jorlow

You can usually only get a few h100s at a time unless you're committed to reserved instances (for a longer time period)

Gpt (and many others) just add these embeddings together in the model, so you could do that and have one vector that encodes both things together

In your example I'd say it's more down to the team. If you have a solid team then you're set, but if you have low performers (which is sometimes the fault of the manager but certainly not always) then I imagine it could be more difficult to manage someone out remotely.

Much of the discussion is philosophical about commutes and keeping an eye on kids while wfh. But those issues aren't new.

What's egregious to me is companies committing to something which has major life and financial implications to the people that work there and then pulling the rug out.

They for sure all have contracts. If nothing else with non disclosure clauses.

Even in places that get a healthy amount of snow (e.g. in the Midwest of USA), many don't have snow tires -- it's a cost and inconvenience they try to avoid.

Starship is all about finding ways to whittle down margins from the status quo

I think this is mainly because such people are very career focused (and politically savvy to get to that point) and careful not to take a role that could even signal a step back. That and they typically will have enough financial security that they don't need to rush anything. "Lower level" folks are less likely to have both of these be true.

This is only partially true. The problem with LED lights is that they deviate quite a bit from a Planck and/or daylight frequency curve. So you can end up with colors looking different even if the color chart looks right. And even the best Aari LED lights have ssi scores of only ~75. I wouldn't be surprised if this is a part of the effect the author describes.

It's quite overkill for even intensive video editing. There are a number of YouTube videos of people downgrading from a m1 max to pro because the max was a waste of money and battery life.

This is the nature of tools, specialization, and how industries mature. This is the evolution of software eating the world -- now it's also eating development itself.

I was at Google (off and on) from 2007-2019. The company has definitely backed away from it both officially and unofficially.

> I'd venture this was written by someone who has a more traditional machine learning background that wants new people to the industry to have that same foundation

This is the vibe I got as well. Which is fair enough (to each his/her own), but I thought I'd mention fast.ai which takes the opposite approach:

> Harvard professor David Perkins, who wrote Making Learning Whole (Jossey-Bass), has much to say about teaching. The basic idea is to teach the whole game. That means that if you're teaching baseball, you first take people to a baseball game or get them to play it. You don't teach them how to wind twine to make a baseball from scratch, the physics of a parabola, or the coefficient of friction of a ball on a bat.

> TensorFlow feels like it was built from the ground up to scale up to billions of users and all kinds of devices, with developer productivity and happiness a secondary priority. PyTorch feels like it was built the other way around, prioritizing developer productivity and happiness; other considerations were secondary.

I recently moved from Google to Facebook and this is how I'd characterize most of the differences I see: Facebook optimizes for your ability to make progress above everything else. Google, not so much.

This video talks about the quality of spectrum from various LED light sources and why it matters. IIRC they suggested amber LEDs in addition to white. https://youtu.be/5U-F7EhLp7g

You develop features behind compile or runtime flags and keep it off until it's ready to ship. This is what chromium.org does so that might be a more accessible way to see it in practice.

At this point, androids kernel is really just a fork of Linux that's wildly more popular. I'm not sure why they need to abide by anything unless they want to.

I saw Brendan's talk live at re:Invent. Highly recommended.

> They advertise themselves into etcd, and any dependencies are pulled from etcd. That means that the Django app gets the addresses for the PostgreSQL and Redis servers from etcd, and connects that way. If these values change, each container restarts itself as needed.

Could you explain a bit more how that works?

The problem (which the article touches on) is that you have to resort to interface{} to make things reusable since Go doesn't have generics. So you have to pick between using a library (which will handle edge cases better) or compile time type checking.

Down for us as well

Updated: https://newton.apptimize.com/faqs.html

If you register (takes 2 seconds) there's an example project where you can see the integration steps with the exact code you need to paste in. You're right though, we should definitely add examples to the main site.

One of the biggest reasons is probably the reason behind using any software as a service rather than doing it yourself: you can concentrate on what you do best and not spend time managing things yourself. A couple of our users switched from using open source alternatives for that reason. We've put a lot of time into our infrastructure and decoupling things so we're very resilient against outages so you don't have to. We've also put a lot of time into making our library more robust and well tested than any alternatives we've seen to date.

For what it's worth, using Clutch was one of the reasons we ended up making Apptimize. We ran into several bugs including one that was crashing the app we were working on and issues with its probability distribution. And the reporting was very simplistic.

Are you using Clutch now? If so, it'd be awesome if you could give us a try.

Chromium compiles leveldb using the standard Xcode toolchain, but has a dependency on Chromium's "base" library, so it's for sure possible. I can't tell you the exact steps, but I suspect it wouldn't be too hard to get the posix port compiling with Xcode as well though.

All leveldb access is done in the main process (which also handles UI, network, most file access, etc).

I believe it's available in later 7 builds and all dev channel builds after that. Note that it is not yet available in the beta or stable channel builds (since we're still perfecting it).

Also, I really recommend you upgrade your Chromium version! I believe security updates are only back-ported to the current stable release, which means you haven't gotten any such updates for a while! (Stable is now at 8.)