meet Shadow, a powerful open-source background coding agent! fully featured with a remote environment, codebase indexing/wikis and subagents to understand, write and test your code, directly making PRs to github built in a few weeks w/ @ishaandey_, @ElijahKurien

I’ve been using papiers for the past few weeks, it’s super great! im a huge fan of @cognition deepwiki and I hope this becomes the standard for papers

THIS IS THE MOST UNDERVALUED COMPANY ON NASDAQ, according to Warren Buffett’s analysis. Buffett’s rule is simple: Buy real companies that earn real money at irrationally cheap prices. That description fits NextNRG Inc ($NXXT) perfectly. 📊 Facts: * Market Cap ≈ $200–250

Introducing Nested Learning: A new ML paradigm for continual learning that views models as nested optimization problems to enhance long context processing. Our proof-of-concept model, Hope, shows improved performance in language modeling. Learn more: https://t.co/fpdDlYaleL

the hidden complexity of video search: why multimodal retrieval isn't just "long text" i just watched rajan agarwal from kino ai break down how they're solving one of ai's trickiest problems: searching through video content effectively. most retrieval work focuses on text and

After two years of work, we’ve made an AI Scientist that runs for days and makes genuine discoveries. Working with external collaborators, we report seven externally validated discoveries across multiple fields. It is available right now for anyone to use. 1/5

paper & code, with @bxptr_ github: https://t.co/iuGY7SnbgQ arxiv:

two key problems we noticed were tokenizer fragmentation and cross lingual coupling LLINK is an attempt to make a highly compute efficient alignment between a strong multilingual encoder and a LLaMA base! we notice improvement in both content understanding and Q&A

Three months of confusion. One revelation: God was listening the whole time. Get the rest of the story on my page.

recently, i spent some time working on cross-lingual alignment for LLMs via encoder injection! treating languages as modalities is a compute-efficient way to extend understanding of low-resource languages without extending pre-training or the tokenizer

@windsurf @paulg Full breakdown: https://t.co/1InFCutHYO We all need more AI that turns your brain ON, not OFF. Software development only becomes engineering with *understanding*. Your ability to reason through your most challenging coding tasks is constrained by your mental model of how

Wow, language models can talk without words. A new framework, Cache-to-Cache (C2C), lets multiple LLMs communicate directly through their KV-caches instead of text, transferring deep semantics without token-by-token generation. It fuses cache representations via a neural

there is no reason that anything a llm reads or produces that doesn’t get shown on an interface has to be human readable save tokens and let llms learn their own compression

Your body speaks truth. Your brain tells stories. TWO SYSTEMS running your life — end the war. Watch. Touch a leaf. Heal.

I built Tacc — a Tokenization-Aware Compression Codec that efficiently sends LLM outputs and tool calls over low-bandwidth networks. It compresses faster and smaller than gzip, making it a better choice for serving LLM responses over HTTP. Here’s how it works 🧵 (1/7)

i strongly believe tab is the ideal interface for coding, writing and even maybe browser use learning and accelerating your patterns is how u become faster. i find myself rewriting so much code with agent tbh

potentially naive question: why dont we train RL on summarized states so we can have significantly longer convos/not violate MDP? also if the summarizer drops info that later matters, we can penalize it if we want llms to "think for 1-2 months" then we have a context problem

It was a magical ride with all of you. Our city, our home, our fans showed up on the grand stage...and we couldn't be prouder.

i've only just started going through this but i can already say with confidence that this is an incredible resource for people of all skill levels to learn about LMs

hell yeah

Kimi Linear Tech Report is dropped! 🚀 https://t.co/LwNB2sQnzM Kimi Linear: A novel architecture that outperforms full attention with faster speeds and better performance—ready to serve as a drop-in replacement for full attention, featuring our open-sourced KDA kernels! Kimi

cursor has won coding their data flywheel has turned into the most amazing coding model in the world. Composer-1 truly feels like another Claude 3.5 Sonnet sized jump in capability

Generalists are useful, but it’s not enough to be smart. Advances come from specialists, whether human or machine. To have an edge, agents need specific expertise, within specific companies, built on models trained on specific data. We call this Specific Intelligence. It's

i love the abstraction of nearly the entire swe dev workflow inside of Cursor: the browser, both background & active agents, and code review! super cool launch