super_linear

The Jassy "Thoughts on Gen AI" memo was released a month ago and this article doesn't seem to reveal any new information on this beyond just suggesting "amazon might soon reduce cost to serve" without providing any real information https://www.aboutamazon.com/news/company-news/amazon-ceo-and...

Previously discussed: https://news.ycombinator.com/item?id=25222601 https://news.ycombinator.com/item?id=1627246 (the original)

Tried to link to the specific timestamp in the video, where the p99 lie is discussed. Core takeaway for me was that most users of a page will experience the p99 latency of the system: https://youtu.be/lJ8ydIuPFeU?si=HasLt5QBsB1tzww_&t=648

Amazon is using math to help solve one of artificial intelligence’s most intractable problems

Seems like pretty bad way to start this article off. AI is math at the end of the day how is this special.

Generally the article seems to lack a lot of substance or specifics on what or how effective automated reasoning is at improving results. Actual substantive studies/papers might be more interesting...

Also see: https://news.ycombinator.com/item?id=42952936

I liked how many of these solutions were pretty "simple" and easily applicable to many engineering teams. Our team similarly saved a ton by just "right-scaling" our AWS resources, applying TTL where possible.

https://archive.ph/1EfaF

Sorry, yeah I was a bit unclear. I was curious if there was any metrics to show how maybe server CPU/Memory or client-side latency changed after moving your app to Rust and running for a year.

But that's fair, if adding open-telemetry is maybe overkill to add to the application it might not apply as much to your use-case.

Nice article, but was looking forward to more of the "in production" angle (e.g. more operations, monitoring, performance). Instead this seemed to more focus on the the maintainability, developer experience using Rust and the ecosystem.

Similar to "The Cult of Done Manifesto" (2009) https://designmanifestos.org/bre-pettis-and-kio-stark-2009-t...

I love CarPlay/AA, and maybe this is hyperbole but not doing CarPlay/AA isn't that crazy of a decision. Tesla, GM, other big automakers have done this too.

https://github.com/SimHacker/MicropolisCore

https://news.ycombinator.com/item?id=36639530

https://archive.ph/Yffbh

Potentially vaguely related, the Eclipse foundation project GM is backing to establish an open source protocol for vehicle app / service communication: https://projects.eclipse.org/projects/automotive.uprotocol

Have been using around "USB condoms" to protect against this: https://www.amazon.com/gp/product/B00T0DW3F8

Is anyone successfully using this outside of AWS?

A toy Go project to run code as a non-root user and some CPU/time/memory limits

Seems like it's time for end of the quarter AWS demos and someone got a little too eager for launch.

Pull request eventually succeeded. Just "retry until the load balancer gives you a good node"

My bad, missed that the blog post was submitted already: https://news.ycombinator.com/item?id=35554460

It's a bit tedious to even try it in my experience: you have to use Unreal Engine for Fortnite and interact with the code through that (no publicly available compiler/VM afaik). They have some docs at least on setting it up but still depends on having UEFN which is a little unfortunate: https://dev.epicgames.com/documentation/en-us/uefn/verse-api, https://dev.epicgames.com/documentation/en-us/uefn/starting-...

(Not the commit author, just an "interesting" commit I saw)

https://github.com/closedai

This organization has no public repositories.

A nice touch

Sounds like a much cheaper "layoff" for Amazon.

re: https://fortune.com/2023/02/13/layoffs-cost-companies-billio...

For the hand-drawn style ones, I think it's https://excalidraw.com/

This looks awesome, I'm definitely going to try it.

...but I do think this quote on the homepage is a bit silly:

Used by the most productive developers at Stripe, Google, Microsoft, Deep Mind

(and so does Tesla)

A decent article on this: https://www.tomshardware.com/news/flash-memory-wear-killing-...

We have this example as a "don't do this" example for excessive logging in our engineering guidelines.

Absolutely no way to prove this but maybe Q1 deadlines coming up and people trying to launch things and make changes?