Hasnain says:

This was pretty interesting. I’ve seen huge page remapping be quite useful at $work so it’s cool to see it talked about more broadly.

“Hey, sorry for the long article, but there was a lot to cover. I hope that it sparked your interest in using Huge Pages for code especially if you’re maintaining a large codebase. For further reading, I would recommend the paper “Runtime Performance Optimization Blueprint: Intel® Architecture Optimization with Large Code Pages”, which was instrumental to me.”

Hasnain says:

“Under Brazil's constitution, indigenous people have a right to their traditional land, and access to the land he inhabited, known as the Tanaru Indigenous Territory, has been restricted since 1998.

The areas surrounding the 8,070-hectare territory are used for farming and landowners have in the past expressed their anger at being banned from entering the indigenous territory.

In 2009, a Funai post in the area was damaged and cartridge shells were left behind in what was considered a threat to the Man of the Hole and the Funai agents protecting him.”

Hasnain says:

TIL. I understand why they have to make it harder to bypass warnings, but this is so inscrutable and hard to find.

(this will make my life so much easier)

Hasnain says:

This is some really cool stuff.

“Now that we can clone running VMs quickly, we can enable new workflows where you don't have to wait for development servers to spin up. Together with the GitHub App, you will have a development environment for every PR so you can quickly review (or run end-to-end tests).”

Hasnain says:

Great read going into a bunch of engineering tradeoffs. The follow up post is also solid.

“Here’s one of the first passages to jump out to me, from Chapter 3: Embracing Risk:

The low-latency user wants Bigtable’s request queues to be (almost always) empty so that the system can process each outstanding request immediately upon arrival. (Indeed, inefficient queuing is often a cause of high tail latency.) The user concerned with offline analysis is more interested in system throughput, so that user wants request queues to never be empty. To optimize for throughput, the Bigtable system should never need to idle while waiting for its next request.

This is a profound and general insight. When I read this passage, my last decade of abject suffering suddenly came into focus for me.”

Hasnain says:

The story they highlight here is really interesting. One frustrating takeaway I had here was just how much folks in computer science tend to reinvent things from first principles and ignore valuable work and research done elsewhere. Especially with AI, the stakes are super high - let’s not skip the ethical decisions, please. Algorithms are never neutral. It’s never just math.

“Along with all its faults, I think the Seattle committee also gave us much to admire. It was profoundly, even uncomfortably, honest about the hard choices at the center of kidney medicine. It refused to pretend that such choices were—or ever could be—entirely technical. And it tried, albeit clumsily, to democratize the values inside a complex, high-tech system. The Seattle physicians and their lay colleagues were rationing a scarce supply of dialysis treatments. But even after Congress provided dialysis for everyone, the shortage of transplantable kidneys was destined to spark similar questions, ones we still face today. And Scrib’s experiment with sharing the moral microphone (along with other stories I tell in the book) helped spark the system we have today.”

Hasnain says:

“But the same system also helped you to become badminton and football stars. Can you see why I’m asking what makes you go against the system now?

“No, we can’t. We’d have done better under a civilised, democratic system. Even without the failed system. Our society would’ve been better off,” says the former footballer, Hao Haidong, and he raises his voice:

"If the system made us into the stars we became, why aren’t there 20 more like me or 10 more like Ye?””

Hasnain says:

This estimate is staggering.

““It’s a ridiculous decline,” Anderson said. “When I saw a 6.6 year decline over two years, my jaw dropped. … I made my staff re-run the numbers to make sure.”

Life expectancy isn’t really a prediction for a single individual. It’s more like a check engine light — an indicator for the health of society as a whole. When more people die than would be expected, or when they die at younger ages than expected, then life expectancy will decline.”

Hasnain says:

Need to resist the urge to just spend a few days playing with stable diffusion at this rate. I do have some old art projects I could dig up for this (need to find the images I liked, still have the code though).

“4.2 gigabytes.

That’s the size of the model that has made this recent explosion possible.

4.2 gigabytes of floating points that somehow encode so much of what we know.

Yes, I’m waxing poetic here. No, I am not heralding the arrival of AGI, or our AI overlords. I am simply admiring the beauty of it, while it is fresh and new.

Because it won’t be fresh and new for long. This thing I’m feeling is not much different from how I felt using email for the first time - “Grandma got my message already? In Florida? In seconds?” It was the nearest thing to magic my child-self had ever seen. Now email is the most boring and mundane part of my day.

…

I’m just thinking about those 4.2 gigabytes. How small it seems, in today’s terms. Such a little bundle that holds so much.

How many images, both real photos and fictional art, were crammed through the auto-encoder, that narrower and narrower funnel of information, until some sort of meaning was distilled from them? How many times must a model be taught to de-noise an image until it understands what makes a tiger different from a leopard? I guess now we know.

And now I suppose we ride the wave until this new magic is both as widely used, and boring, as email. So it goes.”

Hasnain says:

It’s been crazy watching the explosion in this technology since DALL-E was announced - openai must be a bit annoyed that free alternatives are now available. I’ve been playing around with stable diffusion and the results are really good with only a couple minutes of first attempts - can’t even imagine how they would be with more work.

“As such, each image in the training set contributes only a tiny amount of information—a few tweaks to some numeric weights spread across the entire network.

But... the people who created these images did not give their consent. And the model can be seen as a direct threat to their livelihoods. No-one expected creative AIs to come for the artist jobs first, but here we are!

I’m still thinking through this, and I’m eager to consume more commentary about it. But my current mental model is to think about this in terms of veganism, as an analogy for people making their own personal ethical decisions.

I know many vegans. They have access to the same information as I do about the treatment of animals, and they have made informed decisions about their lifestyle, which I fully respect.

I myself remain a meat-eater.”