It was much safer before the indexes decided to throw off all the safety hurdles designed to make sure stocks were relatively healthy and the price was reasonably well settled before they were included. And the core idea that a random investor probably can't reliably pick stocks that will out perform the broad indexes is probably going to remain true, even at the high evaluation SpaceX is a relatively small piece of the total index, it's just disasterous for public trust to see the safe guards thrown down like this imo.
if you're invested in a broadly diversified index you can be annoyed by this thing, but it won't impact you a lot.
E.g. if you owned something based on MSCI ACWI, which is free float weighted and global, SpaceX will end up being less than 1% of your portfolio even at a trillion dollar valuation.
It's just NASDAQ which is complicit in this scam by overweighting SpaceX.
I love go, but I find it did little to make concurrency management easier to reason about. Race conditions are easy to write. Go routines have all the same concurrency problems of threads.
In the parallel HTTP fetcher, the error is discarded. This will likely result in a panic when the response is nil. Also, what if it a server locks up? Or the underlying socket never connects and never times out?
I know it’s a toy example, but one must consider all these things in a real system. Go does have good pathways for these concerns, but it’s also easy to do it wrong. I still have to manually reason about access to variables/struct fields from multiple go routines.
I find I have to design servers from a data flow view; when my performance demands let me do everything via channels to worker goroutine pools (including sending a response Channel back to the request handler) I don’t have to do any locking at all. I find when I have to add in mutexes it means I haven’t thought enough about the problem (or I am fixing some hot spot from the profiler, rarely).
Its absolutely laughable that Gophers act like their language solves concurrency when its actually worse than Java because it has data races. Go added nothing and moved backwards.
The language that actually solves concurrency is Rust, but they will never acknowledge that.
The documentation reference a “decode” function, and it’s imported to the example code, but it’s never called. I’m not sure what the API is after reading the examples.
It’s the same assumption is required for any hole punching handshake (including STUN).
> This is a property called “equal delta mapping”
FWIW I’ve worked in computer networking for 20 years and have never heard it called this. This blog is the only source that comes up when I search for that exact term. I wonder where the author got it from.
> It’s the same assumption is required for any hole punching handshake (including STUN).
This is incorrect.
Hole punching requires being able to predict external port. That's it. If the port remains the same, it certainly simplifies things, but ports going up (or down) by 1 (or 2, or 5) with each new mapping is quite common, trivial to detect and to punch through.
I wonder how many new technical terms are going to be created by LLMs - not to say that this post was N
necessarily written by an LLM (but, who knows!)
It comes from academic papers on categorizing NAT behaviors which (trust me) is hardly the page turning research most people are used to. In these papers they talk about patterns NATs use between successive external port allocations -- which they call the "delta."
The name "equal delta" just means a type of NAT with a delta that tries to preserve the source port. Not to be confused with "preserving" type deltas (that preserve "the same numerical distance" between successive mappings -- e.g. a "preserving delta" type with a value of +1 means each successive NAT allocation is one more than the previous.)
The foundation models themselves will be cheap to deploy, but we’ll still need general purpose inferencing hardware to work along side them, converting latent intermediate layers to useful, application-specific concerns. This may level off the demand for “gpu/tpu” hardware, though, by letting the biggest and most expensive layers move to silicon.
The output of the transformation layers are a collection of embeddings in the latent concept space. Those can be fed into an addition model to extract semantic segments, bounding boxes etc. IIUC this is how dinov3 works.
reply