Moh

@ClementDelangue pi mono repo

Feels like one of the cybersecurity risks over the coming months will be widely used open-source projects that are simply too lightly maintained for how critical they’ve become. A few ways to help: - fund open source more, and reward maintainers better - bring some of the most critical projects under more resourced umbrellas like @linuxfoundation or @huggingface Which projects would actually make sense to join HF if we wanted to help them on cybersecurity?

@vincentweisser @eliebakouch My pfp would go hard on there

Blessed to work with the absolute dream team, and to have our favorite people in AI join us - welcome @eliebakouch 🙌

@zeddotdev please better support mermaid, its the single most important part of the markdown, with zoom support and dynamic nodes movements.

Tell us how we could make Zed the single best tool for working with markdown

@charles_irl @modal @vllm_project oh makes sense, glad that you are sticking to that then!

@dude474 @modal @vllm_project we're a code-first platform!

New in the @modal docs: deploy an H200 server for Gemma 4 26B A4B with @vllm_project. Not optimized for performance, but still gets ~175 tok/s at concurrency 1. modal.com/docs/examples/…

@Padday @Fin_ai great to hear!

@dude474 @Fin_ai Btw we do have regional hosting for Fin already. fin.ai/help/en/articl…

The evolution of @Fin_ai is a great example of product strategy in this era. First, an Agent built on a deep AI layer. Second, we trained our own model, Apex, that outperformed everything else. Now third, we're opening up access to Apex (and our other models) to everyone, including our competitors! If you want the best, now you can have it.

@willccbb damn, ur hair got taller

we’re hiring! primeintellect.ai/careers

@mkurman88 its cuz they are collecting data to further align the model one last itteration

It’s actually sad that it’s free on OpenRouter and not available in the $50/mo paid plan lol

@joshelgar @eoghan just like gpt 5 on azure foundry

@dude474 @eoghan What do you mean?

@Padday @Fin_ai Basically we are building CX agents for a local market, and Fin Apex and the rest of the stack is looking to be a great option, but one of our requirements is to be deployed within our region, just like deploying GPT 5 on Azure foundry.

@dude474 @Fin_ai We're always taking feature requests! Tell me more about what you want to do.

@oneill_c and they be calling it BALLS

Just one more RL training library bro. I promise bro just one more library and we'll fix async and decoupled training/inference and off-policiness bro. Please bro just one more

thats why claude code is fucked in @terminalbench

big day for the gpu poor startups

@kais_rad haha, I get it, its april ....2nd .... shit

NOOOOOOOOOOOOO

@HaoZhang3438830 great job! but a small question, hows that different than tinker/ skyRL-tx? other than the sandbox part

Excited to introduce ProRL Agent: Rollout-as-a-Service for RL training of multi-turn LLM agents! 🚀 As we move toward complex agentic tasks, rollout infrastructure is often a bottleneck. We’re decoupling I/O-heavy rollouts from GPU training via a unified HTTP API. Why ProRL Agent? Decoupled & Scalable: Treats rollout as a service, allowing near-linear throughput scaling. System-Level Optimization: Includes load balancing and automated sandbox cleanup for high stability. Integrated: Now part of NVIDIA NeMo Gym to help researchers scale RL pipelines faster. The Results 📈 On SWE-bench-Verified, we saw significant gains: +8.4 on Qwen3-8B +8.2 on Qwen3-14B Proven success across STEM, Math, and General Coding agents. Check out the research and open-source code: 📄 Paper: arxiv.org/pdf/2603.18815💻 Repo: github.com/NVIDIA-NeMo/Pr… Huge thanks to the team and NVIDIA for the support! 👏

@ClementDelangue @Pinterest @Airbnb @NotionHQ @cursor_ai @eoghan @intercom Every company has a system and every system is an environment for an agent to interact with which equals to RL the llm to better interact with the system.

After @Pinterest @Airbnb @NotionHQ @cursor_ai, today it’s @eoghan @intercom publicly sharing that they’re finding it better, cheaper, faster to use and train open models themselves rather than use APIs for many tasks. And hundreds of other companies are doing the same without sharing. Ultimately, I believe the majority of AI workflows will be in-house based on open-source (vs API). It took much more time than we anticipated but it’s happening now!

@abidlabs trackio cli when?

When we first released Trackio, we put together a barebones UI just to get it out. We always knew we'd come back and redesign the UI. Now that Trackio is downloaded 1M times every wk, we figured it was time to release a beautiful UI. Out now in Trackio 0.20, try it out!

@eoghan oh great!

@dude474 It scored better in scaled real world deployment. 8k customers and 2M resolutions/week.

By far the biggest technology update in the history of service agents. And a significant development for AI applications and AI at large. Today we're sharing news about Apex, the world's first customer service model that beats all known models at this task.