pradn

Eventual consistency gives you more availability, at the cost of consistency. It's fine for financial institutions to prefer this trade-off, because the any over/under error is a likely a fraction of the cost of unavailability.

Also, in the real world, we can solve problems at a different level - like legal or product.

The alternative is that everyone in the Microsoft security ecosystem gets off kernel-mode drivers. Once ebpf-for-windows lands, it should be possible for Microsoft, CrowdStrike, and everyone else to run their filters in user-land. That puts everyone on a level playing field, and makes the ecosystem more secure overall.

https://github.com/microsoft/ebpf-for-windows

It is a bit odd how they do it in Docs. In Cloud, announcements happen when the feature has already fully rolled out. Note, this is separate from region expansion, which might be delayed for some heavy-weight features (like new categories of VMs requiring special hardware.)

In Google Cloud at least, we're quite aware we're not the market leader, so there's a pervasive humbleness. We're proud of certain technical achievements (ie: Spanner), but the world can catch up quickly (ie: CockroachDB, Foundation, CosmosDB, etc). This might be a departure from the feeling in older divisions of the company, haha.

The NYC chapter of Papers We Love looks to be starting back up, after a gap of about a year and then two years before that. I think they're getting Datadog to host the events. They have a convenient office next to Penn Station.

https://www.meetup.com/papers-we-love/

There are a few examples of oral cultures ensuring perfect transmission across a large expanse of time.

One example is the several different recitation styles used to memorize Sanskrit verses. These methods give memorizers multiple ways to remember a line, and also prevent errors like the inadvertent mixing of adjacent words (" euphonic combination"). The "checksumming schemes" are far more elaborate than you'd imagine. [1] The result is perfect transmission of text and its pronunciation, including pitch accent.

Another example is a "multi-party verification" scheme in some Aboriginal Australian cultures. "... storytelling among contemporary Aboriginal people can involve the deliberate tracking of teaching responsibilities. For example, a man teaches the stories of his country to his children. His son has his knowledge of those stories judged by his sister’s children—for certain kin are explicitly tasked with ensuring that those stories are learned and recounted properly—and people take those responsibilities seriously. ... the 'owner-manager' relationship, requiring a story to be discussed explicitly across three generations of a patriline, constitutes a cross-generational mechanism which may be particularly successful at maximising precision in replication of a story across successive generations". [2]

[1]: https://en.wikipedia.org/wiki/Vedic_chant

[2]: "Aboriginal Memories of Inundation of the Australian Coast Dating from More than 7000 Years Ago", Patrick D. Nunn, https://doi.org/10.1080/00049182.2015.1077539

No, we’re quite aware the world outside has been catching up. There’s even a famous doc by a senior director about it…

The very very important stuff is hidden, and the only two examples anyone ever gives are core search ranking algorithms and the self-driving car.

Even the battle-tested hyper-optimized, debugged-over-15-years implementation of Paxos is accessible. Though I’m sure folks could point out other valuable files/directories.

OWNERS files rarely get in the way - you can always send a code change to an OWNER. They are also good for finding points of contact quickly, for files where the history is in the far past and changes haven't been made recently.

Readability really does help new engineers get up to speed on the style guide, and learn of common libraries they might not have known before. It can be annoying - hell, I'll have to get on the Go queue soon - but that's ok.

Yes, oops - fixed!

Google's monorepo, and it's not even close - primarily for the tooling:

* Creating a mutable snapshot of the entire codebase takes a second or two.

* Builds are perfectly reproducible, and happen on build clusters. Entire C++ servers with hundreds of thousands of lines of code can be built from scratch in a minute or two tops.

* The build config language is really simple and concise.

* Code search across the entire codebase is instant.

* File history loads in an instant.

* Line-by-line blame loads in a few seconds.

* Nearly all files in supported languages have instant symbol lookup.

* There's a consistent style enforced by a shared culture, auto-linters, and presubmits.

* Shortcuts for deep-linking to a file/version/line make sharing code easy-peasy.

* A ton of presubmit checks ensure uniform code/test quality.

* Code reviews are required, and so is pairing tests with code changes.

Andrew O’Hagan's article on Assange is rather famous, not only for its contents, but also for being 25,000+ words in a magazine that still pays per word. The LRB can pull it off because they're subsidized by the editor's family funds.

I don't think the pay angle got serious until after like Google went IPO - somewhere around like 2005-2010. Before that, tech workers were paid well, but not so much beyond like a financial analyst. Now there's a mass market of FAANG-M jobs, perhaps like 500k jobs in total. They all pay way more compared to the average job, in multiplier terms.

There has been a quantitative shift in tech worker pay at the top end in the past twenty years.

In other words, the past twenty years have seen the rise of Big Tech: billion-user products, trillion-dollar valuations, eating up entire industries (retail, publishing, consumer electronics, entertainment). What enabled them are software economies of scale (production + distribution), the US's ability to train and attract talent, cheap money, and access to massive markets abroad.

Windows XP was rather beloved as well.

Taste is a nebulous and rare thing. Microsoft rarely had it - it continues to flail around, with brief rays of hope - many of the Surface products are lovely.

Will Microsoft enshittify Windows? Yes, it already has.

There's a whole term for this:

https://en.wikipedia.org/wiki/Fifth-generation_warfare

This is so baffling to me. Is it just to reduce traffic to OneDrive or something?

Pernicious assumptions!

Google Cloud projects have three attributes: user-friendly names, system numbers, and system names. System names are alphanumeric. They can be chosen by the user, derived from the friendly name if there's no collision.

But! There's some system names from the olden days that are actually all numbers - so not actually alpha-and-numeric. Thankfully we don't run into those often.

Hyperthreading is also useful for simply making a little progress on more threads. Even if you're not pinning a core by making full use of two hyperthreads, you can handle double the threads at once. Now, I don't know how important it is, but I assume that for desktop applications, this could lead to a snappier experience. Most code on the desktop is just waiting for memory loads and context switching.

Of course, the big elephant in the room is security - timing attacks on shared cores is a big problem. Sharing anything is a big problem for security conscious customers.

Maybe it's the case of the server leading the client here.

This shop is brilliant, thank you!

These aren't creative tasks. They're research and summarization tasks. No one is asking a legal AI to write an opera.

Part of the reason why LLM summaries are so attractive IS a UI problem. The economics of the web has led to every publisher stuffing their websites with ads. No one wants that. Its much nicer to see a clean paragraph of text.

But we clearly have an ouroboros situation. If publishers lose views, they lose money and the ability to craft good information. Less new info to incorporate into LLMs.

LLM training over the internet corpus has really been a massive heist. Pulling a wool over publishers' heads, undercutting their business, hoarding the information.

But it's really unavoidable at this point. Everything has been democratized: compute on cloud platforms, data via Common Crawl, OSS algorithms and tool-kits. No one can put a stop to this, and there's powerful economic incentives to actually get some benefit out of the hundreds of billions that have been poured in already.

Your comment makes a lot of sense to me.

What I wonder though is - we've been a year and a half into the LLM craze and we still don't see a really good information processing system for them. Yes, there's chatbots, some that let you throw in images and PDFs.

But what we need is more like a ground-up rethink of these UIs. We need to invent the "desktop" of LLMs.

But the keys here, I think, are that

a) the LLMs are only part of the solution. A chat interface is immature and not enough.

b) external information is brought in by the user, and augmented by a universe of knowledge given by the provider

c) being overly general is probably a trap. Yes, LLMs can talk about everything - but why not solve a concrete vertical?

Semantic search helps with a part of this, but is just one component.

I'm with you on this. The vector embedding craze seems to be confusing mechanism and problem. The problem is semantic similarity search. One mechanism is vector embedding. I think all this comes from taking LLMs as a given, seeing that they work reasonably well with phrase-input semantic retrieval, and then hyper-optimizing vector embedding / search to achieve it.

Are there other semantic search systems? What happened to the entire field of Information Retrieval - is vector search the only method? Are all the stemming, linguistic analysis, all that - all obsoleted by vectors?

Or is it purely because vector search is quick? That's just an engineering problem. I'm not convinced it's the only method here. Happy to be corrected!

Thanks for extracting the details. It doesn't seem like they'll be competitive with blocklist-based approaches like uBlock Origin, because their features are fundamentally expensive to compute - parsing JS and such, not just matching URLs against a list of regexes.

What's fascinating here is AdFlush is a classical feature engineering approach: define a bunch of features on the data manually, and then use ML to figure out the most useful / impactful ones. This is not the "throw terabytes of data and see what happens" approach we see with LLMs. It's a bit funny to even point this out because I don't recall the last time a feature-engineered ML project made it to the HN front page.

Features can be brittle, but they are understandable. The paper's appendix [1] lists the 27 features that will likely make a request/resource "ad-related". These include interesting ones like JS AST depth, average JS identifier length, the "bracket to dot notations ration in JS", and a number of graph measures for the graph of scripts.

And contrary to what comments in this thread are saying, they do compare against a blocklist-based adblocker: uBlock Origin. That's in section 5.5. They say they outperform uBlock Origin. But even they say they don't reduce overall page time bc their algorithm is expensive.

[1]: https://dl.acm.org/doi/pdf/10.1145/3589334.3645698

I assume the majority of people are autoscaling merely for diurnal and weekly traffic swings, where the signal window could be as high as 30 min or 1 hour. Do folks really see sub-minutely autoscaling?

Are customers that are so cost sensitive served well by public clouds and attendant infrastructure?

Every commodity service like flights needs to be available through a standard API. We need to maximize the impact of market forces, via price discovery and then competition, for the benefit of passengers.

There's a lot of value in coming up with a clear problem statement and giving it to the right people. We see this with DARPA, at a much bigger scale.

A lot of the value in engineering leadership is in making sure the right problems are being solved.

There's a lot of value in coming up with a clear problem statement and giving it to the right people. We see this with DARPA, at a much bigger scale.