Join us at NeurIPS for a happy hour on AI transformation, co-hosted by Percepta, GC and @nvidia: RSVP at https://t.co/6af38RxjlJ, and solve some fun puzzles at https://t.co/kaKAK2ECYu to skip the waiting list!

Today marks an important milestone. I’m launching Percepta together with @htaneja, @hirshjain, @tmathew0309, Radha Jain, @marisbest2, @KonstDaskalakis and an incredible team, with the goal of bringing AI to the core industries that run our economy. For AI to deliver

Excited to release our first set of reasoning models Phi-4-reasoning and Phi-4-reasoning-plus, available today on HuggingFace and Azure AI foundry. Some interesting insights below and more deep dives in following days!

New Phi-4 reasoning models have been released. Offer performance that is comparable to GPT-4o and o3-mini! https://t.co/S2ErrUHJVA #AIReasoning #GenAI #AI #Phi4 #OpenAI #AzureOpenAI

Introducing Phi-4-reasoning, adding reasoning models to the Phi family of SLMs. The model is trained with both supervised finetuning (using a carefully curated dataset of reasoning demonstration) and Reinforcement Learning. 📌Competitive results on reasoning benchmarks with

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.

We are hiring interns to work on LLM reasoning for summer 2025! Apply at

What is reasoning? Do LLMs use it? Does it help? Is o1 really that better than sonnet? How do you even measure all that? MSR AI Frontiers is working to figure it all out, and we're looking for interns to work on evals to better understand LLMs. Please apply!! Link below:

#Orca I'm thrilled to announce our latest work on Generative Teaching: generating vast amount of diverse high-quality synthetic data for language models to teach a specific skill (e.g. RC, text classification, tool use,math) without the extensive human effort typically required

With Orca, we're excited about the potential of redefining the reasoning capabilities of smaller LLMs. We're still at the beginning phases of this intriguing journey, but our preliminary explorations have yielded encouraging results. 1/7

Cleanly labeled and weakly labeled data are both crucial to few-shot NLU. WALNUT features a unified setting with both few-shot and weakly supervised learning. We hope to bring more attention to semi-weakly supervised learning for NLU. #NAACL2022 #NLProc @MSFTResearch

paper: https://t.co/1Gp0GIbtlF, code: https://t.co/oCsNM1WLMP. Joint work fom Microsoft, Dartmouth and Northwestern.

Integrating knowledge to NLG typically requires training/fine-tuning PLMs on knowledge sources. With KID, we show it's viable to infuse knowledge on the fly in decoding phase without requiring modifications to the LMs. Check out our #iclr2022 work on Knowledge Infused Decoding.

Two new NLU benchmarks for robust evaluation of PLMs.

Open to everyone! The first-ever 2021 Microsoft Research Summit, Oct 19 - 21, with over 150 sessions across 16 tracks, provides the global research community with an opportunity learn from experts pushing the frontiers of technology. Register now:… https://t.co/cK5qBSomc4

Just released a new survey on prompting methods, which use language models to solve prediction tasks by providing them with a "prompt" like: "CMU is located in __" We worked really hard to make this well-organized and educational for both NLP experts and beginners, check it out!

Excited to share our recent work on leveraging meta-learning for cross-lingual transfer.

https://t.co/bwz1PkfLzF #TheyPlayYouWin #TI8 #TeamLiquid via @NVIDIAGeForce

W

An Equivalence between the Lasso and Support Vector Machines http://t.co/31go7AqTTl