freeqaz

I work in security and this surprised me to see. Not that these companies got hacked, but the scope of the attack being simultaneous. Coordinated. Popping multiple companies at the same time says something about the goals the PRD has.

It risks a lot of "noise" to do it this way. Why not just bribe employees to listen in on high profile targets? Why try to hit them all and create a top level response at the Presidential level?

This feels optics-driven and political. I'm not sure what it means, but it's interesting to ponder on. Attacking infrastructure is definitely the modern "cold war" of our era.

Dang, that is unfortunate. I come here to comment that I'm really happy this happened. It is something that legit worries me. Bee populations are already suffering enough!

I hope the government is able to continue keeping the population from "stabilizing" to eventually get it eradicated. Crazy that you actually saw one!

What are the benefits of Kotlin over Java? Something I wish they went into!

Yes haha exactly. I'm sure I could figure out the code to slinkify any object which would be pretty fun

For a non-circle that still works for slinky purposes, I'm sure some sort of geometric prism would be cool. Maybe I'll take a crack at it over winter break!

I've been 3D printing slinkies (helical coils) and I used build123d for it. Somebody on HN recommended it as a first pass if you don't know CAD.

It's a little rough but I was able to get it to do what I needed it to! Here's my code: https://github.com/freeqaz/light-coil-model

Let me know if you want one. They're pretty rad!

Is Hackintosh even usable these days after the transition to ARM? I've been wondering if that will kill it or if the new Windows on ARM machines from Qualcomm will revive the market.

Just feels like too much special magic in Apple's hardware. But who knows?

I wonder if ESP-NOW[0] would be useful for this. I've been toying with building some mesh-based lighting controllers for synchronizing lights. Fortunately it can be entirely static (number of nodes) which makes it an easier problem than dynamic meshes.

0: https://www.espressif.com/en/solutions/low-power-solutions/e...

I wonder what inference of big LLMs might look like with that much cache and memory bandwidth. Not trivial to get a benchmark for that but I wonder

There are also a lot of geopolitical interests in helping Intel succeed. As much as investors might balk at Intel today, the amount of money that Western governments can toss at the problem is effectively infinite. (Looking at Defense contractors as an example; horrendously inefficient, but tax revenue is effectively perpetually guaranteed to keep flowing.)

It's silly to frame it this way, but even though the US and allies pay gobs for planes like the F-35, it's also what continues to allow the stream of taxable money to continue.

I view Intel in a similar light. Even if they stay inefficient, there will still be money for a long time because the West needs a way to be self-sufficient. The cost of not having a backup is too high (tens of trillions of dollars when I last looked into it).

It's a weird market that doesn't obey a lot of the conventional rules of capitalism. Yes, it would be better if Intel were able to compete without government subsidies, and maybe they will. But even if they don't it's very unlikely they'll totally crater because of how central this "power" is, and TSMC will never cede full control of their latest tech until the UN recognizes Taiwan as a distinct entity from China (imo).

I think the author means that it stores data on disk in a single file (aka SQLite!)

I think one of OPs points is that it is more of a boost for juniors and a "tax" for seniors. The senior engineer wouldn't open a PR without cleaning up the code; the junior can't tell the difference.

Seems like he wants to cut it and it will be hard because the Red majority wants it (lots of jobs in the South).

Seems like maybe the SLS++ could get canned. I am very curious to see what moves happen with this guy in charge.

Figma | Senior/Staff Security Engineer | Remote or Hybrid (US/CA) | Full-Time | https://job-boards.greenhouse.io/figma/jobs/4214880004

Figma does web-based software for Design and Productivity, kinda like Photoshop merged with Google Docs (live collaboration). We use FigJam a lot internally for meetings which is like a shared whiteboard.

Anyway, I'm on the security team and we're looking to hire more folks as the rapidly company scales.

The Security team is unique in wanting engineers with both a coding and security background. If you are a classic SWE but possess security-fu then you're welcome to apply! That's the route I took and many of the other folks on the team have too.

My sub-team, AI Security, works on the security of our new AI features (mostly hardening infrastructure and making the data we train on anonymized). My personal focus has been on security of our data recently (Snowflake mostly) which has been a new domain for me.

The other teams are focused around Application Security and Infrastructure Security (with some "endpoint" aka Mac security sprinkled in).

The roles are broad and if you're interested you're welcome to apply on the link above. I can respond to questions at fwortley at figma dot com too, but heads up that I may be high latency with the upcoming holidays!

Also note: The bar is really Senior and up. The company doesn't really hire below that unless you go through the internship program or have a strong personal referral for a team with an opening. (~5 years of experience being the rough rule of thumb unless you're already Senior at a 200+ engineer company.)

Anyway, happy holidays y'all and thanks for reading!

I can't tell if this is similar to PHA or not, but that's a plastic you can buy for 3d printing today that is touted to break down in seawater. https://www.fabbaloo.com/news/pha-colorfabbs-fully-biodegrad...

This is their site: https://colorfabb.us/filaments/materials/pha-filaments

Trying to run ZFS on a bunch of disks was hard. When I ran it on a laptop power management was miserable. It's just hard to use as a desktop system -- I switched to Arch and it's a lot easier for day-to-day.

One of the killer features of Qubes when I used it was the ability to "pause" a VM and all of the apps running in it. That's something I've tried to replicate with tools like tmuxp but I've never found an abstraction as clean as "serialize the whole process tree to disk" like Qubes has.

I gave up on it for usability reasons, but that feature is killer. Anybody else aware of anything similar?

Anybody know if there are any accessible large format 3D printers that have big nozzles? (Ie like >1mm)

I briefly looked but there were all crazy $$$ so curious if others know!

Where do you go for this information though? I think that's the tool that many people struggle to understand here. If you have the desire to learn more, how do you get there?

Damn they're giving under 3 weeks to export before you lose all of your data? That's insane! If I were on an extended vacation and missed this, my data would be gone when I get back.

They _really_ should increase that to be longer! Switch to read-only, but give at least 3-6 months to export. That's crazy otherwise

It's have to be deflation for them to be worth more. The only thing here that would be the equivalent would be the value of the silver. I'm _guessing_ that the silver is worth like 1% of the valuation (from looking at the photos). Doesn't seem like it's 2500 kilos of silver (~$1100 per kg at current prices)!

Is there a good POW mechanism that would test primes?

I found this but curious what else exists! https://en.wikipedia.org/wiki/Primecoin

But if the light is "warping" around the object then would you be able to grab the light from either side by intercepting it there? If you are at the X AU required for lending, but you sit at X/2 AU with some slight offset, you would be in the path of the light that would be traveling to X AU eventually.

I understand if what I'm trying to describe is impossible, I just don't fully understand why. (Is it out of focus? Is the sun too big/bright?)

Is there anything stopping you from putting 2+ satellites out "closer" but in the path of the lensed light, capturing the light simultaneously, and then resolving the image via async computation later? I think this is called interferometry and I know it's hard because you need _very_ precise timing, but I'm curious if that would be possible or not. (Maybe you can get the timing in sync with atomic clocks, or by sending a laser to both from a central point that lets them keep time with some very tight tolerance?)

Weird idea but I wonder if there are ways to take this from "crazy tech" to "hard tech".

I'm able to follow most of what you're saying. It's unclear to me what "convex hull" means though.

Also, where is each softmax happening here? For each attention head?

A 30bn param model, hell even a 7bn param model, is still incredibly useful and I feel like that could have been doable a decade ago!

We have GPT-4 (or at least 3.5) tier performance in these much smaller models now. If we teleported back in time it may have been possible to build

Oh my god, that's insane! This is the clearest example I've seen of why we need better privacy regulations.

There is nothing stopping a company from just spinning up and doing this to compete with PimEyes, and then you'd have to opt-out of that one too.

Absolutely wild!

That's about the same price as a 3090 and it's also 24GB. Are they faster at inference?

What's the best bang-for-your-buck AMD GPU these days? I just bought 2 used 3090s for $750ish refurb'd on eBay. Curious what others are using for running LLMs locally.

It did say that it was methylated and made bioavailable. Not great I think.

Sure, but there is some requirement to not just blindly copy everything over-and-over, and that is where I've seen things get tricky before. If you enable encryption you have to re-upload the entire snapshot periodically.

It's annoying because if you have TBs of stuff that blows. I'm just curious what systems exist for incremental, encrypted backups that don't require full uploading new snapshots.

See here in the NOTE section. Re-reading this, it might a limitation of Duplicity. https://www.rsync.net/resources/howto/duplicity.html