This is a good point - having your code broken up into standalone units that can fit into working memory has real benefits to the coder. I think especially with the rise of coding agents (which, like it or not, are here to stay and are likely going to increase in use over time), sections of code that can fit in a context window cleanly will be much more amenable to manipulation by LLMs and require less human oversight to modify, which may be super useful for companies that want to move faster than the speed of human programming will allow.
Interesting, it seems that the actual surface material of walls and/or furniture makes a large difference in how long VOCs stick around, due to differences in surface area at the microscopic scale.
I have a couple HEPA filters in my house that hopefully keep particulate exposure down. Does this mean that I have to run them longer? That I need more of them continuously running to keep exposure to VOCs low?
As pointed out in another comment HEPA filters don't work well for VOCs (Volatile Organic Compounds), which are gaseous in nature. They're intended to filter particulate matter.
For VOCs you need activated charcoal/carbon filters usually and replace them from time to time.
The GP comment is talking about active ventilation though, through an ERV/HRV system. Also the article states this:
> The lifetime of these compounds indoors can be extended via partitioning to the surface reservoir as modulated by ACR. Higher ACR, which may be achieved by opening windows or through mechanical ventilation, leads to shorter t_half_surf because once indoor compounds partition from the surface reservoir to the gas phase as controlled by gas diffusion across the boundary layer, they would be removed from indoor air more quickly before repartitioning to the surface reservoir.
So they do state active ventilation can help, as you reduce the vapor pressure of VOCs allowing them to partition back into the gaseous env, where they can be promptly ejected. How much exactly is hard to ascertain from their graph since I don't have the exact data they used in the plots. But from squinting at it, it seems 1 OOM change in ACR gives you close to 1 OOM change in the VOC half life, which seems substantial to me.
So adding an active ventilation system might be a good idea for this particular concern. Of course it will add to your energy bill.
True, but simply using a low volume exhaust like a bathroom fan can give you a phenomenally greater effect than zero.
And that's for the entire house, zero is such a small number.
Then when you run it 24/7 it's 24 times as effective compared to a single hour. That's an impressive multiple itself, on top of bumping the baseline above zero to begin with.
This can really add up to a lot more ventilation than commonly assumed from some of the crummiest fans.
If you can't tell the difference when you walk in, between zero and running one of these all day before you get there, you're gonna need a bigger fan.
But you may be surprised and you never know until you try.
But when stuck inside the porous surfaces isn't the problem mostly when they become airborne again?
Most of us don't eat wooden furniture -- granted my toddler didn't get the memo :)
Thus, continuous ventilation (while not perfect) is hopefully still a decent alternative. Probably better than active charcoal filter.
Granted I should probably out a charcoal filter on the ventilation intake to reduce pollutants coming in from nearby traffic.
(All depending on your level of paranoia)
If the porous surfaces are saturated then you'll basically be maximizing the vapor pressure of these gases in the air you breathe. Check out my sibling comment, extrapolating just from the data in the article an active ventilation system should help.
EDIT: And yes, charcoal filters aren't as effective if they're not part of your critical airflow/ventilation path. :D
This kinda makes sense. Water vapor diffuses out through the building materials so why wouldn't VOCs diffuse into those materials?
What you're looking for are not HEPA filters but organic vapor filtering. If you were shopping for a respirator it would be easy but organic vapor extractors I think are a lot more expensive than HEPA filters. I looked in to it when I was doing a couple of oil based coatings for a home renovation project.
A lot of air purifiers are advertised as HEPA but really contain a filter stack consisting of a pre-filter, a HEPA filter and an activated carbon filter. Those would presumably help against VOCs, assuming you change the filter frequently enough
Compare those air ‘purifiers’ with the activated charcoal setups they use on cannabis grow operations, and you’ll get a sense of what volume of charcoal and air circulation is necessary to combat those small particulates. Purifiers help in theory but are nowhere near effective or active enough to combat off gassing or VOC dispersals in practice.
Frequent replacement is critical, my understanding is the activated carbon filters typically provided have very limited capacity. More so when compared to the lifetime of the hepa.
Thats why ecological buildings use lime and clay for plastering indoor walls. They can absorb a lot of things (water, fumes) and thereby regulate air quality and humidity.
The paper posits this is a problem. Large amounts of VOCs are absorbed by these complex structures. Then the structures with the embedded VOCs flake off and are absorbed by breathing, dermal contact and ingestion. Particularly by small children. This is literally their point.
I’d think you’d want the VOCs to be captured by something, rather than floating around in the air where you could breathe them in. Combined with a HEPA filter in the air circulation system, this should be a good solution.
Absorption is usually not a one-way street, though: Surfaces absorb gasses when the concentration in the air is higher than that on the surface boundary, but often also release them back into the air otherwise (which is why you can e.g. smell cigarette smoke in clothes – if they only captured it, there would be nothing for you to smell).
The only difference are some materials like charcoal, which does permanently bind many substances (but as a result can also saturate).
No idea which kind lime and clay are (i.e. "absorb and permanently bind with limited capacity" or "act as a buffer both ways").
> Combined with a HEPA filter in the air circulation system
Actually not that many types of things will bind to the carbon permanently, mostly it's the affinity for such a wide variety of contaminants to the carbon, combined with the porosity of the carbon structure which can have a very impressive amount of surface area to come in contact with the fluid being filtered. Whether filtering air or water. It hangs onto contaminants tightly.
Because carbon is such an effective adsorbent for contaminants, the partitioning coefficient for contaminants to remain in the solvent being filtered is lowered quite dramatically compared to so many other kinds of affordable alternative filtration media.
Most times people do need to afford to discard the carbon eventually, but it doesn't even really absorb contaminants like it's supposed to unless it is activated carbon to a good degree.
Activation only means that is it porous enough to begin with so it has enough surface area to be effective, then it is heated with adequate air exchange to about 250 Celsius for as many hours as it takes for virtually all of the VOC's or moisture it may have accumulated to be baked out. Then sealed up tightly, otherwise it can sit around for ages and gradually become saturated passively with any contaminants or humidity admitted through leaks to the ambient environment.
Sometimes, you can reactivate almost indefinitely to keep reusing the same carbon, and it works with VOCs because by their volatile nature they are basically baked back out easily and virtually completely each time. Different amounts of time if using different temperatures though, if equipped.
The stronger the activation, the more tightly with higher capacity the carbon wants to absorb things it encounters that are dissimilar to the fluid being filtered.
I assume they absorb VOC until you tear down the chalk or clay plaster.
With clay the indoor problem is more about radioactivity, but it's best in terms of humidity control. Chalk creates an alkaline environment on the surface which makes it inhabitable for mold (however the wooden furniture you put in front of it can still get mold if the indoor air humidity is too high).
Does that work if it's painted over? Or can you mix colorants in as with (exterior) stucco? (Maybe this is considered a kind of stucco? I just had to look it up: wikipedia says "The basic composition of stucco is lime, water, and sand".)
Nope, I dont think it works when painted over. Some vendors recommend colors which are very open for diffusion such as chalk colors, but every other "common" color based on acryl/latex/etc basically seals it from the air and destroys it over long term.
For clay I know you can add color pigments to the clay itself, most likely you can do the same with stucco for some limited amount of colors. But painting over it with modern products mostly destroys the diffusion properties.
Many people put plastics or other sealing products on top of a clay or lime-based wall and it's a shame.
I would assume if you paint it over with a latex based paint at least it would massively affect absorption. For oil based paints I have no idea though.
Polymarket gives a >50% chance of the tariffs being ruled illegal, not that they would be refunded - the market only gives a ~8% chance of the tariffs being ruled illegal AND and order to refund: https://polymarket.com/event/will-the-court-force-trump-to-r...
I think, all things being equal, higher home prices should lead to higher rents, since at the margin people on the verge of buying a home would be more likely to choose to keep renting when prices are higher, thus increasing demand for rental units.
Anyone can use coupons. Even if they don't want to spend the time to do it, they could. Same with store brand products made by the name brand manufacturer the choice is up to the consumer.
Uber's price discrimination is opaque. Even if they aren't doing dastardly things with it, people don't like feeling ripped off. We have no way of knowing when we are.
But most big box stores have moved to digital coupons that are indeed customized based on their creepy individualized spy dossier on you. At our grocery store, my partner and I get different coupons or even different deals for the same items.
LLM model training costs arise primarily from commodity costs (GPUs and other compute as well as electricity), not locally-provided services, so PPP is not the right statistic to use here. You should use nominal GDP for this instead. According to Wikipedia[0], the median country's nominal GDP (Cyprus) is more like $39B. Still much larger than training costs, but much lower than your PPP GDP number.
A top of the line consumer desktop, the Mac Pro, costs $7000. The commonly acknowledged first non-mechanical computer, the ENIAC, cost $400,000, which adjusted for inflation is $6,594,153 (see note). Will AI models follow the same pricing trajectory? Probably not but they no longer cost even close to $100 billion.
Note: 1946 CPI = 19.5, 2025 CPI = 321.465 which makes for an increase of 16.49.
It’s not enough to have the biggest model, or the best model per dollar spent, you still need to figure out how to make money with it. It’s not clear than vastly increased expenditure will produce a good ROI.
I think there is perhaps a different conclusion that I come to - email is not the right tool for long discussions with multiple points of disagreement, because it is, generally, a linear medium, which makes it difficult to maintain different threads without careful formatting by every author in the email chain.
I am not sure if there exists a good tool for threaded discussions with multiple different focus areas - something like git but for conversations?
I've written one, and people who use it, love it. People who look at it from the outside, hate it, hate on it, say hateful things about it, and generally make it clear that it should be set fire to, then buried.
But I and colleagues use it regularly, and it absolutely hits the spot for non-linear discussions that are intended to find conclusions.
Very interesting. From my read, it appears that the authors claim that this attack is successful because LLMs are trained (by RLHF) to reject malicious _inputs_:
> Existing large language models (LLMs) rely on shallow safety alignment to reject malicious inputs
which allows them to defeat alignment by first providing an input with semantically opposite tokens for specific tokens that get noticed as harmful by the LLM, and then providing the actual desired input, which seems to bypass the RLHF.
What I don't understand is why _input_ is so important for RLHF - wouldn't the actual output be what you want to train against to prevent undesirable behavior?
Are products allowed to have labels? Am I allowed to tell my friends I like a product? What if I put a video on youtube and accidentally include a brand name in it?
Last time I checked, a product label (on the product or on the package) is not an advertisement. It's just the name of the product and/or brand, and maybe some lines about what it does. Even if you call a product label "a sort of an advertisment" it's fine.
When people complain about advertising today, do they refer to product labels? Or to their friends telling them about a product? If not, why are you bringing this up?
>Am I allowed to tell my friends I like a product? What if I put a video on youtube and accidentally include a brand name in it?
Sure, as long as you aren't getting paid for doing it (directly or via affiliate kickbacks). If you are, and you're discovered, you pay a fine - or go to jail.
You try to paint a "it's impossible" all or nothing scenario around marginal advertising and edge cases. Doesn't matter. If we can get rid of 90% of overt advertising - tv ads, streaming ads, posters, billboads, radio jingles, that's enough, even if "you put a video on youtube and accidentally include a brand name in it".
