breatheoften

I also tend to enjoy the spaces / service ... it seems like could be a bit short sighted of commercial real estate loan holders to let it shutter ... seems like a consortium of lenders with commercial real estate holdings could/should take it over and use it as a management service to try and shave their losses on all the vacant commercial office space that's gonna be everywhere ... it's not like there's really anything more valuable they can do with that space in the foreseeable future ...

I used a technique like this a few years back in a production product ... We had an integration partner (who we had permission to integrate with) that offered a different api for integration partners than was used for their website but which was horribly broken and regularly gave out the wrong data. The api was broken but the data displayed on their web page was fine so someone on the team wrote a browser automation (using ruby and selenium!) to drive the browser through the series of pages needed to retrieve all the information required. Needless to say, this broke all the time as the page/css changed etc.

At some point I got pulled in and ran screaming away from selenium to puppeteer -- and quickly discovered the joy that is scripting the browser via natively supported api's and the chrome debugger protocol.

The partners web page happened to be implemented with the apollo graphql client and I came across the puppeteer api for scanning the javascript heap -- I realized that if I could find the apollo client instance in memory (buried as a local variable inside some function closure referenced within the web app) -- I could just use it myself to get the data I needed ... coded it up in an hour or so and it just worked ... super fun and effective way to write a "scraper"!

OnDocumentReady -> scan the heap for the needed object -> use it directly to get the data you need

As much as I hate to say it -- I really think there is no room for the white house to negotiate on this. The reality is, appeasing the terroristic political element here only perpetuates a deeply corrupted "budgeting" process -- one that is entirely political and not remotely related to any form of rational governance. The long term effects of continuing to gift the terrorists in this way (like those that manifested under Obama as decades long sequestration) are ultimately worse than the incredibly dumb and incredibly avoidable consequences that will come from a default.

After the negative consequences of default manifest -- the one thing we can know with certainty is that there will be a powerful consensus around not allowing this effect to manifest again. The damage of the idiocy will ultimately be worth it to avoid continuing this tragic hostage taking drama forever...

The move here is to steal yourself that you _will_ allow default and say "the united states does not negotiate with terrorists" -- and admit absolutely no discussions about anything other than resolving the debt limit without any strings attached.

It's hard to imagine how anyone can view collocation of stock buybacks, employee layoffs/comp reductions and ceo pay raise as evidence of "good management".

So far there's a lot of talk about how ai will end lower-level jobs -- but the real opportunity for labor savings will be to evolve new company structures which don't have the modern form of centralized leadership with more or less fraud-driven leverage over company profits.

New companies modeled after teams that are assembled more like movies, with crediting processes for everyone involved that tie to broad-base profit sharing mechanisms will be the new way. The c-suite can go away/be dramatically consolidated into real leadership roles -- more hands on and direct leadership -- more like the director of a movie than a ceo of a mega cap, more like an Andrew Kelley than a Sundar Pichai)

More collaborations, more loosely affiliated, and more transparent profit sharing structures will be possible as more and more of the business and legal structural bullshit (which significantly over-determine modern business outcomes) can be automated away.

I have the same issue with slack and any facebook app frequently -- especially if you type, delete, type quickly. I think there must be some damn crdt within the draft edits feature that causes the final state not to match exactly what would be expected from the input sequence.

It's very annoying because it feels like I'm being gaslit by the device -- the errant results can pop in after some delay .... "was I misreading what I saw had been typed earlier on screen??"

I'm wondering what is going to happen to the banking sector as the commercial real estate apocalypse starts to manifest in full ... banks have a lot of interest bearing loans for commercial real estate that is going to continue turning into increasing quantities of illiquid pure junk in the very near term as leases end coincident with a mass cost cutting environment and occupancy rates drop even more ... obviously this has been a concern for awhile but seems like we are approaching a real inflection point

Anybody want to predict how the market will react to this?

My take is that it will be a negative response. They've already shed a lot of workforce -- at this point they are still facing revenue growth loss more than they'll save in costs from reduced headcount -- and I think it's unlikely they'll be able to fire folks into a convincing story about future profit growth ...

To me it's clear they are no longer a machine able to turn ever expanding quantities of engineering talent into ever ever expanding monetary growth ...

so what's left to value in their overall organizational brand?

Personally I think they have an enormous amount of technical and organizational talent still in their ranks -- but ... why? Why will they stay and try hard to chase after whatever the next gravy fad train might be? Seems like it won't be the meta verse ... and if the next fad is really going to be "ai" -- could there really be enough talent that can really stomache the thought of allowing instagram to be the one to control the worlds first super intelligent agi ...?

Who is commenting on this and what will they say??!

1-2 trillion tokens of training for gptx class models

It would take 20 years for a human to read that many tokens at 8 hrs of reading per day (although i'm pretty unclear on whether that's unique token sequences or randomly selected token pairs -- feels like a big difference between the two!!)

We are all here now -- who thinks we are anywhere of note together?

Personally I think that models that do not experience incremental change remain an entirely separate class of intelligence from those which evolve continuously -- but I am open to being persuaded otherwise (certainly I never thought making gpt2 bigger would be as impressive as gpt4) ...

It's crazy to me that no other hardware company has sought to compete for the deep learning training/inference market yet ...

The existing ecosystems (cuda, pytorch etc) are all pretty garbage anyway -- aside from the massive number of tutorials it doesn't seem like it would actually be hard to build a vertically integrated competitor ecosystem ... it feels a little like the rise of rails to me -- is a million articles about how to build a blog engine really that deep a moat ..?

I really enjoy Software Unscripted by Richard Feldman.

He interviews well-spoken programmers and skill leaders from a variety of language ecosystems -- and usually the guests share his appreciation for simple, strongly typed, functional and pure functional programming languages (like Elm).

Currently Richard Feldman is building a new language and ecosystem inspired by Elm but for non-browser application domains -- and he often discusses concepts and engineering trade offs as related to the design and implementation choices that are being made while growing this (to me quite exciting) new language.

I think "zero-copy" is the actual gem which might actually move us closer to making more of our systems have "less visible" plumbing. There's always going to be plumbing somewhere and it's not so actually directly actionable to observe that we have "too much of it". There needs to be a signal to guide a process by which we remove unnecessary necessary plumbing to obtain benefit.

I think there are significant advantages that will accrue by moving toward systems designs that encourage "more zero copy plumbing" -- not totally clear on all the ways the posix legacy might prevent that but the parts of posix that do prevent that will prove to be the most important pieces to shed I think.

Yah which is exactly why this response is not going to work.

The issue right now with "full employment" is incentive based -- after covid large numbers of people opted out of the jobs available to them because they finally recognized how much they have been being screwed over by the employer classes. The social contract is bad -- people are unhappy with raw deals.

I think the only thing that could actually work right now will require actually addressing the structural social wealth imbalances.

Too much employment --> introduce mandatory hour caps and increase mandatory vacation allowances and decrease retirement ages.

Too much overhead per employee -- get healthcare the hell out of the employment contract ...

It's insane to me that when we saw relative cross class wage growth for the first time in decades -- the immediate conclusion of all the decision makers had been that the labour market is over-exuberant and needs to be tamped down...

I can't think of a program more dangerous than one which intercepts and has option to receives every byte read from the filesystem -- while also having its own arbitrary network access ... the idea that employing such a tool could ever be better than _not_ having that tool is wrong at such a fundamental level that I don't understand how our computing society allows it to go on ...

Aside from any quirks in the rendering model -- at a more fundamental level I can't imagine what apple is thinking with the observable object data modeling approach -- and what [seems to be plans](https://forums.swift.org/t/pitch-observation/62051) to expand it ... it feels like the ObservableObject style modeling approach has been tried a million times in various frameworks and always results in complex and difficult to debug systems -- codebases which make it inherently incredibly difficult to reason about "what happens when this data changes"

A Unidirectional flow + data flow-like solution makes so much more sense to me as the basis for sane data modeling for ui's ...

why does it cost 8.6 billion dollars to build a marine corps base?

My big problem with the high frame rate in the hobbit was the slack of camera stability -- I think whatever image stabilization solution they used performed very poorly for that film -- not sure what went wrong but always hypothesized it was related to the higher capture rate.

It felt like the camera was always bouncing and moving around randomly in a way I found very displeasing...

I've wanted this for awhile as well. Squash only merges, which are enforceable in github, get you close but leave you without any automated way to determine if a given branch was ever merged to main or not ...

How/when do changes propagate to clients? Do clients cache the contents they read forever ?

$465 million for one season of TV seems very expensive.

But $465 million as part of the bootstrapping/rnd budget for some new gpu-adjacent compute offering would be pretty easy to justify ... and potentially being able to move some of the development cost associated with that new cloud offering out of aws and into some other budgetary boundary is almost certain to provide stock boosting effects.

It seems obvious to me that the tech companies with streaming platforms have started using tv/movie production budgets as a way to fund research into offerings appropriate for development teams that are highly-creative/technically competent/and driven by result oriented pressure. Adding value to the production stacks used by those kinds of teams is hard but also technically worthwhile and pretty measurable. And having discerning and hard to please customers as early partners/beta testers for a new service could be very valuable.

I suspect Apple is doing this kind of thing as well with their shows where you can see the same or similar high-end visual effects appearing on multiple of their shows -- feels like the kind of thing you'd expect to happen with different creative teams using converging 'in-house' tooling ...

wow I couldn't disagree more ... the search/replace interface of atom is the thing I missed the most from atom when I migrated to vscode ...

vscode _has_ added an non-sidebar find interface somewhat similar to the atom one -- but its not as good ... for one thing it doesn't support replace (find only)

There's no reason for layoffs to be time-correlated if that is the reason. Unless you think the sports analogy is strong enough to support ideas like "seasons"?

Dream team reasoning should ideally be applied semi-randomly -- if you need it to be enforced make the "reminder we are a dream team" messaging different and randomly chosen for different aspects of management and maybe even just introduce artificial delays to stagger firings so they are not clumped into groups ...

group firings are a wall street signal not a management strategy ...

I really wish it preserved key order ... is quite annoying losing this at the storage layer ...

I used to do Oceanographic research and we had a piece of equipment called a "shipboard gravimeter" originally built in the 1950s.

It was basically a very sensitive spring in a gyro stabilized platform attached to a bunch of analogue circuitry and was designed to measure the strength of the gravitational field (on a moving ship).

The manual for this equipment was amazing -- it literally taught you everything about the device and how it worked -- from the theory behind it to the purpose and function of each component.

In some sense writing a complete manual was easier to accomplish then than an equivalent manual would be now -- the a to d circuit used in a modern implementation would likely be more complex than the whole 1950s device ... but old manuals really are of a different quality and thoroughness than anything written for a modern audience ...

I tend to think it's going to succeed. Node is already getting displaced (by various server runtimes that hew more closely to the browser) - the true future of javascript remains where it has always been -- the browser.

"browser compatible" javascript that runs in places other than the browser can (and should) only grow over time ...

deno is kind of a rust-wrapper around v8 -- how long before it becomes the api for the javascript engine used by the browser rather than v8 ...

I think (eventually) "import some_browser_only_api" will become an optional thing you can do when you want to make your code depend on browser only functionality -- with code that doesn't depend on the browser-only functionality able to run without it ...

I think the oo features tend to not play as well with many styles of functional programming -- at least the forms of it that work well in typescript ... in typescript I tend to represent data as structurally typed "plain old javascript" objects and my program becomes largely just functions that operate on values and return new values. The idiomatic ways available in the language to copy and produce new values from old values tend to be straightforward and without as many gotchas when the values themselves carry all their meaning without the prototype chain. Once the prototype chain is a important factor in the program behavior you tend to have to keep track of "how exactly was a value with this shape acquired" -- i can't as easily just roundtrip it through some json for example -- and copying a value tends to not be as composable an operation with a tree of objects where the reachable sub objects all have behavior based on prototype chain.

When the prototype chain is involved I think the value gained from structural typing tends to decrease -- or at least exposes the programmer to more sharp edges.

Is anyone still using the class keyword in javascript or typescript? private field syntax doesn't matter in the first place if you don't use class {} anywhere ...

I feel like most of the typescript code i've been in recently looked like it needed 0 more class declarations.

How good is `debug.gem`? I'd be excited to try it if I was still using ruby ...

I'm not using ruby anymore but the debugger experience I had from vscode via the rdebug-ide library a few years ago was _mostly pretty_ nice aside from very performance limitations ... a massive performance improvement on that front must feel great!

Just because I haven't seen it called out anywhere ... I have an actual compliment about Monterey -- its a change which stood out to me as a major sign that perhaps old flaws are starting to become eligible to be fixed again - despite what Dragons they might've had to dig up and rebury to get it done :fingers crossed: ...

There is a completely new display settings UI for configuring multiple monitors AND ... -- it works great and its a huge improvement!

Seems like they are proposing a wealth growth tax rather than a wealth tax.

I wouldn't be surprised if it's not largely driven by personal profit motive/asset exposure of the decision maker class ... executives, boards, and upper management at such companies are nearly uniformly composed of wealthy individuals who probably have diversified investment portfolios including tons of commercial and residential real-estate interest in places like the Bay area ... what happens to real estate prices in the bay if tech becomes remote? Personally I can't imagine choosing to live in San Francisco in the first place -- but how many of the folks who did for work, then left during the pandemic to live in more livable places, would really want to go back if given the option not to ...?