Shang Hong Sim

ML @ Oumi

[Shang Hong Sim @shanghong_sim]

@natolambert yes!!

[Shang Hong Sim @shanghong_sim]

@danqi_chen @__howardchen @ZexuanZhong @danfriedman0 @xiamengzhou @gaotianyu1350 @_awettig congrats @__howardchen !!!

[Shang Hong Sim @shanghong_sim]

@JIACHENLIU8 in the age of agentic coding and execution, what’s the best way to build taste?

[Shang Hong Sim @shanghong_sim]

i’m curious as to what “stop believing in yourself” looks and feels like.

[DAN KOE @thedankoe]

At some point, usually in your 20s, you'll notice that the people around you stop believing in themselves. And no matter how hard you try, you can't save them. By all means, do not let it infect your mind. Stay on your path.

[Gary Marcus @GaryMarcus]

Imagine if your car randomly went out of control 10% of the time.

That’s commercial-grade generative AI web search.

[Shang Hong Sim @shanghong_sim]

@Oumi_PBC This also unlocks more frontier methods.

A big part of my work was on on-policy distillation.

Instead of training on static datasets, the model learns from data sampled from its own policy in real time.

[Shang Hong Sim @shanghong_sim]

@Oumi_PBC Oumi connects the full loop: evaluate → generate data → train → improve

All in one system.

So instead of fragmented steps, you get fast, reproducible iteration.

[Manos Koukoumidis @Koukoumidis]

🚨 The era of general-purpose AI is over.

Today we're launching Oumi. 🚀

The platform that lets any team build custom AI models — in hours, not months.

Just describe what you need. Oumi builds it. #VibeML

Higher quality. Lower cost. Fully yours.

[Shang Hong Sim @shanghong_sim]

@SarahChieng @Oumi_PBC would love to cohost the next Cafe compute! We attended the previous one held in Seattle and had a blast XD we could support Seattle, SF and NYC, and open to more!

[Shang Hong Sim @shanghong_sim]

@jennyzhangzt So awesome Jenny!

[Aniruddhan Ramesh @aniruddhr04]

Just had a PR merged to @oumi_ai!
Multi-turn data synthesis support is now in the codebase. This enables programmatic generation of full conversational dialogues. Here's what's coming and why researchers will care 🧵
https://t.co/Xau1LczLSA

[Manos Koukoumidis @Koukoumidis]

Want a prediction for the Super Bowl or your future based on your AI strategy? Ask ChatGPT.

🏈 🏈 🏈
❓ Respond in a single sentence. Who will win the Superbowl 2026?
❗ The Seattle Seahawks are widely favored by odds and expert predictions to win Super Bowl LX (2026) over the New England Patriots.

🤖 🤖 🤖
❓ Respond in a single sentence. Which enterprise will win, the one only prompting closed off the shelf models or the one only fine-tuning open models?
❗ The enterprise that only fine-tunes open models will win, because ownership of the model stack creates compounding advantages in cost, performance, and strategic independence that pure prompting can’t match.

You heard ChatGPT...
(Feb 8th 2026, 3:21pm)

The one thing that ChatGPT doesn't know "yet" is how easy it is to finetune models using Oumi. What are you waiting for? 🚀

[Shang Hong Sim @shanghong_sim]

@nw3 @Zoe_ZouYi Real, it’s multipurpose

[Manos Koukoumidis @Koukoumidis]

When we started Oumi, we set out to build a better future for AI – grounded in the benefits of open source. Eight months ago, we open-sourced the most comprehensive repo for training foundation models end to end.

Today, we’re building on that by enabling every developer and enterprise to train custom models in hours, not months ⏱️ – a foundational shift in how teams build, ship, and own AI.

I shared why custom models are the future – and why now is the moment – in our new blog post (see first comment for link). 📖

Ready to jump to the new era? Sign up for early access and join our limited design partner program. See more at our refreshed website.

Proud of the world-class Oumi team: builders and operators who are shaping what comes next. 🚀

[basvanopheusden @basvanopheusden]

A few weeks ago, I started a new job at @OpenAI. I wrote a document about my interview process and recommendations for anyone on the job market for AI research positions. I hope it's helpful!

https://t.co/0I6f6UrAqD

[Shang Hong Sim @shanghong_sim]

🚨 Announcing DCVLR: The Data Curation Challenge for Vision-Language Reasoning! 🚨

@Oumi_PBC launches DCVLR, an open-source initiative to build the best reasoning datasets for Vision-Language Models (VLMs) 💫

🔍 Why: Vision-language data generation lags behind language-only efforts, with most strategies being closed. DCVLR aims to promote more open source efforts.
📦 Your mission: Curate ~10k high-quality instruction tuning examples that improve vision-language reasoning using any strategy (eg synthetic generation, smart filtering, or novel approaches). Reproducibility and scalability are key; novelty is optional.
🧪 Evaluation: Performance on VMCBench-DEV, LiveXiv, and OlympiadBench and others
💰 Prizes: 🥇$3,000, 🥈$1,000, 🥉$250 USD. Best projects will be presented at #NeurIPS2025
💻 Compute: No big GPU budget needed! Oumi handles all training & evaluation. Students can apply for free Lambda compute credits!
📜 Rules: Teams of 1–20, open to all (students, researchers, industry pros), max 3 leaderboard submissions per team.

📎 Starter kit, baselines, cloud credits & more: https://t.co/uOSH9MHvfr

Best of luck! 🍀

#NeurIPS2025 #AI #ML

[Shang Hong Sim @shanghong_sim]

Excited to be presenting Trust-Score, a metric measuring the groundedness of LLMs in a RAG setup and Trust-Align, an alignment method to improve LLM trustworthiness at #ICLR2025! 🥳

Interested? Join us at poster 228 tomorrow 10-12.30pm and for our oral at 3.30pm (Oral 4A).

📄Paper: https://t.co/VHAbP0h2ID
💻 Code: https://t.co/qezTKp10MW…