yalok

Exactly. Antiviruses are evil in this sense - crippling battery life significantly.

Wherever possible, I send “pkill -STOP” to all those processes, and stall them and thus save battery…

How does it compare with VMWare? I’d rather not use Parallels…

edit: whoever downvoted - please explain, what's wrong with preferring VMWare? also, for me, historically (2007-2012), it's been more performant, but didn't use it lately.

Chrome uses tons of APIs from MacOS, and all that code is very well optimized by Apple.

I remember disassembling Apple’s memcpy function on ARM64 and being amazed at how much customization they did just for that little function to be as efficient as possible for each length of a (small) memory buffer.

> the administration probably done some basic calculation: they get more in donations from alumni who want legacy admissions for their progeny than they get from Cal Grants.

The calculation was beyond basic - I read somewhere here that it was around $3m that they were getting from Cal Grants.

Around 8 years ago, I heard (from a friend of mine) that the min donation to guarantee admission to Stanford was ~$10m. Wouldn’t be surprised that it’s even a higher number nowadays…

Push notifications for e2e messaging apps carry e2e encrypted payload, which can’t be decrypted unless Apple reads the private keys from those apps sandboxes…

Just anecdotally, but this aligns with my observations on the trend/growth of successful useful open source projects that go with permissive BSD-like license. ~20 years ago there were way less of those than now.

And as a SW developer doing client side/apps as well, using GPL/LGPL is a total pain and basically cost prohibitive, unless I work on my personal small project where I don’t care about having to/risking to open source the rest of the code and getting sued/cloned…

Real life example from ~2010 - we ended up including an LGPL library in our mobile app code, and published/upstreamed all the modifications we did to that code (mostly ARM optimizations). Once the app became popular, our competitors came to us demanding the source code of our app - just because iOS didn’t support dynamic libraries (so we had to statically link it), and giving them the object code to relink it wasn’t enough for them (which would satisfy the spirit of LGPL), because they really wanted to see how we hacked around iOS camera input APIs…

What’s the expected additional latency due to running this re-ranker?

Still waiting for someone to do the same in Bay Area. Many parts of it don’t have any fiber optics options, even though Sonic does provide some in the north.

AT&T put an optic cable at my curb 10 years ago (most likely due to imminent competition from Google Fiber internet), but then never lit it (most likely because Google dropped their effort due to complications with cities)…

a friend of mine is a SW developer in Nvidia, working on their drivers. He was complaining lately that he is required to fix a few bugs in the drivers code for the new card (RTX?), while not provided with the actual hardware. His pleas to send him this HW were ignored, but the demand to fix by a deadline kept being pushed.

He actually ended up buying older but somewhat similar used hardware with his personal money, to be able to do his work.

Not even sure if he was eventually able to expense it, but wouldn't be surprised if not, knowing how big companies bureaucracy works...

even in regular languages with similar structure, sometimes the ending of a sentence forces you to change how you would say the whole sentence. Human synchronous translators usually correct themselves in such cases, which is a trade-off of having better latency in most cases, at the cost of having to correct yourself once in a while.

But if other countries chip in, this may snowball quickly.

I’m surprised nothing similar was done in US yet…

And yet even with the code we risk to miss some things from AI… so similar problem there…

Reminds a Conway’s Game of Life on steroids.

Did not realize it happened just so recently in California:

> In the 2022-23 school year, California became the first state to provide free school meals to any child regardless of whether they were eligible for the free or reduced-price meals as defined by the federal government.

Last 1 year I’ve been working full time on an integration layer between an end-user service and a few realtime LLM models that are part of that service.

The amount of code needed to achieve stability/predictability and address all kinds of edge cases is huge, and I have yet to see at least 1 use case where we can rely on LLMs answer 100% if it concerns any fixed state machine implementation etc.

Yes, these models are really good (just amazing!) at what classical CS approach can’t do around media and text processing, but they have such a hard time playing by specific strict rules…

So, CS focus will change, but it’s not going away… it’s more like we will end up with a better abstraction layer - like in 50-60s it was all in pure machine codes, then assembly, then C/etc, OOP, etc - here we will probably figure out even more elegant way to express unambiguous algorithm in a very succinct and very readable/maintainable way - and let LLM-based compilers convert it deterministically into some c++ code… (and those compiler may end up still having tons of classical code for speed/reliability/etc)

I’m not sure it was because they cared about security - looks more like accounting for 32-bit timestamp rollover would be very disruptive to the huge (legacy) code base and it was a quick fix to work around the problem :)

oh, ~4 years ago I talked to Sonic guys at length (great company, btw!) - they were too far north from us, and their estimation was to make it viable for them, they'd need to have around 200 of my neighbors commit to switching to Sonic at once if they lay out the fiber in our location.

Around 8 years ago I saw an AT&T truck on our street, and the guys installing some fiber into our street conduits. I was ecstatic & started checking AT&T website periodically to see when the service will be enabled.

Guess what? It's still not enabled. AT&T only did it because there was a risk that Google Fiber will do it in our city. Unfortunately, IIUC, Google never could overcome local regulations and abandoned the project. So, AT&T didn't care to light up their fiber (that was already in the ground & ready to go!!!).

Comcast doesn't offer any cable in my location neither.

I've been seriously tempted to do it myself too, but doubt I'll ever have time for that - mostly to overcome the local bureaucracy to get all the permits...

Huge respect to Jared!

but also they compare reasoning and non-reasoning models - e.g. Meta's Llama 4

anecdotally, NFC payment is my #1 use for Apple Watch, by far. It's so much work to pull out a phone and unlock it! :)

No quotes, just my observation - that the buttons are more optimized for the good look (symmetrical, sleek) vs usable. Building a plastic molding form with some shape/tactile difference shouldn't be significantly more expensive.

Re Touch screen, I agree - that's pure cost cutting. But that's just crazy. I'm surprised regulators are not preventing it.

I wonder how many people have literally been injured or died due to this pursuit for better "design". And it applies not only to touch screen (which is a total madness), but to real buttons as well - those buttons need not only to be physical, but have different form/some tactile differentiation.

What's happening with touch screens is total mess due to car designers trying to circumvent regulations (e.g., can't have that screen in front of a driver, or projected over the windshield) - so they are forced to move it to a side, which makes operating it while driving more dangerous.

But even button's design is falling prey of it, with buttons hidden way below the top of the dashboard, made of the same size, etc. It takes forever to scan & find the right button (e.g. internal air) while driving.

And I don't see it as just some recent trend (although there's much more of it). Even 10-12 years ago, this way already a problem.

One example of stupidity: Toyota Sienna 2013 has brightness control for its media screen (which also shows back up camera) via a touch screen button. Turning brightness at night works ok - but then turning it back on during day light is impossible - nothing is visible on that dimmed screen!! Latest Toyota Sienna has all of the climate control buttons (8?) in 1 row, very low, of the same size, barely visible signs on them - recipe for disaster...

I wish these designers and execs would have their kids life depend on it - I wonder if that'd change their thinking...

This has been going on for years. I used to do a lot of iOS development, and have an eye for bugs. Almost every Apple app/service has been regressing in quality.

Take basic functionality - a phone app (calling). After certain audio sessions use (calling via WhatsApp) I can’t make regular calls over cellular - the UI app immediately cancels the call. Only reboot helps.

Or notes - for many years/iOS versions, they lived with a bug where a text note may just become blank - and only restarting Notes app makes it visible again.

Or AppStore - if an app has to be updated (I have auto updates off) - and I press Update - it gets downloaded, installed - and then AppStore is back to showing “Update” button! If you just go to the app, it’s a new version. But if you press that “Update”, it will redo update from scratch.

Sometimes I’m so frustrated, and thinking of my options - it’s either move to Android, or go get hired at Apple with a mandate to fix bugs in various products… but knowing Apple secrecy culture/silos, it’s not going to work, and requires change in their hiring process/perf review/QA.

folks are working on those efficient lasers - e.g. https://xcimer.energy

The guy who literally was laying his life down to defend his nation, and who has hundreds of his fellow citizens die every day

vs

guys who are just afraid of the compromat that Putin has on them.

> You've spent so much time optimizing ML models that your own brain now refuses to process any thought that could be represented more efficiently with fewer neurons.

this! especially because lots of their content doesn't have audio on a specific language unless it's bought in that country's iTunes store.

This wonder if there’s similar research on reducing the amount of data (by improving its quality) for pretraining

What’s the size of the model used here?

What are the specific reasons that this is a problem?

Asking without any bias, just curious if there’re any other reasons than what I already know…