gopalv

There's a point when learning's fun, I think the OP is still there.

I wrote a bunch of realtime C++ in 2003, hated it. But last year, I wrote most of my code in C++ and liked it finally.

Lambda and auto was the tipping point where I stopped hating it. Some templates thrown in, but mostly to avoid writing a lot of branches for variants of the same function.

With lambdas I could now write the way I was initially taught programming, mostly by Lispy teachers channeling SICP.

Didn't hate the allocations either with unique_ptr.

This entirely depends on the teacher or the book author.

The "do not try anything new, but repeat after me" is a popular method of teaching things when we talk about skills, but often the next step is not just pursued but entirely discouraged.

There is a particular point where a teacher says "some of you might fail, but it is a sacrifice I am willing to make" & ask people to do something entirely new by themselves.

That is cross-over point where being self-taught has some advantages, because you have external guidance but not instruction.

If you're the kind of person who has failed a lot learning basics, then those new failures are less of a knock on your ego than if you had coasted through the early years without learning the parts which come after a failure.

Maybe if we all had a clear memory of "learning to walk", it would help - six+ months of falling down and not giving up, which most of us have done once.

> many was the evening when the Sandwich Maker and the Tool Maker could be seen silhouetted against the light of the setting sun and the Tool Maker's forge making slow sweeping movements through the air, trying one knife after another, comparing the weight of this one with the balance of another, the suppleness of a third and the handle binding of a fourth

The only morally acceptable form of bigotry in society is against the lazy & stupid.

So if you want to hate, you need to turn everything into one of those two things.

Apple's neural engine shows that you can live in between those two worlds.

As you said, the trouble is the latency, the programming model is still great.

The bugs that matter aren't potholes, they're trails with no markers, cut by someone you don't know who has left the organization years ago.

This is more like "What was your harder part of the Rubicon trail?"

I spent 7 months debugging a single issue which happened once a billion requests across 100k web servers or so.

For those 7 months, most machines did a hard restart every 4 hours and the bug never reproduces, so I didn't even know if I fixed it until I had enough confidence to turn that restart loop off.

Finding the issue had a fork of valgrind in the middle, found a CPU errata with AMD hardware when 2 locks placed in the same cache line (but not all failing machines had that CPU version), found a "bug" in UFS FreeBSD where deleting a file + recreating it uses the same inode id (not a bug, but not how linux worked)

The fix was 2 lines.

Your answer is not the wrong answer, but the bugs that you encounter in the wild is often like a tiger hunt in a jungle, alone.

The fudge factor always existed, it is now the airline's job rather than mine.

They have an incentive to pick a good-enough one, while adjusting it daily/weekly while I tend to fly once every couple of months and that too to a completely new airport most of the time.

If you do the fudging right and leave enough padding, it makes everyone happy - if you pad too much, you lose business as several normally possible connections become impossible by the published schedule.

Indigo did this in India a few years ago and the flights were 30-45 minutes longer on paper (like a 90 minute flight marked as 120), but because of their on-time reputation they got a lot of business customers who take multiple flights a day over other airlines.

AI models are more like a programming language or CPU architecture.

OpenAI is Intel and Anthropic is AMD.

At least in CA, rain causes more grass growth for fires to connect up through.

The fire material just builds up as growth pull carbon out and builds up a pile of it (because when a tree grows, most of its mass comes from the air & rain).

And then the fires were started by a "Lightning complex" on top of a hill.

Tractors and farming.

By turning what is traditionally a labour intensive product into a capital intensive one.

For now, the farmers who own tractors will beat the farmers who need to hire, house and retain workers (or half a dozen children).

This goes well for quite some time, where you can have 3 people handle acres & acres.

I'll be around explaining how coffee beans can't be picked by a tractor or how vanilla can't be pollinated with it.

In India, the 4-stroke came in big right around the time when leaded fuel got banned. The anti-knock stuff only really worked properly on a 4-stroke, so you'll notice that it took somewhere between 2000 and 2005 to really kill the new vehicle market for 2S.

However, the sound, that is what is completely different about these - you can actually talk on your phone sitting in the electric versions, while the petrol ones need you to shout over it when it goes up a hill.

The most interesting intersection for me was the mobile app and the EV charger combo the drivers use.

In my last visit to Kerala there were plug points under some street lights which could charge these 3-wheelers at 240v at 15amps off the pole.

These guys would just be parked right under one of those, with their mobile apps open & they'll unplug only once they get a pickup location, keeping their batteries perpetually topped up when idle.

The whole "no build step" was the most underrated factor for its success, because it was such a low-skill tool to pick up.

The tightness of that edit-test-move loop was why I used it, even if the "edit" included "sftp it to a vhost".

That bit is what made it successful & I can't think of anything I've used in the last decade which felt as easy as PHP for that particular style of self-taught programmer.

Digital tools mostly seem help me with long-term memory and doesn't do a thing for the working memory exercise.

I can look up a phone number or have a calendar on what I'm doing next Wednesday, but the working memory of "what am I doing right now" hasn't gone digital yet. Navigation has vaguely the same, because it's more long form memory rather than the hot list in my brain (maybe "what gear am I in?" on a manual car? doubtful).

Of all the day to day activities I do, cooking is one of those things where I've really felt like my working memory is working hard & despite quite a few tools like timers or automatic cookers, the way the kitchen is organized with closed drawers is a most-recent working memory exercise like those face-down card games.

Of course that happened and will happen again, but with the same results.

That's almost the exact origin of the word "Luddite", where people went around breaking steam powered looms.

The "treat AI like a smart intern" is basically the best scenario out there. I see this approach more often than the wielder of AI being entirely non-technical.

Github co-pilot at least is more like a power drill or a CNC machine rather than a robot writing all your code - or what a spreadsheet tool was to an accountant. The math is no longer the hard part, the rules outside turned into their domain expertise.

Teams of five are now cut down to one experienced dev or two.

That detail sucks much more for someone who is going to start in the industry next year than to the people who are already here. They're going to have to learn from the AI instead of people.

In India, when the dust kills computers by clogging fans, you'll see workplaces with computers have a "leave your shoes outside" board. I'm not sure what happens today, but this was true when I lived there.

That's mostly because everyone coming in is usually spent the last half mile on foot and the shoes will basically flake-off indoors in the lowered humidity.

Also, for some reason my dentist's office has a "employees must change into shoes" notice, suspiciously in the bay area.

Maybe they had some contamination issues ("work shoes" that could be santized properly?) or maybe they had employees wear sandals into the office & they had to enforce the rule after someone dropped a sharp on a toe?

The problem with "given more authority" I see is that management plucks these engineers out to make their day job basically "sit in meetings" if you're even slightly effective at simplifying life for everyone else.

Because that is the place of most leverage to place those people, but then those people are in a constant tug-of-war with the first group of "fresh ideas".

Eventually, the people who are in charge of the "prevention of bad architecture" become the bad guys because they (or me, I'm projecting) get jaded into just finding out what's wrong with something as fast as possible to be able to keep up with that workload.

You go from a creative role to a fundamentally sieve role with destructive tendencies, where you are filtering out the good from the bad as fast as possible, be.

First of all, not all new ideas are bad and "there's something bad about X" is not a "let's not do X".

Secondly, going from making things to shooting down things is intellectual suffering if you have a bit of empathy.

Some people on the "committee" with you don't have empathy & literally enjoy it - you're either trying to damage control on a case-by-case or building a "these assholes need to be fired" doc out of the meetings.

I realized what I would become if I conflated authority and respect ("respect my authoritah!").

Quitting was really the only way out of it. But it wasn't hard to explain to my spouse that i needed to leave for a job a level down and which paid half as much, because she could see me bringing my "why the hell do we have to do this" attitude home & to family decisions.

There might be no latency benefits to that, but there's always a throughput gain from using fewer cpu cycles over all - the cpu cycles you would have spent decompressing can be used for other things.

For that reason single-core performance would still be nice to have and always great to have open source code which is performance centric so that we all get to find out "How the hell did they do it?" (it is very readable assembly [1])

> I'd like to see compression speeds and performance on AVX if AVX-512 is not available.

The ARM is probably more interesting in the elastic compute world right now, the AVX world is unlikely to get cheaper.

And unlike before, I don't need to wait for a vendor to sell me an SoC, Gravitons are already here.

[1] - https://github.com/SnellerInc/sneller/blob/master/ion/zion/i...

I ride a motorcycle and I give Teslas a lot of cushion when riding behind. They're common enough in the bay area to be treated specially, because of the one pedal driving (my car is set to "creep", but it is still strong at 70mph).

The way the car changes speed and the way the Model S looks from behind, it takes a bit of time before you realize it is getting too big in your PoV without lights coming up, particularly as people lift off through turns or highway onramps.

The cars which are bigger or higher up like an ID4 or Rivian seem to not suffer from this sort of "hidden width" that the Tesla has, the wheel flares melt into the backdrop when looking at it.

I don't think Technology Connections is ever going to drive a Tesla or test it, but the low profile fast cars are much harder to figure out speeds off from parallax without the brake lights.

Well, if you have an incident list and nobody's checking it twice ...

The gradient of current moment is that whatever approach is optimized to use more data and more compute is much easier to invest in than something which can do more with less, but with a significant number of possible dead-ends.

At some point, this will have diminishing returns, but until that is hit, this makes sense as a purely return-on-investment for both research progress and business returns.

If you look away from the senators and consider the people the senators represent, it was intended to be even without the filibuster.

But what should be truly opposed is the cowardly way in which the filibuster is done today.

You should make 41 people vote against the bill, on record with their names, then go back to their constituents and explain why they did it in a town hall sometime soon.

Right now a senator can rely on the fact that their re-election is five years away when killing a bill which is supported by their constituents.

41 of them cannot assume they have years for the public to forget their vote on this particular thing (like hurting small businesses by inaction too).

The systems are scaled up for the "Black friday" sales event.

With cloud, you will plan for the steady state and just boost it up when you need to.

The anti-pattern is usually that when on-prem hits 80% load, some engineer is usually tapped to weed out all the slow code to push past procurement delays.

While in cloud, people just throw more hardware at it and ignore low hanging performance problems which are costing them money (& unlike the on-prem, fixing it will immediately reflect in the budget the next day).

My uncle is a 60 year old software developer, but he started writing software much later in life.

He started off in EE working for Sperry, went with the divisions wherever they went and switched over at the ripe old age of 47.

His previous work was all about state machines, so is his new work.

So "40 years of experience" isn't the same as 60 years old developer.

There's more of the EE/Physics grad turned software developer around than people who wrote software in the 80s.

The "securing" part aside, ssh still gives you a whole terminal experience very similar to the telnet one & the ability to draw into it (like if you were using Emacs on the shell host).

You can set something like that up by forcing a command on ssh instead of a shell (like what github does for git).

This is such a circular thing, that I feel like it is amazing to see it.

The reason LLMs use a NN is because they're trying to encode a probability function for generating the passage.

And now, you are encoding another n-gram follower exercise (i.e 1+1 = 2) on top of it :)

In Model 3 and Model Y, it is clear that Tesla is cutting corners where it can & the gap between those and a Model S Plaid is stunning.

The Rivian, EQS, Mach E and Taycan were head and shoulders above my Model 3, but also in a slightly different price range.

My car feels like a Toyota with a Tesla touchscreen the way it feels inside (at least it understands my accent when I ask it to change the temp on voice). The Kia EV6 GT is probably a close comparison or the new Prius in the way it looks.

Definitely overpaid to get access to the highway Supercharger network. My friends drove from SF to Chicago round-trip in a Model 3 in the winter break and when Tesla starts charging Rivians everywhere, it will be somewhat of a game changer.

Because of that, I think Tesla as sort of a Ford meets Exxon, they make a trickle of money off the chargers, more if they have CCS - I'm actually waiting for them to stab me in the back and ship a Tesla with a standard CCS, making me carry around a "dongle" to charge my old model.

The interesting part for analog is obviously going to be wear and repeatability, because the digital layer auto-corrects the errors until it becomes damaged and unusable clearly, while analog will slide off somewhat gracefully but you need a way to detect it happening multiple layers later.

AI is somewhat tolerant of errors since, the cumulative error across a lot of biases can approach zero if the errors are random.

[1] - https://www.youtube.com/watch?v=GVsUOuSjvcg#t=15m44s

Apple customers historically are often the most accepting of new things, even if that means change that involves spending more money, because Apple will cover the other end of change well (& expand a little that way).

The 3.5mm jack loss is tolerated because of airpods.

I'm old enough to remember the furor caused by Apple taking the 3.5 floppy drive when every publisher used Apple hardware for QuarkExpress & everything Adobe, replaced with this new-fangled usb thing.

The crucial thing over there was that Apple didn't bring in their own USB drive thing, but instead let IOMega try to deliver it & it went poorly for Apple customers who had big data swaps between them & others (DTP shops, magazine layouts, Photoshop artwork etc).

But it needs to be said and repeated, right?

Because people feel time-poor when it comes to matching what they want to do against what they can do. Building up a backlog is probably the worst way to kill the fun there.

And if everyone in that scenario feels like they are somewhat alone in that feeling where the "Books I wish I had time to read" turns into a prioritization exercise where you end up reading the "most important book" while thinking of a book you aren't.

You and the OP are saying the same thing, but it is worth repeating.

The longer you've been out of a structured learning environment like a school/college the more sense that makes because that is a constrained environment where optimization actually helps & the fun reading part isn't.

As for me, my library holds list is a good way to have a "river of books" where I can dip out of it and let it pass through my bookshelf on a schedule whether I read it or not.

This observation is sort of downstream from the Baumol effect rather than the effect itself or more accurately when the effect is countered explicitly.

The core loop of the effect is when you need to pay a healthcare manager a ballpark similar salary when putting out a "for hire" ad to get applicants or alternatively hide the salary filter till much later in the interview sunk cost.

The problem is that you end up having the managers first apply for other jobs before getting to the low-paying job, so the initiators of the interaction (i.e "apply for jobs") sieve out before getting to the job that pays almost the same but needs less competence.

So you pay about as much, but get even less value for money than paying more.

When you put Baumol effect, Dutch Disease[1], Gale & Shapley[2] and the Market for Lemons[3] together, you get to see the job market from a lot of different angles in my immediate neighbourhood.

Silicon valley has a Dutch disease for math teachers for instance, but also the Baumol effect for the English staff. Not complaining about them, I'd like my kids to learn history, math and english from great teachers & don't want to do Kumon or whatever else the other kids are doing after school.

The way the schools try to fix it is by making the schools initiate hiring through temps and do extensive adjunct periods before any concept of tenure to work around the market for lemons (you can't hire a temp managing director, which is what's different there).

This really sucks for the good teachers who want to have a happy late 20s in the career they prefer.

[1] - https://en.wikipedia.org/wiki/Dutch_disease [2] - https://www.jstor.org/stable/2312726?origin=crossref [3] - https://en.wikipedia.org/wiki/The_Market_for_Lemons