inference.… https://t.co/CuPj9owJjS

Obviously naive RAG sucks, but it's pretty clear that agents still need the right retrieval tools in order to access massive amount of enterprise context. 🤖🔎 This is a fantastic blog post by @thesourabhd showing how we've carefully crafted a universal retrieval interface… https://t.co/7I8hDiscxR

Within the past week there were like 4 papers doing KV cache for diffusion language models Here's another from NVIDIA now: https://t.co/IRARcjXy0j

Can I use the same model for both the main task and evaluation? https://t.co/IkGmOZQSjm https://t.co/xfaFmaG52C

on production monitoring and real time evals https://t.co/lFDR16kniy

today's office hours w/ https://t.co/jdjFasak7B https://t.co/9th8fENHsP

AI as Hedge Fund Analyst https://t.co/RVVWkUxCDp

Excited to share my first contribution here at Illumina! We developed PromoterAI, a deep neural network that accurately identifies non-coding promoter variants that disrupt gene expression.🧵 (1/) https://t.co/U71ENH5F1t

An Operating System for Memory-Augmented Generation in LLMs Lots of great ideas on how to think about memory and better manage it in LLM-based agents. Must read! Here are my notes: https://t.co/zhFpaCvQhJ

Wow. Sequoia just backed a startup going after Devin, Cursor, and OpenAI Codex. Their agents can read your entire codebase, write, test, fix, and merge PRs automatically It’s like having a fully autonomous software engineer by your side, 24/7. Examples 👇 https://t.co/qb69mPOLSR

lmfao beautiful @AmpCode one shotted this and installed into my shell. now I just run autocommit to generate conventional commit messages. <3 sonnet https://t.co/WfjlQLYfzR https://t.co/NH9VFMkxwS

What began as a side project in late 2024 has grown into something pretty useful! With the latest case studies from @LangChainAI's Interrupt conference, we've now curated over 700 real-world LLMOps implementations. Fresh Production Patterns from the Field: 🏭 Factory: 5,000+… https://t.co/LGy2ixniwv

Edit your notebooks using the jupymod mcp Built and deployed the mcp with @AmpCode. Download it from pypi with uv add jupymod https://t.co/6S4tCxZBDc

Talk to https://t.co/CpQTspHXbi 🔊, the most modular voice AI around. Empower any text LLM with voice, instantly, by wrapping it with our new speech-to-text and text-to-speech. Any personality, any voice. Interruptible, smart turn-taking. We’ll open-source everything within the… https://t.co/ZatL9tS1OG

LlamaIndex now supports the new OpenAI Responses API features: · Call any remote MCP server · Use code interpreters by using it as one of the built-in-tools · AND generate images with streaming. https://t.co/AyBOUodXK3

DeepSeek’s R1 leaps over xAI, Meta and Anthropic to be tied as the world’s #2 AI Lab and the undisputed open-weights leader DeepSeek R1 0528 has jumped from 60 to 68 in the Artificial Analysis Intelligence Index, our index of 7 leading evaluations that we run independently… https://t.co/EmilinKZLe

Building a FAQ from our evals course. This one has been coming up lately. https://t.co/N984YUjgjS https://t.co/Li2dMvkzI9

yesterday, I finally managed to train a good enough keypoint detection model that identifies 33 characteristic points on a @NBA court I’ll use this model together with homography to map player positions from the video frame onto the court map sports repo:… https://t.co/bHVAqBXncb

Sonnet 4.0 is cracked at document understanding. With our latest update, we’ve built a sonnet 4.0-powered agent to help convert the most complex docs into markdown, detect layouts and tables/images. The agent loop helps prevent hallucinations and join tables across multiple… https://t.co/10KCvKOCly

+1 more

You can run full browser automations for AI agents without worrying about Chrome, Puppeteer, or infrastructure. Steel is an open-source browser API that wraps Chrome, manages sessions, handles proxies, and exposes everything via a REST API or SDKs. https://t.co/9XKUl7zPxg

This is a great example. I am telling you, listen to me—it's the needle that matters, not the arc, for fast cognition in gauges. This is not disputable. So much noise was made in the comments in that quoted thread. I know it is correct. https://t.co/DyDLqMZRaN

SOTA AI agent that reliably works... where Claude, Gemini, and o3 fail... to do the boring chores in life... @FeatherlessAI is making this possible, as part of our work into AI reliability Surpassing existing frontier models & agents by 50%+ https://t.co/FwTAjBJ72q

Iykyk. Programming with claude: Yes, you’re right. ah good catch! @HamelHusain @johnowhitaker @hugobowne @JnBrymn @awakenjake https://t.co/8M25BexhKO

Lol @AmpCode going to be eating my credit card once I finish my free credits at this rate lmfao https://t.co/cuiqU1RYz0

I know what I’m doing as soon as I start coding on Monday! 🥳 *please dear Universe may it work and allow small, incremental changes 🙏🏻* I wonder if I can at least try Claude Code if I’m on the pro plan? 🤔 https://t.co/4dEpm2xvUx

Using Veo 3 to create fictional product reviews (unsurprisingly, it does YouTube review style very well, sound on.) https://t.co/0N4TO8W2Rp

Welp got bored with that gonna try gemini pro and see if I feel like its worth it now - I literally used 4.5 for google search qa and thats all Ive really used any pro feature for https://t.co/lopbhofZTZ

You can now crawl entire websites and extract LLM-ready data with a single tool. Crawl4AI is an open-source repo built for AI agents, RAG, and data pipelines. It supports both browser-based and HTTP crawling, with real-time Markdown generation from any site. https://t.co/9Zwnk5AKAi

Introducing Rork 1.0 Make any mobile app you want, save it to your phone or share it with the world – in minutes. Powered by Claude 4. Video by Veo 3 AI. https://t.co/xr3uibte7g

Learn to Reason via Mixture-of-Thought Interesting paper to improve LLM reasoning utilizing multiple reasoning modalities: - code - natural language - symbolic (truth-table) representations Cool idea and nice results. My notes below: https://t.co/U4pCcy7JUu

Create stunning AI-generated images with precision using the Image Generation Agent, by @itsclelia ! 🎨🤖 This open-source project automates the prompt refinement-generation-visual feedback loop, helping you produce images that truly match your vision. ➡️ Multimodal agent that… https://t.co/xbI550NOnc

Microsoft releases NLWeb NLWeb uses MCP to make it simple to interact with websites in a standardized way. Devs can now convert any website into an AI app. MCP is to NLWeb what HTTP is to HTML. This went largely unnoticed this week, but it looks like a big deal. https://t.co/98M5UmGYP9