unusualmonkey

My 2c:

1) Is this worthwhile? Looks like ~500 CPU cores were saved (are these real cores, or does this include hyperthread cores?). I don't know cloudflare's costs, but this seems like single digit servers and probably savings only in the $XX,XXX range. Not nothing, but do you expect a positive ROI on engineering?

2) If you do want to go to this detail, did you consider in putting the filter at the deserialization step, preventing the headers from being created in the first place?

The Bellagio in Vegas is pretty awesome.

I find it ironic that you decided to try to justify natural diamonds... just to say that synthetic diamond purchasers always try to justify it.

Pot, Kettle and all that. With probably a large helping of selection bias.

How does a LLM compare to say .ZIP at compressing Romeo and Juliet?

How does ZIP compare to a LLM at answering a prompt to write a short story about cats in the style of Romeo and Juliet?

Because compression and generation are very different concepts.

Compression schemes take an item, reduce it in size, and return the same item (lossy or lossless) when asked.

LLM's and other generative AI are not designed nor are particularly good at this. What they are good at is returning new results using 'learning' it has developed via training.

If I want to send someone a sample of code, I'll use zip. If I want to generate code from a prompt, I'll use a llm.

Changing from png to jpeg is a copy (albeit lossy) - which is the exact kind of thing that copyright is designed to protect against.

If you took a pen and paper and traced over an image that's also a copy, even though you used a human.

Saying that Generative AI is a compression algorithm shows a lack of understanding of the issues and technology at play.

I think it's absurd to say that only say 3% of crypto transactions are criminal, if the majority of other transactions are meaningless.

Surely what we actually care about is how many useful, legal, meaningful transactions there are.

For example if for every 1 legal transaction there is 3 illegal transactions and 96 speculative or maintenance transactions... it starts looking like this is predominantly for criminal uses even though only 3% of transactions are criminal.

I love how you quote that, but not other lines like:

"We first document that 90% of transaction volume on the Bitcoin blockchain is not tied to economically meaningful activities".

How much of the actually meaningful use is legal? E.g. buying good and services, not evading laws, regulations, sales taxes etc.

So you can use it to conduct illegal cross border money transfers... of course most cryptos aren't anonymous and even those are you still need to transfer it to a local currency etc. for that functionality to be useful.

> Ask yourself, do Canada, Australia, New Zealand, the US, Britain, Ireland, etc, all have the same inability to build...

Why wouldn't they? They are relatively similar countries in many ways, any problem that one has could easily be replicated in others. For example I imagine zoning laws are all somewhat similar, favoring suburbs and strictly zoned areas vs market driven mixed use areas.

I also suspect they all have significant migration, high labor costs and high building standards.

> But our discussion is not about that. It's about the very basic thing in probabilities, which is some event is not as likely to happen as others. This is so trivially true.

Except that's not a given.

Any equally long random string is as likely as any other equally long random string.

Different length sets of random strings may differ in probability.

Finding what might appear to be meaningful structures in large data sets, e.g. shapes in 2T galaxies, doesn't inherently suggest anymore than chance.

> My point is that the probability of getting a proper English word is very unlikely. Most of the time, you'll get gibberish strings.

Sure, but given a large enough sample both will likely exist. So the fact that one happens to be english should not surprise anyone nor does it suggest meaning.

> Also, I didn't say the sentence to be encoded in morse code. Instead, the galaxies form the literal shape of "W", "E", and so on. I hope you can see that in this case, it's borderline impossible to happen.

I used morse as its easy to reason about. There's no reason to think shapes are impossible - you just have to define what makes a shape and then look for patterns that match.

Humans have been finding patterns in clouds, stars and even toast since time immemorial.

https://svs.gsfc.nasa.gov/30505

> So yeah, in this context, some shapes are more likely to form (randomly) than others.

Again I think you are confused. Assuming random distribution, 'We Are Coming' is just as likely as any other similarly long structure to form. You just happen to care about that structure more than others - however that doesn't make it more or less likey to form.

That message, in morse code is .-- . / .- .-. . / -.-. --- -- .. -. --..

There are 200B to 2T galaxies in the obeservable universe. If you found lines of galaxies and interperated them as morse code, I'm sure you'd find some interesting words/phrases being said.

You'd expect that phrase in every 2^28 = 268,435,456 random 28 digit binary strings - which is not very many. Keep in mind a galaxy could be part of many, many strings (different index position, different orientation of string).

Yes, a large set of instances is more likely than a single instance (all things being equal).

However that doesn't mean that a sphere is any more or less likely than any specific other structure. It's an small but important distinction.

No, a pyramid shaped planet is not as likely to form as a sphere shaped pyramid. Definitionally a pyramid shaped planet is impossible.

HH is just as compressible as HT or TH or TT.

You can easily build a compression scheme for any one of these values, but not one that encapsulates all values while using less data than the raw values themselves.

> If you throw 1000 dices, is it possible to get all one? Yes. Is it likely? Not at all.

That's literally as likely as any other possible outcome.

Let's simplfy this to a coin toss, which is more likely:

HHHHHH

or

HHTHTT

or

HTHTHT

They all have the exact same odds of appearing, we might just tell ourselves one formation is more special than any other.

There is no guarantee that local processing is going to have lower latency than remote processing. Given the huge compute needs of some AI models (e.g. chat gpt) the time saved by using larger compute likely dwarfs the relatively small time need to transmit a request.

Fantastic.

It's losing billions of dollars.

It's easy to make something people like... if you're willing to burn money.

IIRC large portions of China's high speed rail might be a boondoggle... so not sure that's the best example.

The other problem here is that the is very little moat here - all you need is a good screen and decent headphones, both of which are commodity.

The AVP is vastly overpriced and overspecced for basic media consumption.

> If you want a store in a mall, you have to pay higher leases and probably additional fees than you would if you operated out of your garage.

An iPhone is not a mall. An iPhone is a device owned by the end user. If the end user wants to use an app by some third party developer, they should be able to.

We don't let car dealers from prevent customers from installing aftermarket addons.

Why are you funding over half a million just for college?

So many words... so little actually said.

Sci-fi is inspirational, especially for people who use STEM to change the world. However, contrary to this opinion, this isn't inherently negative or bad.

Small form factor builds are a thing - for all we know they picked up another.more expensive 4090 that fit better.

If you truly believe this, please try to use 803.11a for daily needs.

I personally appreciate faster speeds, lower latency, better support for multiple devices, less contention, better support for wifi mesh etc.

Heck, I'm far more frustrated at wifi in 'corporate deployments' e.g. At a conference or hotel than I am with WiFi at home, so benefits there still improve my experience.

Seems strong to say .net is mostly showmanship... when your main contention is another framework has alpha support for a different GC.

The top trim was originally announced with 500 mile range. That's changed as if yesterday with no real announcement AFAIK.

Have you used those technologies?

Asp.netbwas great, asp.net cire was a good upgrade, and minimal apis are an optional way to get an asp project up and running with less boilerplate.

Not sure how that constitutes a cluster.