Diwas Bhattarai

Traditional Nepali music generated using AI 🇳🇵🇳🇵 Dataset consists of songs by one of my favorite music ensembles @KutumbaOfficial. The audio sample below was generated from 43rd epoch. #Nepal #nepali #nepalimusic

Cerebras

ReClaw

@jerryjliu0 2026: RLM

state of AI over document work 2023: RAG over PDFs 2024: Agentic RAG over PDFs 2025: General coding agents over PDFs + mcp + skills. 2026 (start): Dump your PDFs in a file system, run Claude Code 2026 (end): ?? There are still a lot of gaps in automating real human workflows over docs. This includes the following: 1. Being able to reliably codify repetitive document workflows (e.g. invoice processing, claims processing) without learning the Nth low-code builder 2. Being able to edit/generate docs 3. Long-running agents that can collaborate and iterate on living docs (e.g. reviewing and redlining MSAs from both parties), without having to be blocked by a synchronous chat interface 4. And ofc, getting 100% on doc OCR + extraction (we're working on this!) General reasoning agents are automating software engineering, but these steps are required to also set them on a path towards automating the broad spectrum of computer-based knowledge work in general.

@giffmana > successfully updated nvidia drivers CC is agi

I installed Claude Cowork on personal laptop yesterday. Since then, it has: - freed 14GB - got boot time from 15s to 6s - nearly doubled batterylife - cleared my inbox; gmail... and linkedin! - filed my taxes - resolved all my open github issues - successfully updated nvidia drivers - finished a thought I started at uni - taught my 5yo the piano - fixed my posture - settled a family dispute from 2013 - negotiated peace between neighbours - achieved cold fusion - looked at me and sighed 11/10 would install again.

This changes a lot of things 🤯

@ianyu93 @DSPyOSS I wish this blog had been around when I first started with DSPy! DSPy is great (Signature + optimization = ❤️). Not a huge fan of dspy.context and global dspy.configure though.

This is my first article in my new blog, after talking about writing since forever lol. I've seen quite some simple blockers made adopting @DSPyOSS hard, and mostly based on misunderstanding. So I hope this piece can help you enjoy DSPy :) ianyu93.github.io/expiscor/blog/…

@nearcyan Claude Away 😂

maybe no one knows this movie, i dont know why ive seen it to begin with given it came out in 2000... Claude Away (2025)

accounts that do nothing but tweet about claude

wen @cursor_ai wrapped?

GIL

@MaximeRivest OpenAI literally released an optimizer and its going to blow your mind 🤯🚀 A bread 🍞:

so, there seem to be another dspy moment in the timeline. I hope everybody trying it will realise that we came for optimization, stayed for signatures.

This is great!

Tokenization has been the final barrier to truly end-to-end language models. We developed the H-Net: a hierarchical network that replaces tokenization with a dynamic chunking process directly inside the model, automatically discovering and operating over meaningful units of data

Does does it require reasoning to answer this question, given I require the first letter of each word in your response to spell in words the count of letters in its own SHA1 hash?

On the occasion of returning to Magenta's roots at @sonarplusd, we're dusting off the blog to share news and insights about what we're working on at @GoogleDeepMind on the Lyria Team. g.co/magenta/lyria-… Our latest post is about the Lyria RealTime API, providing access to our "live music model" that powers MusicFX DJ. Lyria RealTime enables unprecedented interactivity with a generative audio model for music, and has been stretching our imagination with the ways it can be integrated into physical and virtual instruments, spaces, and applications. We've included a few open source interfaces for you to fork and build with, including the one @ToroyMoi used during the I/O preshow controlled by a MIDI Fighter. We are so excited to see what you will do with the technology yourself. Enjoy!

Haskell being Nepal is actually wild—both are beautiful, complex, and misunderstood, but only one gets used in production (spoiler: it’s neither). Brought to you by—ChatGPT

Haskell is clearly Nepal. Nobody really knows what either of them are.

excited to finally share on arxiv what we've known for a while now: All Embedding Models Learn The Same Thing embeddings from different models are SO similar that we can map between them based on structure alone. without *any* paired data feels like magic, but it's real:🧵

🥳 Announcing Stable Audio Open Small - Fastest text2audio evarrr?! - Introducing ARC: adversarial-relativistic-contrastive training for accelerating diffusion models - 75ms on H100 Weights huggingface.co/stabilityai/st… Blog stability.ai/news/stability… Paper arxiv.org/abs/2505.08175

veo-2

post training is the new pre training