Neel Joshi

Researcher at Microsoft working in computer vision and machine learning

[Neel Joshi @neelsj]

My team at Microsoft Research, working in multimodal, AI is hiring! Please apply if you are interested in working at the cutting edge of multimodal generative AI. https://t.co/j134QSbPe4

[Microsoft Research @MSFTResearch]

MagenticLite's models are now fully open source.

MagenticBrain and Fara 1.5, previously available on Microsoft Foundry, are live on Hugging Face with open weights.

The app, the harness, and every model in the stack are now open.

[Microsoft Azure @Azure]

Meet Phi‑4‑reasoning‑vision‑15B in Foundry: high‑fidelity vision reasoning + high‑quality perception for real‑world dev workflows.

Learn more: https://t.co/wv22EnQO5R

[VentureBeat @VentureBeat]

Microsoft built Phi-4-reasoning-vision-15B to know when to think — and when thinking is a waste of time
https://t.co/exMXVFEss6

[Neel Joshi @neelsj]

So proud of this work from my team
and excited to share it with the world!

[Microsoft Research @MSFTResearch]

Vision-language models improve multimodal systems, but can make them slower, costlier, and harder to deploy. Learn how Phi-4-reasoning-vision-15B, a compact and fast multimodal reasoning model, blends strengths of different methods while reducing their limits: https://t.co/jP5L3AXRzX

[John Langford @JohnCLangford]

At NeurIPS, @KwangjunA is giving a semi-tutorial on distributed orthonormal updates Tuesday 4pm-5pm https://t.co/wihJcErNfe at Upper Level Ballroom 6CDEF .

These updates are worthl +10% to +100% more data, and the new Dion2 approach makes them tractable at any scale.

[Vidhisha Balachandran @vidhisha_b]

📌 You can now find all the evaluation logs (and reasoning traces for common benchmarks!) from our inference-time scaling report and the Phi-4 reasoning report at  https://t.co/skNOjClLxQ. The evaluation code can be found at Eureka ML Insights: https://t.co/FjviLeU889.

[Ahmed Awadallah @AhmedHAwadallah]

One of the secret weapons we had in doing the phi-4-resoning report is Eureka: https://t.co/Ncp7ZhbQr9

Eureka and our eval team have been doing an amazing job with adding new benchmarks to doing deeper analysis of results beyond single-score statistics.

[Dimitris Papailiopoulos @DimitrisPapail]

We’ve been cooking... a new open weights 14B Phi-4 reasoning model, SFT’d on ~1.4M carefully curated reasoning demonstrations from o3-mini and RL’d for a tiny bit. This model is a little beast.

[Suriya Gunasekar @suriyagnskr]

I am thrilled to share our newest Phi models. This time we went all in on post-training to produce Phi-4-reasoning (SFT only) and Phi-4-reasoning-plus (SFT + a touch of RL) — both 14B models that pack a punch in a small size across reasoning and general purpose benchmarks🧵

[Ahmed Awadallah @AhmedHAwadallah]

Tech Report: https://t.co/DWZjnO3SsU
Azure Foundry: https://t.co/fAf6HaiI1g
HF:
https://t.co/7kSAmy4kZ0
https://t.co/jITBpUXWUq
https://t.co/zxkgBKknse

[John Langford @JohnCLangford]

A new post: Headroom for AI Development https://t.co/smJu33g02M .

It's quite interesting to compare biological and silicon capabilities.

[Microsoft Research @MSFTResearch]

Announcing AutoGen 0.4, fully reimagined library for building advanced agentic AI systems, developed to improve code quality and robustness. Its asynchronous, event-driven architecture is designed to support dynamic, scalable workflows. Learn more: https://t.co/N7iSeR7ZJk

[Neel Joshi @neelsj]

Excited to announce the release of Eureka, an open-source framework for evaluating and understanding large language and multimodal models! I’m really proud of the team. This is important work and crucial to have it be public and open.

[Microsoft Research @MSFTResearch]

How can we rigorously evaluate and understand state-of-the-art progress in AI? Eureka is an open-source framework for standardizing evaluations of large foundation models, beyond single-score reporting and rankings. Learn more about the extended findings. https://t.co/3MM6aZn09O

[Neel Joshi @neelsj]

@besanushi @vidhisha_b @Jingya_C @hmd_palangi @VibhavVineet @AhmedHAwadallah Thanks to all! So proud of the team and the great and work. This is important work and crucial to have it be public and open.

[Taein Kwon @taeinkwon1]

I am delighted to announce our HoloAssist challenges at the EgoVis (https://t.co/YnOMwbXH6v) #CVPR2024 workshop.

HoloAssist is a large-scale egocentric human interaction dataset, where two people collaboratively complete physical manipulation tasks using HoloLens2.

[Ece Kamar @ecekamar]

Join us in shaping the future of AI!

AI Frontiers is a lab inside Microsoft Research with a mission to unlock new AI capabilities and solve real-world problems. We are hiring researchers and engineers in our teams based in Redmond and NYC.

Apply here: https://t.co/NlliHqF60S

[Besmira Nushi 💙💛 @besanushi]

📣Spring hiring: AI Frontiers at Microsoft Research is looking for researchers passionate about in-depth understanding and rigorous evaluation of foundation models and multi-agent systems: https://t.co/a8lKwqTJvS.

Learn more about the group here: https://t.co/oijt5OP9dm. This is an excellent opportunity to work with our AI&ML crowd: @ecekamar @neelsj @hmd_palangi @VibhavVineet Safoora Yousefi @vidhisha_b @SaleemaAmershi @julia_kiseleva @adamfourney @bansalg_ and many more.

Examples of in-focus research topics include: mechanistic interpretability, training and instruction data understanding, measurement methodologies and sanity checks for evaluation practices, evaluation of multi-turn and multi-agent interactions.

[Microsoft Research @MSFTResearch]

HoloAssist is a new multimodal dataset consisting of 166 hours of interactive task executions with 222 participants. Discover how it offers invaluable data to advance the capabilities of next-gen AI copilots for real-world tasks: https://t.co/LQ6lFQxLrh

[Philadelphia Phillies @Phillies]

It's been an unbelievable journey. We didn't just hear your support, we felt it. None of this would have been possible without you.

Thank you for everything.