Congratulations to the Orca team on pushing the capabilities of SMLs to a new frontier. This release builds on two fundamental innovations: Using agents for more efficient data collection and building self-improvement loops for models. We'll see a lot more on these ideas.

Please join us tomorrow to hear about the AI work coming out of MSR and how we are organizing ourselves in this new era.

Episode One of Microsoft Research Forum is next week!

January 30 at 9:00 AM PT.
Register now ⬇️
msft.it/6017ivaL1

On “Abstracts,” Jordan Ash & Dipendra Misra discuss the parameter reduction method LASER. Tune in to learn how selective removal of stored data alone can boost LLM performance, then sign up for Microsoft Research Forum for more on LASER & related topics: msft.it/6016ia6As

2023 has been the most consequential year in Microsoft Research's 33-year history. In addition to countless internal (to Microsoft, OpenAI, etc) contributions, I bet that again our labs contributed more per researcher than any others to open research. Read more microsoft.com/en-us/research…

Don't miss Chi's talk on AutoGen today! #NeurIPS2023

Early attempts at automated/assisted evaluation of agents in AutoGen with very interesting results
by Julia Kiseleva Negar Arabzadeh

At Microsoft, we’re expanding AI capabilities by training small language models to achieve the kind of enhanced reasoning and comprehension typically found only in much larger models. msft.it/6019iGtzH

We just released a new version of #AutoGen and added compatibility with #OpenAI Assistants!

This means you can now make multiple GPTs collaborate to solve complex tasks 🤖🤖🤖

Checkout our new blog post for details: microsoft.github.io/autogen/blog/2…

📣We are excite to share that our team at MSR are looking to hire machine learning researchers and engineers who are passionate about working at the frontier of AI towards measuring and understanding progress in AI. Besmira Nushi 💙💛 jobs.careers.microsoft.com/global/en/job/…
jobs.careers.microsoft.com/global/en/job/…

Multi-agent frameworks is the next frontier when it comes to reliably solving complex tasks with LLMs. Read about AutoGen here: aka.ms/autogen-pdf

One of the most important questions of AI today is future platform technologies that build on top of large AI models towards reliable execution of complex tasks. AutoGen offers a new, multi-agent approach that shows a lot of promise. microsoft.com/en-us/research…

Yes, there is plenty of room at the bottom. MSR is open-sourcing Phi-1.5, a 'small' 1.3B parameter LLM that behaves like much larger models. arxiv.org/abs/2309.05463

Great summary of our efforts to better evaluate, understand and deploy multi-modal models in the open-world in a responsible way.

Highlighting amazing work from and with Hamid Palangi Ece Kamar Neel Joshi Vibhav Vineet Sean Andrist Xin Wang Hanna Wallach (@[email protected]) Solon Barocas Furu Wei Sebastien Bubeck Eric Horvitz Daniela Massiceti Cecily Morrison Su Lin Blodgett Dean Carignan Forough Poursabzi

At #ACL2023NLP , I will present a work on combining LLM-based data generation with human supervision to train smaller classification models that can have accuracy even higher than few-shot classification of the original LLM.

arxiv: arxiv.org/abs/2306.04140

details in 🧵:

Universitemi ziyaret etmek ve yapak zeka dunyasindaki sasirtici gelismeleri tartismak buyuk bir zevk olacak. Sabirsizlikla bekliyorum.

I'd never think the work we do in MSR would become a story in This American Life. Listen in to hear MSR researchers' reactions to interacting with GPT-4 first time. Peter Lee Sebastien Bubeck Ronen Eldan
thisamericanlife.org/803/greetings-…