juancn

A large chunk of the world doesn't have near enough infrastructure to support EVs. In many places they don't yet make sense.

It is too uniform, which makes it hard.

Everything is a list or an atom. There's no variety and it doesn't take advantage of pre-existing knowledge.

For example, in most curly braced languages a block is very different from an argument list, not so in Lisp.

There's not many visual cues of what function something does, anything can be anything.

Even math expressions are weird (because there aren't any), you cannot leverage years of training reading simple math notation to make sense of stuff.

I use the "what's going to kill us next" philosophy. Figure out what that is and do something to avoid it.

Wash, rinse, repeat.

Isn't this kind of obvious?

Parties (any kind) are entertainment and they compete with other forms of entertainment.

We have way more options to amuse ourselves now than a few decades ago.

Here's the full interview referred in the article in case you're interested:

https://www.youtube.com/watch?v=giTi8KWSZl4

It's not odd, it's a for profit company that's using it for marketing.

If you're far left the center is right, if you're far right, the center is left.

It's rarely a single cause, there's no root, it's a forest.

It's usually a combination of things that are not quite right, specially in mature orgs where all the simple fixes have already been applied.

It's semantic, if you do RCA right you don't stop at a single cause, but that's what usually ends up happening.

One thing that's often left out that tends to grind organizations to a halt is that continuous improvement processes tend to be additive.

You always add rules, and alerts, and so on. You need to revisit existing processes to and see what can be removed or replaced.

My other nit with these is a term that's abused a lot "Root Cause", most complex issues have many contributing factors, not a single root cause, and if you force the teams to find one, they will.

I dislike RCA (Root Cause Analysis) it puts you in the wrong mindset, CFA would be better (Contributing Factor Analysis).

That could be easily be corrected by relaxing merge gates for changes only to the `docs` folder (or some suitable naming pattern).

You can even do live edits on the web if you don't want to use a command line.

It's the price you pay for privacy

I still don't get how are they dealing with copyright here. Looks like the ostrich algorithm (hide your head in the sand and pretend nothing happens).

Those patches are not copyrightable unless human made. There's a reason why Oracle doesn't accept AI contributions to Java.

It would be much more user friendly if there was a way to check the serial of a particular PI and see it's original hardware config, manufacture date, and so on.

How do they define consciousness precisely?

That's where all these discussions tend to break down, if you get accurate enough with your definition you end up saying that a thermostat is conscious in a sense or reach a point where you cannot define it well enough.

Account lockout as the author mentions is the biggest risk on personal accounts.

Passkeys just make it harder/riskier.

Apparently it only happens on P-cores, recent E-cores have a fast path for subnormals.

I think the only way for now it through distillation of your own models, which can get expensive fast.

Wow! Close to boiling! /s

I wonder how (and if) continuous learning models will achieve stability.

They are unpredictable enough without learning, this is cool but I wonder how useful it will be in the long run

I like it conceptually, but the main issue for me with reduce is that it's hard to know exactly how the reduction will actually be executed.

The FUBAR potential with map and filter is much smaller, with reduce it depends on deep knowledge of the internals of the reduction itself, which makes it not as useful as a safe abstraction.

It would be interesting to have a LUT to exclude existing country flags and turn this into a random flag generator.

There aren't many places where a wearable doesn't get in the way, even an Apple watch it's unusable if you do kettlebell training (I severely scratched one doing a kb snatch).

I also stopped wearing my wedding ring because of lifting, it mangled beyond all recognition and had to cut it off with pliers and send it for repair.

Wealth inequality is mainly a statistical phenomena.

If you start a perfectly fair society where everyone has exactly the same amount of money and you play a game were people interact and with equal random chance (i.e. each exchange you get a 50/50 chance of earning 1 buck or losing one), after a few exchanges you get a Boltzmann-Gibbs (exponential) distribution.

If each agent wealth chance of increasing depends on what they already have, you get a pareto distribution.

It's extremely hard to avoid at large scale.

All my serious use notebooks run on Mac OS, my gaming PC on Windows, and my servers on Linux (I do keep an old-ish Thinkpad with Kali for tinkering).

The hardware+software combination of MacOS is unbeatable in terms of things like battery life and overall reliability, and it's a Unix, so the command line is almost the same. The M-series are a beast.

Linux has always been messy for desktop use for me (I've been using it since 1994). Sound always gives me trouble, and getting the power profiles right takes way too much work.

It tends to get in the way whenever I want to get something done, which happens a lot less often for Macs. A large part of it is the hardware, I'm sure, but still. I value my time.

Servers are fine, fewer hardware quirks to deal with.

Learning from stats (AI) has two big problems: it's non deterministic and much more expensive to compute than the deterministic alternative (when available).

The strength is dealing with ill-defined problems.

This last point is what makes it feel like magic, but the first two are the ones that can lead to disillusion.

The statistical approach only works if we can get computation cheap enough, sufficient data and if the problem can tolerate "good enough" results.

Any hot beverages: tea, coffee, mate, cocoa, etc.

Logo in Spanish in a C128 when I was in third grade started my journey in software.

After a while I was running laps around my teachers and I knew computers were for me.

Anything shitty that enters the context window shifts the entire thing to shittiness. So far this has been my experience on pretty much any model.

Context feeds on its output.

Once you it goes that road, unless you stop it and give it enough counter examples and details of what you want (i.e. you're nudging it on latent space towards a better spot), it keeps degenerating.

It gets even worse if the context window is compressed before you get a chance to correct.

Long horizon agents can degenerate at machine speed.

I still think you get much better results if you give them short horizon, well specified tasks.

That's usually true of all common hash table implementations (when objects don't have a defined order, if they have you can get O(1) average and O(log N) worst case), regardless of language.

The O(1) is the expected average case, which usually holds.

Yeah, O(N^2) is theoretically possible, but unless you're defending against some sort of denial of service attack, in practice it rarely matters.

Still, if you can guess a sensible initial size for a hash table you can avoid a lot of the overhead of rehashing.

I wish they would make one without all the health crap. Drop the sensors and add more battery.

I only really care about a few things (besides telling the time):

    - notifications
    - unlocking a Mac on proximity
    - Apple Pay
    - alarms
    - skipping a song

I couldn't care less about all the health sensing stuff, I have to take it out when I work out because it's either in the way or would break.