dilyevsky

the capex vs opex arugment would be persuasive if you couldn't lease servers. and obviously colo was always opex. basically the only time you'd get into serious capex was when you were building your own datacenters

> Its addresses are MAC addresses, serial numbers burned into each card at the factory.

oof

Nice idea, but execution could be better - way too many factual errors. Actually, considering it appears to be just 100% ai slop it's a little bit shocking how many just plain wrong statements it put out. Almost every section is at least a little bit off...

I met a number of really successful founders who came in just wanting to build a business and found a niche later (sometimes waaay later). I suspect majority of founders with “great ideas” discovered them after the company was started and the great idea they started with wasn’t actually that good

yes it's true, us court has ruled destructive scanning is required to make it "fair use" (ianal, etc etc) - https://www.ropesgray.com/en/insights/alerts/2025/06/from-bo...

author here, thanks! i tried putting both on the same plot in FIG 1 but it was kind of a mess...

https://apoxy.dev/blog/netstack-bbrv3#behind-the-scenes has more of the thought process and choice of tools.

The posters interpretation of why Envoy is using downstream on listener side and upstream on backend side is wrong. There's a formal definition in RFC 2616:

  upstream/downstream
    Upstream and downstream describe the flow of a message: all messages flow from upstream to downstream.

Now HTTP is obviously bi-directional but since for most reverse proxy usecases responses far outweigh the requests upstream means "the backend"

> security companies are going to comb over the weights and discover every hidden secret of the model

is this even doable? it seems plausible the model can be order-66'ed without it being so obvious

A lot of software projects don't make money directly. Our company makes money. This would have cost thousands of dollars paid to a contractor previously with unclear results so yes it's 100% justified.

You're citing a paper that does not reproduce in the year 2026 and a guy who built a following on doom&gloom projections with zero actual verifiable sources and demanding proof. SMH

Here is a TCP simulator I built using Fable 5 to reproduce BBR paper results: https://ccsim.apoxy.dev It took it ~4 hours to bang out initial version which was slightly broken then it took me another day to fix some bugs. A year ago this would take me more than a month of dedicated work, probably longer. A 50x productivity boost on pure coding tasks easily.

Try pointing Fable 5 to almost any CS paper - it can reproduce results in mere hours using maybe couple hundred bucks in retail token prices. This has been obvious to anyone caring to spent even just 30 mins for months now and the writing has been on the wall for more than a year.

"do not fool yourself" proceeds to fool himself... if you are honest about your bias that's still fooling yourself if your conclusions are just demonstrably wrong.

If you try actually filling out DS-160 and going through the rest of the process the difference will become really clear.

Which effort setting did you run this with?

A year isnt very long considering sums involved here. At a certain hyperscaler a full dc build out took 2-3 years but we heard that aws did that in a year all in. And also musk proved you can do it literally over holiday break but you don’t have to go that extreme - just dont let cya pmcs do it

Like 90%+ cost of a dc is actual computers which are pretty shippable

Why cant anthropic just sign up to those resellers, rack up some usage and then just start banning tf out of them? Reverse uno

Everyone says support but great support requires culture of excellence and customer obsession and that were the first things to go when those zirp-era scaleups went on hiring sprees

I know a few who are really feeling the pressure from customers now being able to vibe code part or their product and also their cloud bill is about to explode because hardware prices are through the roof

Pretty sure Carmack's idea of slack is what many companies would call "working as hard as possible"

Seems like this is a problem almost entirely solved by llm+vector database setup.

Don't get me wrong - AVP, Airpods, M-series chips are all amazing. Nothing as revolutionary as the first iphone, ipod, or an Apple II

Why didn’t those amazing engineers develop great hardware before or after jobs?

The context is AI made some knowledge work less cushy so now some folks are trying to point out random imaginary flaws (e.g TP or "water usage") when they're not busy trying to convince everyone AI doesn't work.

If it's http2 then Go's stdlib is pretty unoptimized to say the least. Huffman decoder is really cache unfriendly (pointer chasing) and I think allocation heavy too. Same probably goes for http1 and http3.

I dont think this comparison really works. Firefighter would be goalie or a defender and like you said in sports they are less appreciated/compensated for a simple reason - usually they don’t bring in views. There are exceptions ofc like Pippen or Seaman

Tom DeMarco had a whole book about this approach: https://www.penguinrandomhouse.com/books/39276/slack-by-tom-...

sure but there are tons of products from faang and other large companies that can be considered feature complete

Alternative ways of working

Twitter had laid off like 75%+ of staff and everyone, including on this site, was convinced it would crash and burn, yet it's still working. Explain?

nearly all big companies have failed too just on longer horizon

They're basically not building anything, and what they are building will be borderline unusable because of the route they chose so the car will be faster, cleaner and more convenient. If you want fast you can take the plane. So unless the problem is "how do we siphon as much tax dollars as possible banana republic-style?" I really can't agree here.