cheald

Not only is it competitive, it's actually superior in many instances.

https://arstechnica.com/gaming/2025/06/games-run-faster-on-s...

Basically the only games that don't work are those which include anticheat which intentionally borks Linux. Check https://www.protondb.com/ for any game you're interested in to see if it'll run or not.

The one I was referring to was from this paper, first published in May: https://arxiv.org/abs/2505.20211v1

I don't recall how I found out about it, but it was either paperswithcode or an LLM research session working through the intruder dimensions problem.

In my Stable Diffusion tests, it substantially improves LoRA training speed and fidelity, though I've got some experiments that seem to even further substantially improve on it by adding learnable rotations of the singular vectors.

Standard LoRA (W_delta = B@A with standard inits) generally underperforms FT, primarily because of "intruder dimensions" (new high-ranking singular vectors which misalign with the singular vectors of the underlying weights) as outlined in the paper.

There are techniques like PiCa and SVFT which can mitigate much of the loss, though.

I'm not a doctor, but I am passionate about this stuff in my own life.

tl;dr: Exercise, sleep, and diet. Plus a zillion different supplements and medicines as adjuncts to a healthy lifestyle.

First, consider what inflammation is. It's fundamentally an immune response designed to attack unhealthy tissue and to facilitate repair of healthy tissue - the effects of inflammation are largely driven by cytokines like TNF-alpha (which is responsible for killing unhealthy cells and recruiting immune cells), IL-1 (which recruits immune cells), and IL-6 (which drives cRP production - the biomarker that you usually look for to gauge systemic inflammation). The production of these is mediated by nuclear factor kappa B (NF-kB).

Other major factors are things like reactive oxygen species (or "free radicals"), which can oxidize all sorts of things in the body and cause damage (which is good when the thing being damaged is a pathogen or damaged cell, bad when it's healthy tissue). Damaged tissue provokes immune responses.

So, if you want to "reduce inflammation", you want to:

1. Reduce stimuli or downregulate processes which are causing the production of inflammatory cytokines

2. Upregulate the production of anti-inflammatory cytokines

3. Ensure sufficient antioxidant capacity to deal with ROS production and oxidative stress

If you've got a chronic illness or autoimmune disorder, you're dealing with inflammation just because your "make immune defenses" signals are stuck on. But you can also have chronic inflammation through too much fat (adipose tissue is an endocrine organ!), environmental or diet factors, or just behaviors which result in an imbalance between pro- and anti-inflammatory responses in the body (for example: smoking induces consistent tissue damage, which drives immune responses).

Exercise upregulates production of anti-inflammatory cytokines and improves mitochondrial efficiency, which results in less ROS production during cellular respiration. Sugar surges cause elevated ROS production, and chronically-elevated blood glucose results in insulin resistance, which promotes inflammatory cytokine production. Lipopolysaccharides from gut bacteria in the bloodstream stimulate immune responses. Insufficient sleep upregulates NF-kB directly, but also contributes to dysfunction of other systems which can upregulate NF-kB.

If you're sleeping plenty, eating well, and exercising, and you don't have a chronic health condition, your inflammation levels are probably pretty good. But you can generally further reduce them with supplementation of things like:

* Omega-3 fatty acids (which compete with omega-6 fatty acids - "seed oils", which produce inflammatory prostaglandins) - this is why your doctor wants you to take fish oil

* Turmeric, resveratrol, and green tea extracts (which contain compounds which inhibit NF-kB and are ROS scavengers),

* Vitamin D (which inhibits cytokine production and supports your natural antioxidant systems)

* NAC, which replenishes glutathione (the primary driver of the body's antioxidant systems)

There are medications, of course, like your regular old aspirin and ibuprofen, which reduce prostoglandin production (which is one upstream of NF-kB), corticosteroids (which block NF-kB), as well as more exotic entries such as GLP-1 peptides (which, among other things, improve insulin sensitivty and reduce adipose tissue, which results in reduced systemic inflammation) or BPC-157 peptides (which acutely inhibit NF-kB, upregulate antioxidant enzymes, and help regulate nitrous oxide, which is how they can help heal NSAID-induced leisons).

This is by no means comprehensive - there are plenty more mechanisms and interventions to explore - but it should be a pretty good clue as to why "diet and exercise" are standard health advice. You don't want to turn off your inflammation responses - they're responsible for taking out pathogens, killing tumors and maintaining a healthy body - but you don't want them chronically upregulated, either.

Acetaminophen doesn't deteriorate the stomach lining like NSAIDs, but it also isn't anti-inflammatory. It's an analgesic and antipyretic.

The use of Stingrays to conduct mass surveillance dates back decades, yes.

FWIW, I replaced my Nests with Centralite Pearls a few years ago, and have been extremely happy with them WRT Home Assistant. The Pearl doesn't seem to be widely available anymore, but any Zwave or Zigbee thermostat + a local hub gets you a thermostat that should work with Home Assistant and will be immune to being sunsetted like this.

Same boat here. I'm 6'10", 260, and athletic. I'm very acutely aware of the shrinking space available on aircraft, and there's precious little that I can do about it. I typically just get an aisle seat and raise the armrest after takeoff (and then spend the whole flight dodging people going to the bathroom and refreshment carts, but it's better than nothing).

There is a problem with heavily obese people taking up significantly more horizontal space than is actually fair, but it's absolutely compounded by the airlines shrinking the per-passenger space to the size of the average passenger.

This would end up having effects like causing employers to discourage their employees from drinking coffee or exercising, both of which acutely raise cortisol levels.

They also breed extraordinarily quickly. Females are capable of reproducing at only 3 months old, are fertile year round, have a gestation period of only 115 days, and produce litters of 4-6 per pregnancy. Even with control efforts, populations have been growing at a rate of ~20% annually.

That's true of semaglutide, but newer peptides like tirzepatide (a dual-agonist) and retatrutide (a triple agonist) have additional effects like improving insulin sensitivity, and simultaneously slowing the release of glucagon and activating glucagon receptors, which directly increases fat oxidation and thermogenesis.

Well, if you can convince my wife of that, you're more than welcome to try. I just crossed 100k miles on my Model 3 and I think I'm on my 5th set of tires.

It's a combination of the weight of the vehicle and the torque that the electric motors produce.

Tailwind solves a somewhat different problem. It's maybe easiest to think of it as a bunch of aliases for writing inline styles. The fundamental problem with CSS at scale is leaky abstractions, right? Tailwind moves the abstractions down a level, so that rather than attaching semantic classes, you're just composing the properties of an element directly, except you don't have to remember if you used px or rem for this padding, or what the media query for a mobile breakpoint is. This tends to reduce the incidence of high-gravity classes which tend to accumulate features, at the cost of some verbosity vs something like Bootstrap, though it's significantly less verbose than equivalent inline styles would be.

@apply in CSS modules to compose the above sort of thing into something more semantic and readable without creating a bunch of global artifacts is definitely the way to go, IMO.

I've been writing CSS by hand since the late 90s. I resisted Tailwind for a long time, but I tried it on a pet project and I get it now. It doesn't fundamentally do anything that you can't do in CSS already, but it sort of "quantizes" the DX such that the vast majority of the CSS you'd usually end up writing boils down to a set of conventions.

You can still write your own CSS as needed (and probably should, for some of the more esoteric Tailwind cases), but stupid stuff like flexbox directives and padding/layout that you do literally everywhere become a lot easier, especially with things like the Tailwind VSCode plugin which provides autocomplete, reflection on defined variables, and linter errors for duplicated or incompatbile expressions.

And it still requires assembly, is sensitive to tolerances, and is functionally a plastic zip gun - an inferior variant of what you could easily produce with nothing more than a trip to your local hardware store. A better comparison is the FGC-9, which is specifically designed to be manufacturable without access to any regulated parts, and even then there's quite a bit of specialized materials sourcing and assembly that has to be done (including rifling your own barrel out of hydraulic tubing using electrochemical machining). The story that people are just rolling fully-functioning firearms off of their 3D printer is a tale told to frighten hoplophobes.

"Creating a weapon is as easy as loading a print file in your print queue"

Well that's just flat-out false. You can print things like frames, receivers, and misc hardware, but you still need machined metal components for parts that need to tolerate heat and pressure, and specialized hardware like springs of specific dimensions and tension which are really not easily home-manufacturable.

The author explicitly links to another article[1] on Tom's Hardware which specifically says this, so, uh, top notch journalism there, fellas.

[1] https://www.tomshardware.com/3d-printing/i-tried-3d-printing...

If you like the genre, the 2019 game "Control" treads in SCP space. It's great fun.

Giving a tool which is specifically good at finding relationships - no matter how strained or tenuous - between arbitrary concepts to paranoid schizophrenics seems like you might as well be lighting the fuse on an unstable bomb.

AI generations of people look different if you know the subject well, too.

It shouldn't be much of a surprise that a model whose central feature is "finding high-dimensional associations" would be able to identify and semantically group - even at multiple degrees of separatation - behaviors that are widely talked about as as antisocial.

How would this work in the case of, say, identical twins?

We've been there since the release of the Steam deck. Really, the only things I can't run on my Linux machine are the games that intentionally break the game with anticheat mechanisms. And it's not all anticheat - there are plenty of titles with anticheat that run just fine.

Check whatever you want to play at https://www.protondb.com/ - chances are, if it isn't intentionally borked with anticheat, it runs just fine. Looking at the top 300 games by Steam player count, 17 don't work, and probably 5 of those are utilities (like Crosshair X and Lossless Scaling).

If you have genuine reliable alpha, leverage exists. But you'd better make damn sure your edge is as good as you think it is.

Right, and I think that in this case the person making the original statement was probably erroneously using "ChatGPT" to mean "LLM" (in the same way that "Kleenex" is used to mean a disposable tissue).

It's true that ChatGPT-the-product can search the web (using a traditional search engine as a RAG datastore), and it's also true that o4-the-llm is not a go-out-and-find-external-truth search engine. I'm more sympathetic to the original point because many people do misunderstand what LLMs are and what they're capable of, and it's worth dispelling those myths (though it's valuable to do it precisely specifically to avoid this sort of confusion).

My household of 5 has been running exclusively Linux desktops for the past several years. Coming from Windows, I installed KDE distros and nobody has had any trouble with it.

They average user needs to be able to turn on the machine and have it boot, log in, use a web browser, connect Bluetooth devices, and print stuff. Linux desktops are more than capable of that sort of thing with zero additional training.

This weekend, I actually booted an old Windows XP machine that I've had sitting in a closet, and was astonished at just how...clunky Windows XP felt to me. It's not that it was hard, but it really helped highlight to me just how actually-usable Linux desktops have become.

It's been a long time since I've played WoW, but I don't really understand how permadeath would at all be viable. It's a game that's built around the expectation that the player is going to die, a lot.

I'd imagine you could exploit something like the stochastic denoising approach from DDIM and descendants, where you add some noise back after each denoising step, essentially randomly remasking unmasked tokens and giving the model a second chance to unmask them "properly" as the denoised response becomes better-known.

Another way to look at that time series is that US debt as a percentage of GDP has doubled from 62% to 121% since 2007.

https://fred.stlouisfed.org/series/GFDEGDQ188S