Lightnews — Scholar-powered news

Larry Osterman

@larryosterman.github.io

2.2K followers 510 following 1.2K posts

Long term MSFT employee, Dad, Singer. Expect cat pictures, tech nerding, etc. Previous: SMB filesystem, Exchange Store, Windows Audio, Windows Runtime, COM, Confidential computing. Current: Azure SDK for C++ Proud LGBTQIA+ ally. Profile pic from Not

Posts Media Videos Starter Packs

Larry Osterman @larryosterman.github.io · 1d

You see, only democrats like government functions like the scientists who track emerging epidemics.

MAGA has no use for that nonsense, after all, if they stopped testing the disease goes away.

So to hurt the demon rats they fire all the doctors.

Larry Osterman @larryosterman.github.io · 1d

Floured and browned in a pan, then put into a baking dish with the "puff".

Larry Osterman @larryosterman.github.io · 1d

Just saw that

Larry Osterman @larryosterman.github.io · 2d

As mentioned, this is a family recipe, it's not in any cookbook (well it might be in some church cookbooks from New Jersey in the 1950s)....

Larry Osterman @larryosterman.github.io · 2d

Mazel Tov!!!

Larry Osterman @larryosterman.github.io · 2d

It's not puff pastry - it's closer to popover.

Larry Osterman @larryosterman.github.io · 2d

Sequim (already mentioned) and Puyallup

Larry Osterman @larryosterman.github.io · 2d

Oh. These were whole eggs. None of that separated stuff. Crack in a bowl with flour and milk and whisk. Couldn't be simpler.

Larry Osterman @larryosterman.github.io · 2d

Happy Eggs. We can't have chickens, the bobcats, coyotes and raptors would use them as late night snacks

1 2

Larry Osterman @larryosterman.github.io · 2d

A half dozen cholesterol free eggs and almost a pound of cholesterol free butter.

Works for me. So yummy and healthy too!

2 4

Larry Osterman @larryosterman.github.io · 2d

I normally don't do food pics but tonight was worth it.

And it was so tasty.

And we don't ask how much cholesterol there is in the puff.

1 4

Larry Osterman @larryosterman.github.io · 2d

Sharron made chicken puff tonight and it came out pretty well...

Chicken puff. Chicken pieces baked under a popover like pastry.

An old Osterman family recipe.

Spooning a part of a homemade popover like crust and chicken into a bowl (off camera)

3 35

Larry Osterman @larryosterman.github.io · 3d

Hands down my favorite of the Dolly show.

I post it whenever I have a reason to.

Larry Osterman @larryosterman.github.io · 3d

Obligatory post of my chorus singing this amazing song

Travelin' Thru | Seattle Men's Chorus

YouTube video by Seattle Men's Chorus

youtu.be

1 2 6

Larry Osterman @larryosterman.github.io · 6d

When we anthropomorphize the actions of a cat, we never forget they are a cat, but when we anthropomorphize the actions of a llm, we often forget they are giant probability matrices spitting out the collection of words that most likely respond to the prompt.

Larry Osterman @larryosterman.github.io · 6d

One core problem I face every day is that we *constantly* use anthropomorphic language to describe the actions of llms - "ChatGPT suggested...", etc.

That's largely because we use it as a shortcut for what is actually happening, but means people *think* llms are more human than not.

1 2

Larry Osterman @larryosterman.github.io · 8d

The primary concern I have is that it will mis-measure concurrent operations because of concurrency issues inherent in the proxy/mock.

It's all solvable, it just takes work.

Larry Osterman @larryosterman.github.io · 8d

And it has its own set of challenges...

1 1

Larry Osterman @larryosterman.github.io · 8d

That is a still to be written feature of the perfect tests.

Larry Osterman @larryosterman.github.io · 8d

In this case the test had a warmup phase and a run phase, both of which were run from the same function and I forgot to clear the warmup results from the run results so the test measured the work done during warmup AND run, but only counted the time spent in run. So work/time was way wrong.

Larry Osterman @larryosterman.github.io · 8d

That's actually why C++ and Rust have the same performance characteristics - when you are measuring how many operations per second you can perform, if there's a 30ms network delay on each operation the network delay overwhelms any language specific effects.

1 3

Larry Osterman @larryosterman.github.io · 8d

No, the Azure SDKS have a defined methodology which is used for all the languages. To be honest, I have issues with the methodology (among other issues, network latency usually completely overwhelms the measurements), but it is what we have standardized on.

1 2

Larry Osterman @larryosterman.github.io · 8d

Can you increase the amount of work performed by performing the work items in parallel? If so, it is more scalable, if not, it is not.

IOW, can you increase work done by trying to do work in parallel or not. For the C++ SDK, the answer appears to be "no", for Rust, the answer appears to be "yes".

Larry Osterman @larryosterman.github.io · 8d

The Rust and C++ SDKs are essentially identical, *except* the Rust SDK is MUCH more scalable than the C++ SDK - there's something stopping the C++ SDK from scaling up.

1 4

Larry Osterman @larryosterman.github.io · 8d

I've spent the past week working on a perf measurement framework for the Rust sdk, and I was really excited because I was seeing consistent results showing the Rust SDK was 50% faster on the same operations as the C++ SDK.

And then I found the bug in my framework which invalidated the results.

3 9