Accepted (conditionally) at PLDI 2022: "Hardening Attack Surfaces with Formally Proven Binary Format Parsers" ... describing multiple years of work getting correct & secure parsers into the Windows network virtualization stack. #everparse #fstarlang Microsoft Research #pldi2022

Hate it when this happens 🟩🟨⬛⬛⬛ 🟩⬛🟩🟩🟩 🟩⬛🟩🟩🟩 🟩⬛🟩🟩🟩 🟩⬛🟩🟩🟩 🟩🟩🟩🟩🟩

I'm very excited to co-host with Mary Hall this virtual roundtable on creating stronger #regional #computingresearch #partnerships focusing on the #Atlanta area. We have fantastic speakers including Erwin Gianchandani, Charles Isbell, Greg King, Phyllis Schneck and Peter Williams.

📣📣📣 Social event for women PL scientists at PLDI 22 (W@PLDI) There will be food, booze, and other women scientists. Need I say more? pldi22.sigplan.org/track/pldi-202…

We are looking to hire researchers working in cloud systems and reliability. careers.microsoft.com/us/en/job/1321…

MSR Fellowships are open! Please apply! This year we are including Australia and NZ as well. Yay!!! Thanks to Steve Blackburn (@[email protected]) for prodding us to fix this. And to Jessica Mastronardi Lexie Hagen for doing all the hard work! Please DM me for any questions.

github.blog/2022-06-21-git…. This required a herculean task spanning multiple orgs. But happy to say that PL/HPC/Systems research significantly improved the performance of Copilot.

To the coffee aficionados out there, I am in the market for an espresso machine. What do you use and why?

A great keynote so far, with bonus Simpson’s content!

Cant get the picture of Marge and Homer and how they confronted the dragons of GPU communication out my head! Such an amazing keynote talk at PPoPP23 yesterday by Saeed Maleki! Read more about MSCCL at github.com/microsoft/msccl!

An excellent talk by Aashaka (Aashaka Shah) at NSDI 2023. Aashaka presented TACCL (usenix.org/conference/nsd…), a system that synthesizes algorithms for collective communication for distributed ML training. Joint work with Madan Musuvathi @vj_chidambaram Meghan Cowan and others not on Twitter

The Wooden Firehouse: An allegory for computer security [A/B test] You have lived all your life in a quickly growing town, whose growth has been sped up by constructing all the buildings out of wood. Some buildings in town are huge structures that have been repeatedly expanded/1

Okay, took a week long break from Twitter and I found this. Yikes. Roopsha, continue to rock wherever you are.

Thanks a lot, Sriram. And thanks to my collaborators who pushed the boundaries of what is possible with concurrency testing: Thomas Ball Sebastian Burckhardt Pantazis Deligiannis Lenin Ravindranath S Suman Nath, Shan Lu and many others.

I have seen the future and it's AICI github.com/microsoft/aici I've always wondered if foundation models will eventually support a more formal definition of prompt, beyond just natural language. This seems like a really interesting approach.

This is actually brilliant work by microsoft: github.com/microsoft/aici. Sending mini WASM programs instead of prompts to LLM providers would be incredibly powerful. Multiturn, , Grammar, COT, RAG, function calling, etc could all massively benefit from this

This repository enables you to control the output of LLM using code, allowing you to specify the format of generated sentences through regular expressions. This makes it more reassuring to build products based on AI. github.com/microsoft/aici

This is a fascinating idea not just for AI, but for APIs in general: ship WASM controllers to the servers which execute with a context, but allowing arbitrary flow control. github.com/microsoft/aici

Folks, give this new tool from RiSE a spin: github.com/microsoft/aici. Prompts are WASM programs and gives you a flexible/programmable way to control the output of an LLM. Semantics matter :)