pradn

This problem exists with all shared code. Shared code, shared fate.

You hope that with more eyes on a library, it'll be more secure. It may also mean vulnerabilities have a higher blast radius. You have to choose where you want to lie on the balance.

Spectre and Log4J - problems on widely deployed tech (Intel CPUs, Java services).

The general shift to infrastructure-as-code is tied to a number of new trends and requirements:

1) Code review (config code can be reviewed by humans)

2) Automated pre-submit checks (config code can passed through automated pre-submit checks - such as preventing huge changes, or giving you a nice diff to look at

3) Auditability / history-tracking (you can look at the history of a config file to see who, when, why changes were made - this may even be necessary for compliance reasons)

4) De-duplication (you can extract common components into templates/functions - this supports variation across envs/regions/customers and repetition across tasks/machines/DBs)

All of these features help build large-scale systems in modern corporate environments.

Did they have samosas from Ken’s at grad tea time back then? :)

It’s not novel so much as an invitation to do the same.

The reading experience on print is infinitely better, just because it’s not tied to an electronic device. It’s too easy to be distracted on a phone or even an e-reader.

Every magazine, no matter how good the quality, has annoyances on their sites - cookie notices, newsletter sign ups, having to even log in, ads even for subscribers.

For the two magazines I like to read in print (NYRB, LRB), even the ads are useful (mostly new books).

This is just like FAANG where they forget to include a major player like Microsoft. (Maybe it was legacy before Satya, but Azure and AI have brought it back to the front.)

It's a mixed bag. Only sheer scale allowed Apple to pour in a huge amount of money into their M chips, probably the most important consumer-facing innovation on desktops in the past 5 years.

Google has as good a track record as anyone else for not shutting down Cloud services. Consumer services are a different category of product.

Europe isn't good at building massive tech companies. But its hard to say Europe is not a leader in scientific and technological innovation.

Just in tech, Europe developed ARM chips, ASML lithography machines, and DeepMind's AI breakthroughs. These underpin a few trillion dollars of global economic activity.

Yes, should have local models in addition to remote models. Remote ones are always going to be more capable, and we shouldn't throw that away. Augmentation is orthogonal - you can augment either of these with your own data.

But Steam updates happen in the background! It's pretty rare for me to be waiting for an update.

I'll bite - I've yet to see a big difference between 300mbps and 1GBps speeds. It's completely marginal.

Being able to save 8 minutes downloading Fortnite isn't a big enough deal. For video calling, you're limited by how much bitrate the provider is willing to allocate for you. For looking at photos on social media, you're stuck with Instagram compression. Internet browsing more tied to the number of requests and client-side parsing speed than total bandwidth.

Should our society be spending hundreds of millions of dollars on laying new fiber for marginal improvements?

What’s between the lines is that iMessage is critical as a way to lock in users to iOS. People care about security somewhat but they care way more about being ostracized for having green bubbles. I bet few common users could tell you the security properties of major messaging apps. This app, if allowed, would have shaved off a parentage points off iPhone market share.

> It’ll be several months till Gemini Pro is even available.

Pro is available now - Ultra will take a few months to arrive.

No, Google is on a more level playing field than you think. It certainly can't train on enterprise data, and of course not on private user data like emails. Cross-division data sharing is tough as well, because regulators don't like it for anti-monopoly reasons. OpenAI can scrape YouTube all it wants, but DeepMind may not be able to just train against all of YouTube just like that.

We might soon get to a point where every player is using pretty much all the low-cost data there is. Everyone will use all the public internet data there is, augmented by as much private datasets as they can afford.

The improvements we can expect to see in the next few years look like a Drake equation.

LLM performance delta = data quality x data quantity x transformer architecture tweaks x compute cost x talent x time.

The ceiling for the cost parameters in this equation are determined by expected market opportunity, at the margin - how much more of the market can you capture if you have the better tech.

I don’t see why it would take more than one PSF board meeting to approve funding for an event. Even if the board had questions, the follow-up is supposed to happen quickly. Several months to approve $9000 is silly. That’s probably just the catering budget for a 3-day, 200-person conference in the Bay Area - let alone for a conference for all of Africa’s Python community.

There's a bunch of S3-compatible bucket stores. I suppose one could ask their user for their S3 bucket key. It should be possible for an S3-like service to provide one-bucket-only keys w/ rate limits. The users knows where their data is and exactly when its being read/written. And when done using the app, the user can even revoke access.

Of course, having users manage a storage bucket and its key is a bit tedious. Browsers could generate a per-app bucket and provide the bucket key to the app, all with user permission (like when a site asks for location permissions).

These buckets can be shared between apps too. As long as apps can read the same data formats, data portability is stellar.

It would be great to have an ecosystem of apps that all use compatible buckets/formats.

But this is unlikely to happen bc this sort of transparency and data portability isn't currently a big demand by users / governments. Privacy activists have gotten a great number of wins over the past decade. We need to make progress in many more computing issues.

Well, sure, there's infinite room for scope-creep. That doesn't take away from a minimal solution that works well from the user's point of view - "I don't want to see this, make it go away".

New bots follow me on Twitter every day, spamming my notifications. They're very obviously a bot bc their name and lack of posts. I rarely had that problem before the downsizing. I think my follow count went up 20% in the past year, nearly all bots.

Otherwise, the service is generally reliable.

For something as simple as a dismiss button for a particular promotion, it's reasonable to store the bit on device. This will work for mobile and desktop apps, which I assume are the vast majority of Spotify usage.

Or they could invest in a generic "misc settings" column in their DB - to store random stuff like this in a blob. You could even query/index on them w/ something like Postgres's JSON support.

From what I hear, he directly expedited an office move just to bring Gemini folks together in NYC. So that is true to some degree. Not sure about the whole picture.

I wonder if people fall into a "happy valley of ignorance", where users don't actually see how an LLM can be wrong, rarely actually are met with hallucinations, and their use of LLM output is rarely a big enough problem. Whereas, we technical people who know its a bunch of matrix operations are so skeptical that we don't put this amazing technology to much use at all.

I’ve taught this game to 3 year olds and college students - everyone has fun with it!

I don't think users understand how LLMs are different are different from search engines, especially for information retrieval. Someone I know well has been using ChatGPT and Bard for months, and was surprised that they don't just use the Bing/Google search engine indexes behind the scenes. The idea that LLMs are a bunch of frozen matrices is not obvious.

It's hard to communicate that it's better to rely on LLMs for some classes of "reasoning" or language tasks vs simple information retrieval, particularly given retrieval does work well much of the time.

> I communicate with openai German and English how ever I see fit.

I did not foresee that you could mix input languages. That’s fascinating. Multilingual people use languages for different purposes, often in the same sentence. Ie: technical jargon in English, a joke in Arabic, etc.

I expect the difference in connotation/feeling/mood to be less relevant for an LLM, if you’re working with facts. But there was a recent post showing LLMs performing better when you said you were stressed/scared. Did you notice any such differences for your multi-lingual inputs?

Volume 2 ("The Desolate One") of the 2015 Portuguese film trilogy "Arabian Nights" has a 40-min story about bird trainers, as well. They love their birds, they love their songs, and they love to compete.

https://en.wikipedia.org/wiki/Arabian_Nights_(2015_film)

Ok wow, I didn't know the name comes from American Psycho!

There's an app called Dorsia, which lets you purchase seats at sought-after restaurants. The purchase is usually in the form of a per-seat minimum spend. They start at $75 per seat, but some go into the hundreds. This app solves a few problems at once: customers get to buy their way out of the seat-scouting process (keenly watching the apps for open reservations, walking in and putting their name on the waitlist, bribing the host, knowing someone with pull, etc), restaurants guarantee some of their seats go to high-spending clients (you could walk out with spending $35 at a good number of these spots), and companies have something to offer to their customers as a perk - Dorsia membership (the fancy gym chain Equinox gives their customers Dorsia access, for example). None of this is democratic, but that's the nature of luxury.

https://www.dorsia.com/

I worked from the Buenos Aires office, as part of my week long trip there. I was new so I didn’t have a million PTO days. The office space was beautiful, centrally located, with incredible food, and genial colleagues. We watched a World Cup match together. Same deal in Istanbul.

Engineers at these companies are like little startups. Senior engineers are expected to take a general problem, nail down the ambiguity / get consensus, and own the project til launch and after. That includes writing docs, running meetings, getting other people to do stuff for you (approvals, coding), writing code, testing, and ops/oncall work. In this way, the role requires "vertical flexibility" - a lot of jobs in one product. It's not "horizontal flexibility" - being a programmer in a bunch of different projects. That's the general gist of it, but I suppose everything can vary given how big these companies are.

Still, we like to speak of all these companies in the same breath, but I feel like unless you worked at a few of them, it's hard to say what the actual commonalities/differences are.