markasoftware

455 karmajoined 12 jaar geleden

https://markasoftware.com/

comments

1 uur geledendiscuss

Bazel maintains a system-wide cache that fixes this. At $current_employer a truly cold build (rm - rf the system cache) is 10 minutes or so. In a new worktree with a warm system cache it's under 60 seconds (still needs to build the worktree specific analysis cache). A fully warmed build is 10 seconds. Additionally there's no lock on that system wide build cache.

For this reason alone I want to like Bazel. But at the same time it has like half a dozen caches for different purposes and doesn't feel generally elegant. It saddens me that (a) cargo can't do this afaik and (b) its hard to package Bazel packages under nix. I'm not sure what other system has a shared unlocked cache.

eergisterendiscuss

Wtf do you mean by story? Performance and price are all people care about

4 dagen geledendiscuss

Why is no other inference provider offering similar prices then?

4 dagen geledendiscuss

Multiple providers (who need to make a profit) offer the same 4.40 rate for glm-5.2. It's not subsidized.

Deepseek's 0.86 or whatever is likely subsidized but alternate providers offer it for a price comparable to glm-5.2.

5 dagen geledendiscuss

I have friends who have had serious permanent vision changes as a result of taking a reasonable dose of psychedelics one time (which is one of the main reasons I haven't tried it). If you overdose on either lsd or alcohol you'll have serious ill effects.

6 dagen geledendiscuss

At the same time, using LSD once is higher risk than one night of drinking.

8 dagen geledendiscuss

FWIW I had this issue even with the MDB_NOSYNC flag so it shouldnt be force flushing to disk unless I'm out of ram or whatever

8 dagen geledendiscuss

Not amazing. In certain workloads I ran, once the db reached several hundred gb, writes would hang for longer and longer periods of time, eventually hours, while the db grew drastically in the background. https://news.ycombinator.com/item?id=30023623 seems to be the same issue, and it was serious enough that Shopify decided not to use lmdb.

And yes, I ensured there were no outstanding long lived readers, verified with mdb_stat -r. My workload used one transaction per read/write anyway (never needed larger atomicity). Once the db got into the bad state, running my program on it would almost immediately run into the issue again, so I really think the db is in a bad state such that most writes would cause it to hang, not related to how I do transactions. This workload would pretty consistently hit the issue once the db got to several hundred gb.

Issue #10236 on the OpenLDAP bug tracker might be the root cause, who knows. It's been marked CONFIRMED for years without a fix, while other similar issues are created.

This is extremely annoying. It seems workload dependent (other workloads I've run create absolutely massive lmdb dbs without this issue) and once it happens your only recourse is to make a new db and copy the contents over (thankfully reads still work fine on these borked dbs).

Other than that, though, it's great. Never in any case had actual data corruption, and reads and writes are extremely fast (until this issue happens)

Edit: fun fact, since shopify may have created Bolt in response to this bug, and then Bolt was the root cause of the 73-hour Roblox downtime in 2021, this bug may indirectly have caused one of the worst outages ever!

10 dagen geledendiscuss

Nym is pretty complicated and although they offer some kind of traditional VPN service as well IIRC, their main product is actually the only large scale operational "mixnet", which you can think of as kinda similar to Tor but a lot more resistant to certain types of deanonymization attacks (specifically it's designed to be safe against passive global adversaries, ie an attacker who can monitor every packet on the internet, though with the default settings at least it's not quite there yet: https://petsymposium.org/popets/2026/popets-2026-0055.php).

The research behind the mixnet is quite legit (look up Loopix). There's cryptocurrency involved because they're trying to do this whole complicated thing to incentivize people to operate nodes in the mixnet (though who are we kidding, the real reason is to make money). One can argue this is good for the long term sustainability of the network, and helps prevent sybil attacks by encouraging a large number of legitimate nodes. Of course, the downside is that the company is trying to make money, not just make the best mixnet possible. Having a profit motive also means they're incentivized to get the network to a point where interactive usage works well even at the price of anonymity (an optimization in this vein is sorta the reason that the paper I linked above is possible).

But having a working mixnet at all is extremely cool. People have been researching mixnets for decades without producing anything practical. While Nym is by far the most deployed / most ready mixnet rn, there thankfully are other options coming up, like https://katzenpost.network/ (though I believe most of the devs behind Katzenpost are also in some VC-funded situation, they certainly at least are trying to position katzenpost as more of a community driven project).

13 dagen geledendiscuss

are you able to say more about this (specifically that the researchers who left were concerned that qwen models were being closed sourced)? I was under the impression that the chinese labs and their employees aren't particularly concerned about ideology/safety/whatever and were just releasing open source models because it helps with publicity and does the most damage to the US AI labs. (and to be clear, I strongly support open source models, I just doubt that the Chinese labs and their employees are actually motivated by morals).

16 dagen geledendiscuss

The comment you're responding to is talking about books that students will "barely understand". You're talking about subjects teens aren't interested in. The comment above says nothing about interest and specifically does not advocate against teaching things to teens just because they aren't interested in them; only if they won't understand them.

6 maanden geledendiscuss

the thing is modal is running untrusted containers, so there's not really a concept of "some front facing" containers. Any container running an untrusted workload is at high risk / is "front facing".

If Modal's customers' workloads are mainly GPU-bound, then the performance hit of gvisor isn't as big as it might be for other workloads. GPU activity does have to go through the fairly heavyweight nvproxy to be executed on the host, but most gpu activity is longer-lived async calls like running kernels so a bit of overhead in starting / retrieving the results from those calls can be tolerated.

9 maanden geledendiscuss

I have a friend who did similar tunneling a while ago. It also works on cruise ships.

He discovered that on some airlines (I think American?), they use an advanced fortinet firewall that doesn't just look at the SNI -- it also checks that the certificate presented by the server has the correct hostname and is issued by a legit certificate authority.

My friend got around that restriction by making the tunnel give the aa.com SNI, and then forward a real server hello and certificate from aa.com (in fact I think he forwards the entire TLS 1.2 handshake to/from aa.com). But then as soon as the protocol typically would turn into encrypted application data, he ignores whatever he sent in the handshake and just uses it as an encrypted tunnel.

(The modern solution is just to use TLS 1.3, which encrypts the server certificate and hence prevents the firewall from inspecting the cert, reducing the problem back to just spoofing the SNI).

9 maanden geledendiscuss

if the world was all XHTML, then you wouldn't put an ad on your site that wasn't valid XHTML, the same way you wouldn't import a python library that's not valid python.

9 maanden geledendiscuss

Their example of why Ada has better strong typing than Rust is that you can have floats for miles and floats for kilometers and not get them mixed up. News flash, Rust has newtype structs, and you can also do basically the same thing in C++.

I don't know much about Ada. Is its type system any better than Rust's?

10 maanden geledendiscuss

I mostly agree that, as someone who cooks something intricate <10 times per year, a sharp knife is in fact more dangerous than a dull one.

Scott in the video makes the argument that sharp knives are safer, because you don't have to use as much force. But the only time I've ever cut myself with a knife in the kitchen have been with very sharp knives, eg one time handling it while washing the blade.

10 maanden geledendiscuss

as soon as an LLM makes a significant mistake in a chat (in this case, when it identified the text as Romanian), throw away the chat (or delete/edit the LLMs response if your chat system allows this). The context is poisoned at this point.

10 maanden geledendiscuss

per application, so per person

10 maanden geledendiscuss

same thing, but a game: https://brantagames.itch.io/motus

10 maanden geledendiscuss

I'm pretty confused about why it's beneficial to wait to read the whole compressed file before decompressing. Surely the benefit of beginning decompression before the download is complete outweigh having to copy the memory around a few extra times as the vector is resized?