pradn

It was clearly a "tongue-in-cheek" framing. Got me interested at least!

It's clear that frontier agent swarms are going to be an important cyberweapon for well-resourced groups. Not just governments: organized crime, rebels. It will be feasible to train a deliberately misaligned agent, and to point them at adversaries. We'll have to deal with persistent, ongoing attacks, like we do with DDoS and ransomware attacks today. The limits on this will be intent, hardware cost, and talent.

You need an agent swarm to defend cyber-assets. No way we can respond at human speed.

Another consequence is that every public API with a bug bounty will be pen-tested up til the cost (in expectation) of running the AIs. We already see public AI-discovered vulns being reported to companies.

I'm not sure what this means for AI startups if their innovations can be copied by OSS so quickly (what, like 2 weeks?). There's "consumer surplus" for everyone, to borrow an economic concept. But we do ideally want some of the surplus to flow to the innovator, too. I know there were precursors, but that's fine - it's hard to have a totally novel idea in such a popular field. I don't know what the end game is for TypeSafe - they'd need to demonstrate perpetually better results, or compete in another axis: UX, support, custom solutions, etc. So much of the time, someone proving a concept, or it simply getting enough publicity, is enough for a "Cambrian explosion" of follow-ups and copies. Famously, that was true for "Attention is All You Need", and the general idea of "next-token prediction" being so powerful.

We've stumbled into general differentiable models..

The post is clearly in Claude English. Sorry not going to read a post in such an obnoxious style.

Well, yea. Everyone agrees with these points. Who wants to ruin things for the sake of it?

The reason why things get worse is because of systemic problems: misaligned incentives, information asymmetry, diffuse responsibility, etc. It's actually hard to improve these aspects of large systems. Hence, why the world is an imperfect place.

They have a "before" picture but not an "after"!

Suddenly turning off services when the billing cap is reached is a big reliability risk for customers. All workloads using that billing account would be impacted immediately.

If they weren't turned off at the billing cap, but were given some leeway instead, either that becomes the new hard limit, or GCP will have to give away the difference.

And there's no "middle ground" you could implement that makes sense either - like a "frozen" state. Preventing new writes to a GCS bucket breaks the writer app. Freezing VMs serving web traffic takes the site down.

Even if the service was shut down once the billing limit is hit, how long would GCP wait for the user to add funds or raise the limit? GCP would need to either keep the services in a hidden/frozen state or not turn them off / freeze them at all (in which case GCP would be giving away resources for free).

Maybe GCP can give users a heads-up when they're about to hit the limit? GCP already does - billing alerts do exist. It's just possible to blow past them if your usage is a massive spike.

Moreover, getting the hundreds of GCP services to implement a "frozen" state is difficult. It's hard enough getting everyone to listen to the "billing account disabled" signal, and (soft-)delete the resources (based on the resource, after some time interval). Given these billing overruns happen for smaller customers, it's not really worth solving the problem - which I don't think has a great solution to begin with.

How much do existing services trust new email service providers? It would seem to be an uphill battle for Cloudflare to start a new service. It's easy to automate from the start, meaning it's easier to send spam. I suppose reputation is not simply based on the domain the email comes from?

Brilliant piece of content marketing:

1) Pulls you in with a catchy title, that at first glance seems like a dunk on Cal.com (whatever that is).

2) Takes the "we understand your pain" approach to empathize w/ Cal.com, so you feel like you're on the good vibes side.

3) Provides a genuine response to the actual problem Cal.com is dealing with. Something you can't dismiss out of hand.

4) But in the end of the day, the response aligns perfectly with the product they're promoting (a click away to the homepage!)

This mix of genuine ideas and marketing is quite potent. Not saying this is all bad or anything, just found it a bit funny. The mixed-up-ness is the point!

A blog post like this is half the story. I’d like to see the results. Did your brother get more business? What were the failure modes? Did customers care if it was a bot or not?

They have direct feedback that many people return them bc they’re too heavy, and yet… It’s just Apple being stubborn. I guess it’s not a big enough problem for them, and they don’t care about losing the market. One must laugh.

Is AI-driven clean room implementation a wild west at the moment? I suppose there haven't yet been any cases to test this out in real life?

I'm glad you remember it as well! I didn't think to see if there was a recording or something of this talk, until now. It looks like the text of the talk was published here: https://www.cs.utexas.edu/~EWD/DijkstraMemorialLectures/Tony...

And the talk wasn't a random talk, but a memorial talk for Dijkstra: "The 2010 Edsger W. Dijkstra Memorial Lecture". I forgot this aspect as well!

That's right - it was bubble sort first. Absolutely - frail, yet sharp. I'm happy to hear several of us didn't forget this encounter with him.

Good to hear from you after a while, Gaurav (I think?!).

He came to give a lecture at UT Austin, where I did my undergrad. I had a chance to ask him a question: "what's the story behind inventing QuickSort?". He said something simple, like "first I thought of MergeSort, and then I thought of QuickSort" - as if it were just natural thought. He came across as a kind and humble person. Glad to have met one of the greats of the field!

The problem can be complex, which sometimes means the solution needs to be complex. Often, you can solve a complex problem in simple ways. There’s many ways to do that:

a) finding a new theoretical frame that simplifies the space or solutions, helps people think through it in a principled way

b) finding ways to use existing abstractions, that others may not have been able to find

c) using non-technical levers, like working at the org/business/UX level to cut scope, simplify requirements,

The way you can make complexity work for your career, is to make it clear why the problem is complex, and then what you did to simplify the problem. If you just present a simple solution, no one will get it. It’s like “showing your work”.

In some orgs, this is hopeless, as they really do reward complexity of implementation.

One of the odd things people do with tech is taking someone else's random projections at face value?

What does it mean to say "we were promised flying cars", or "every city would have micro-factories, that 3D printing would decentralize production"?

The people creating these narratives may a) truly believe it and tried to make it a reality, but failed b) never believed it at all, but failed anyway, c) or be somewhere else on this quadrant of belief vs actuality.

Why not just treat it as, "a prediction that went wrong". I suppose it's because a narrative of promise feels like a promise, and people don't like being lied to.

It's a strange narrative maneuver we keep doing with tech, which is more future-facing than most fields.

It's extremely painful that there's are free, OSS dictation tools that can run on-device, that are so much better than Apple's dictation, and yet it's quite difficult to use them on the iPhone. I'm referring to Whispr. Microphone access is a pain for custom keyboards -- for good reason, but still.

> When I was taken to the Tate Modern as a child I’d point at Mark Rothko pieces and say to my mother “I could do that”, and she would say “yes, but you didn’t.”

Actually, no you couldn't. The subtlety of the choice of colors, their shading, and their soft shaping, and the program of their creation over many years - you couldn't do that. They're lovely and sublime, and wonderful and an abyss. If you want to throw all that away and reduce it two boxes of paint, go ahead - but you'll be wasting a lifetime's engagement, of the joy of seeing with your intellect wide open.

> The value got extracted, but compensation isn't flowing back. That bothers me, and it deserves a broader policy conversation.

It bothers me, too. But, look at the history of the internet. There's no reason to expect we'll be able to fix this problem.

1. Search engines drove traffic to news/content sites, which monetized via ads. Humans barely tolerate these ad filled websites. And yet, local news went into steep decline, and the big national players got an ever-larger share of attention. The large, national sites were able to keep a subscriber-based paywall model. These were largely legacy media sites (ie: NYT).

2. News sites lost the local classifieds market, as the cost of advertising online went to zero (ie: Craigslist). This dynamic was a form of creative destruction - a better solution ate the business of an older solution.

3. Blog monetization was always tough, beyond ads. Unless you were a big blog, you couldn't make a living. What about getting a small amount of money per view from random visitors? The internet never developed a micro-payment or subscription model for the set of open sites - the blogosphere, etc. The best we got were closed platforms like Substack and Medium, which could control access via paywalls.

All this led to the internet being largely funded through the "attention economy": ads mostly, paywalls & subscriptions some.

The attention economy can't sustain itself when there are fewer eyeballs:

1. Tailwind docs have to be added just once to the training set for the AI to be proficient in that framework forever. So one HTTP request, more or less, to get the docs and docs are no longer required.

2. Tailwind does change, so an AI will want to access the docs for the version its working with. This will require access at inference time. This is more analogous to visiting a site.

All this measurement is useful only if you change your behavior in response. How often is this the case?

A common pattern I see is data-plane nodes receiving versioned metadata updates from the control-plane. As long as the version is higher than the node's previous one, it's correct to use. So, the metadata is a sort of monotonic counter with a bag of data attached to it. This pattern produces a monotonic counter, which I assume is a naive sort of CRDT - though the data itself doesn't benefit from CRDT merge semantics. In this world, as long as a node gets an update, its going to incorporate it into its state. In the article's terms, the system has Strong Convergence.

I'm trying to figure out under what practical circumstances updates would result in Eventual Convergence, not Strong Convergence. Wouldn't a node incorporate an update as soon as you receive it? What's causes the "eventual" behavior even after a node gets an update?

It seems to me the trouble is actually getting the update, not the data model as such. Yes, I realize partial orders are possible, making it impossible to merge certain sequences of updates for certain models. CRDTs solve that, as they're designed to do. (Though I hear that, for some CRDTs, merges might result in bad "human" results even if the merge operation follows all the CRDT rules.)

We can work toward closing security gaps with new technology, yes. It is necessary for large-scale adoption of LLM tech.

I didn't see this, thank you! They have a follow-up as well:

https://machinelearning.apple.com/research/ferret-ui-2

I have used it actually! It's been years so the fact that it's an accessibility tree just like in a browser didn't come to mind immediately. Both Mac and Windows have such representations for native apps. The actual functionality apps and the accessibility clients support is something like a two-way negotiation. A lot of stuff that should be supported in apps, in theory, is not, just because no client supports it, etc.

It's not a smart move to avoid integrating the most important capability advance in computing in the past decade - LLMs. They do support small cases, like summarizing text. But there's scope to do more.

They're stuck on the privacy angle, because what it means is you can't call remote services. You'll always have access to more resources at a data-center than a phone. So, while the frontier of what's possible with purely-local models will keep advancing, it'll never exceed what's possible with remote models.

People care about extra privacy when the delta in capability is minimal. But people won't allow a massive discrepancy, like the difference between a 8B model and a 700B model.

Apple needs on-device AI to do chores for me with the apps I have installed. Apple has everything it needs:

* Apps are already logged in, so no extra friction to grant access.

* Apps mostly use Apple-developed UI frameworks, so Apple could turn them into AI-readable representations, instead of raw pixels. In the same way a browser can give the AI the accessibility DOM, Apple could give AIs an easier representation to read and manipulate.

* iPhones already have specialized hardware for AI acceleration.

I want to be able to tell my phone to a) summarize my finances across all the apps I have b) give me a list of new articles of a certain topic from my magazine/news apps c) combine internet search with on-device files to generate personal reports.

All this is possible, but Apple doesn't care to do this. The path not taken is invisible, and no one will criticize them for squandering this opportunity. That's a more subtle drawback with only having two phone operating systems.

They have a super fast and slick file storage app. Some of the features that are natural additions to that feature set work quite well, like document scanning. But so much of what Dropbox does seems like they can't stay put and be happy with their core offering. Of course, they have to do this to increase revenue, for fear of becoming a mere commodity. It's tough.