Hacker News new | past | comments | ask | show | jobs | submit | changoplatanero's comments login

Yes its a problem for the grid and the power companies don't allow large clusters to oscillate their power like this. The solution that AI have to do during their training big runs is to fill in the idle time on the GPUs with dummy operations to keep the power load constant. Having capacitors would be able to save on power usage.

Inb4 a startup is created to sell power load idle cycle compute time in AI training data centers.

Good news is that in this scenario it’s simulated suffering and not “real” suffering.

Is my LLM experiencing real or simulated suffering? How many layers of complexity and real-time stimuli-responsiveness would make the answer to that question effectively unknowable?

Suffering is bad because people experience the suffering. There is no fake suffering.

The joke is, they aren't real people. That is what the comment implied.

Of course, from the POV of any such simulation, the simulated suffering is real within the context of the simulation.

But realistically, it's already a giant leap - a thought experiment, mostly - to wonder if we're in a simulation. To then extrapolate our IRL experiences to some parallel In-Simulation-Life (ISL) value is pushing the thought experiment to ridiculous levels.


My point though is that simulated suffering, if simulated effectively, is as real as nonsimulated suffering. The important part of suffering is that it’s experienced. The inverse of fake people experiencing real suffering is that if you put a real person in a torture machine but you somehow prevent them from experience suffering they would not be suffering. So it does not matter whether or not the people are real.

I fully understand some people disagree as you’re saying. But I believe they aren’t being intellectually honest with themselves and are actually just trying to excuse the reality of simulated suffering.


all suffering is real if it is experienced by thinking beings

Same difference, though

Building their own models gives them more leverage for negotiating favorable terms with openai. The key to a good negotiation is having a strong BATNA, best alternative to a negotiated agreement.

I think OP poses a strong question.

If you destroy $500B in market cap in your share in said companies to get a better deal in a small segment of your core business...

That might not end up being a very bad deal overall.

The reality is, if MSFT wasn't delivering this SOTA SLMs now, someone else would be doing it within a few months.


Hmm I saw a different analysis recently indicating that there was a large reduction in fast food jobs in California after the raise to $20/hr. Which one is right?

The first studies that came out failed to adjust for seasonality.

“Here’s the problem with that figure: It’s derived from a government statistic that is not seasonally adjusted. That’s crucial when tracking jobs in seasonal industries, such as restaurants, because their business and consequently employment fluctuate in predictable patterns through the year. For this reason, economists vastly prefer seasonally adjusted figures when plotting out employment trend lines in those industries.”

https://www.latimes.com/business/story/2024-06-12/the-fast-f...


Share your other analysts so we can take a look.


Those tables seem a bit... odd? Are we led to believe that between January and February, no fast food restaurants closed? Not a single one? And same for July and August?

Also, a gold star for prominently featuring the totals cumulatively, oooo scary number go up! I did the needful:

https://i.imgur.com/OziGDLg.png


Something I'm curious to know: How does the efficiency of this new process compare to using regular solar panels to generate electricity and then using that electrical energy to synthesize the same chemicals?

Direct solar-to-chemical systems like this can be more efficient in theory because they cut out the middleman (electricity storage and conversion), but in practice, they're often less mature and have lower overall efficiency right now compared to established solar-electric-chemical setups

Agreed: this is the key question. One effort I follow in this direction is Terraform Industries[0] who are building exactly this type of system.

Their approach is PV + DCC (Direct Carbon Capture) and then simple carbohydrate synthesis, with the goal of establishing standalone autonomous systems that can generate valuable resources on their own in remote areas with ample sunlight.

They have a great blog where they go through their motivation for the approach from first principles [1].

[0]: https://terraformindustries.com/ [1]: https://terraformindustries.wordpress.com/home/


After following the literature down several different rabbit holes, I found this argument in some of the supplementary figures on that tree that seems to address your question:

> "Supplementary Note 1 | Advantages of PEC hydrocarbon synthesis.

"In general, PEC systems have the potential to combine the performance of wired PV-electrolysis (PV-E) systems with the simplicity of photocatalytic (PC) systems. PV-E is an established technology, which can take advantage of commercial solar cell modules with light harvesting efficiencies above 20% 24 and state-of-the-art gas diffusion electrolysers operating at high current densities above 1 A cm-2.25 However, PV-E assemblies require additional components including reactors, membranes, pumps, corrosive electrolytes, external cables and control electronics, increasing the overall system complexity and associated cost.26,27"

"On the other hand, PC powders provide an inexpensive alternative to PV-E, since light absorber particles and any necessary catalysts are dispersed in solution, which greatly minimises the overall system complexity. However, wide band gaps and charge recombination often limits solar-to-hydrogen conversion efficiencies to below 1%.28 While a homogeneous dispersion of the light absorber and catalyst can increase reactivity, this also poses challenges for the subsequent separation of all components and products from the reaction mixture."

"Accordingly, PEC artificial leaves provide a balance between PV-E and PC approaches in terms of complexity, cost and performance, by integrating state-of-the-art semiconductors and catalysts into a single compact panel. These PEC devices can perform reactions beyond water splitting (e.g., CO2 reduction to C1 products, or the light-driven C2 hydrocarbon and organic synthesis introduced here), while allowing product separation between the anodic and cathodic sides. This intrinsic design advantage is demonstrated by lightweight PEC systems using 15-fold less material than conventional solar panels, which combine the high performance of wired systems with the high activity per gram of photocatalyst nanoparticles.29 This applicability and potential of PEC-based fuel production also translates to hydrocarbon synthesis. In addition, direct light-driven hydrocarbon synthesis is carbon neutral, avoiding the energy-intensive Fischer-Tropsch process for indirect hydrocarbon synthesis from syngas (H2 + CO)."

Practically speaking the catalysts in these processes have relatively short lifetimes, so you'd want to incorporate an efficient catalyst regeneration process into the production pipeline, i.e. you might only get 16-128 hours of efficient production before catalyst regeneration is required so that needs to be built into any commercial process. So if you can design a catalyst that's easy to regenerate, that's very important.

Source material with nice pictures of the copper nanoflowers:

https://static-content.springer.com/esm/art%3A10.1038%2Fs419...


Modern solar panels are about 20-25% efficient. Putting it in a battery and using it to drive a car is about 80-90% efficient. Both the battery and motor lose some energy. If you multiply that, you get to about 16-22% efficient (starting with sunlight). And solar panels and batteries are still improving. Perovskite and other multi layer panel technologies provide a path to 35-40% panel efficiency.

Getting a similar efficiency generating some fuel that you then burn at 20% efficiency in a combustion engine results in a net efficiency of about 5%. That might improve on the fuel generation side but electricity generation would improve in a similar way and it would not be as efficient as that (thermodynamic laws and all that). It's basically not going to get much better than being between 4-8x less efficient than battery electric.

BEVs are winning on price and cost for that reason. Batteries are getting dirt cheap (50-60$/kwh). Solar and wind energy basically have no marginal cost. Driving 500K miles at 20 gallons/mile costs 75K$ at 3$/gallon for 25K gallons of fuel. 500K miles is a realistic life expectancy for modern battery electric drive trains. Good luck with that with an ICE car. Grid electricity isn't free but it won't cost you 75K. And honestly, you're going to be spending more than that on fuel in most parts of the world. And there's maintenance, parts, oil (engines use a lot of that too), etc. Bottom line: you could buy a new EV for 30-40K, and use the remaining savings for maintenance, tires, etc. All on the money that you aren't spending on fuel. Even a free ICE car would be a bad deal compared to that. You'd lose more money on just the fuel than you save on the car.

Now are efuels going to be cheaper or more expensive than regular fuel? It's a rhetorical question. We all know the answer (no way in hell). Hydrogen, bio fuels, efuels, etc. don't really stand any chance economically. None whatsoever. This is just greenwashing noise. None of that stuff is going to scale or matter. Some of the technology might matter for other purposes though. Hydrogen is super useful for lots of things and providing chemicals that we currently produce from oils synthetically could be valuable too.


> 20 gallons/mile

I did a double take here, I believe you mean miles per gallon?

> We all know the answer (no way in hell). Hydrogen, bio fuels, efuels, etc. don't really stand any chance economically. None whatsoever. This is just greenwashing noise. None of that stuff is going to scale or matter.

Disagree, you seem to only be considering cars. All of these things are just different forms of energy storage and they are useful if battery technology doesn’t have multiple orders of magnitude of improvement left in it.

There are numerous high impact use cases where you need more density, faster energy transfer, and completely different weight profiles than batteries.


obviously, nice catch

I'm also wondering how it compares to the efficiency of things like "grass" and "trees", which also convert sunlight into very useful things.

"Don't be snarky."

"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."

https://news.ycombinator.com/newsguidelines.html


Imagine you have 100 acres growing corn for biofuels, would it be nice to replace these by 99 acres of wilderness and 1 acre of photovoltaics producing the same amount of biofuels?

If your photovoltaics are 100x more efficient to produce your chemicals, agriculture is the dirty way of doing it.


We could more or less do that now. In fact we should probably just stop growing corn for biofuel, it's not obvious that it's even energy-positive, let alone a good use of farmland.

What do you mean by Biofuel? Like getting it from Biomass.

I don't understand, can't we get Biomass from like undesirable items like (not to gross anyone out, but feces?) whereas corn still has some value where you can actually eat it.


Does it mention it’s 100x more efficient anywhere? Or is it just an example you’re providing, in which case, why not 1000x?

They might be remembering the stat that:

> Looking at land-use efficiency, corn-derived ethanol used to power internal combustion engines requires about 85x (range: 63-197x) as much land to power the same number of transportation miles as solar PV powering electric vehicles.


What's that supposed to be relevant to? We don't make corn-derived fuel because it's cost-effective. It isn't. The idea is to give corn farmers something to do, which won't work if you reduce the amount of corn you're growing.

We make corn-derived fuel in order to power the Iowa caucus.

Corn farmers could be doing literally anything else, including a whole variety of things that rebuild soil or capture carbon or generate electricity, and it would be equally effective at powering the Iowa caucus, as long as we pay them to do it. They could even be producing crops organically, producing free-range livestock, or producing different lower-return higher-nutrition types of food, should we ever be interested in changing our diet a little. Deciding to produce the world's largest excess food supply in an industrialized fashion and then literally burning it was maybe a poor use of resources.


> why not 1000x?

now we're talking - can I invest in your company?


Corn isn't particularly great for producing ethanol. I'm guessing that a synthetic process won't be able to get close to 100x less land usage, but any improvement would be welcome.

The problem I see is that there's not enough money in in to develop a new process. Cellulosic ethanol outperforms corn on nearly every measure, but there's not enough money in it to pay for the development needed to scale it up to industrial levels.


Yes, and don't forget photovoltaics aren't limited to the crust - they can scale upwards, outwards, on top of oceanic deserts and arid lands.

Cover sunny, bleak northern africa in towering photovoltaics panels baby!


Labeling a massive geographic zone in an underdeveloped, historically exploited area—one which plays a key role in how the climate works—-as useless, and then converting it to an extractive industry…what could go wrong?

I’m as optimistic as the next person about energy tech, but I hope it doesn’t turn out like yet more colonialism.


Yeah, it would be terrible to offer the citizens of Chad, Mali, Tunisia, Libya, etc an opportunity to get revenue. Only Western democracies like Norway and Australia are allowed to extract substances!

> as useless

It's not useless, it's arid.

> which plays a key role in how the climate works

As does ever other square meter on Earth, and its natural abundant resource could be leveraged to even greater overall good.

> Extractive industry ... like yet more colonialism.

I'm not sure how we got here, or what your actual concern is.


I'm just picturing whole new swathes of rainforest being clearcut and bulldozed to make way for "artificial leaf farms."

Photosynthesis in nature is 1% efficient so it doesn’t need to be greatly better to beat it

The rubisco enzime is specially ineficient. While most enzimes can usually do thousands of reactions per second, rubisco does up to 10. Organisms compensate making loads of copies, to the point it's the most abundant enzime in nature.

What can a molecule do for such a long time? I mean they move very fast, the distances there is very short, so I kinda assumed that all the molecules do they do almost instantly. But 0.1 sec doesn't seem like an instant event.

Grass and trees are pretty bad at converting sun into glucose. Main enzyme in photosynthesis Rubisco is both slow (few molecules per min vs several hundred per second) and lowly selective (confusing O2 for CO2 regularly).

Which makes sense, for most of Earth's geological history CO2 was more abudant. So chance of mistaking O2 and CO2 was nil.

https://youtu.be/vYVSH2RpHcQ?feature=shared


Make sure to include the time and inputs to make the grass, and especially trees; those don't just appear out of nothing. And we already know how it works, it's called logging.

None of the inputs required for plants to grow require toxic pollution or destructive extraction.

Of course humans can bring in toxic or destructive inputs to try to favor certain plants over others, or humans can do other non destructive things to favor certain plants over others. Or humans can step aside and let the plants do their thing which will create abundance too. (I like the middle of these three.)

Also, trees provide far more value than timber alone.


To be a pedant, the inputs for photosynthesis are pretty toxic to humans. Sunlight burns and causes skin cancer. CO2 also kills people each winter when space heaters aren't properly vented.

Would you give up fertilizer and pest control and stop feeding the 8 billion ?

Please dont be a holdomorehippy.. Those back-to-nature loving massmurderers without a cause creep me out beyond repair.. those that openly hate some humans at least give the monstrous game away.


Eventually, you won't have a choice when fertiliser produced from oil runs out, becomes cost prohibitive, or is made illegal due to greenhouse gas problems; likewise, "pest control" has already resulted in a 40% decline in insect populations; it won't be good if it gets to 100%.

It would be best to find sustainable ways to grow food now, instead of continuing unsustainable ways (including supplying massive food aid to unsustainable populations so they can keep growing) until there is a precipitous crash.

The idea that only industrial scale farming can feed the planet is mostly a myth promoted by producers of industrial scale inputs and the oil/gas industry, by the way.


The production of concentrated nitrogen compounds from thin air is useful enough that we'll almost certainly keep doing it en masse in an electric-only future.

Mining for phosphorous and potassium fertilizers, likewise, but situationally a little different because these aren't very mobile in the groundwater column like nitrogen is, and they don't offgas back into the air like nitrogen compounds do. Quite possibly we'll be mining manure lagoons more, for CH4 and for closing the loop better on P and K.

Ag will continue at industrial scale for cereal grains, because half the population is not going back to the fields.

Within that framework, there's a lot of difference between outcomes in terms of how green we make our farms, what we grow, how we grow it. Herbicide and insecticide practices do not have to be what they are, as we witness massive overuse of things like neonics, glyphosate, and aminopyralid mostly because there's little financial reason to constrain use. We could stand to dramatically reduce the amount of cereal grain we consume, from a diet perspective, but the logistical difficulties of alternatives like more fresh fruits & vegetables will tend to increase carbon emissions. Eating less grain-fed meat and more high-protein legumes is basically a win-win from diet and climate perspectives. Returning to a less industrialized industry where livestock are raised on farms instead of on "feeding operations" seems like a fair tradeoff against something like subsidized corn-ethanol production. Attempting to encourage long-term soil stability with reduced tillage and is another goal that we might tangle with that would reduce yields; We have plenty of yield to spare in the US, so this is an option.


It is a given that bovines should be eating grass (one of the most productive plants there is with the highest calories per acre), not grain, and with the bonus that bovines or other ruminants eating grass improve the soil ecology and lessen erosion. There also isn’t any need for fertiliser inputs, or any oil/gas produced inputs at all.

Chickens can also be raised more sustainably. They don’t need to be raised 50,000 at a time, and don’t need to be fed grain. I don’t feed mine other than in winter when there is snow, and they don’t forage past an acre or so area. We produce a surplus of more chicken meat and eggs than my household can eat, and I still have enough time to work full time doing something else. (The same goes for my cows, but they take even less work and basically sustain themselves - I have not bought feed for them in two years.)

Oddly enough, I now sell my eggs for less than grocery stores charge for them. I could easily plant enough cereals and legumes for my household (about a 10,000 sq ft area or ¼ acre), but haven’t done this the last 2 years since I put my effort into vegetable gardens and livestock instead.

Part of the big myth is that we need industrial scale “farming”. We don’t. A lot more humans need to be a lot closer to producing the food they eat, though. If someone owns/maintains a lawn, they should be using it to grow food, instead of buying factory farmed food. (I give apartment dwellers a free pass, but I see large swathes of land that do nothing but grow grass and then have a lawn mower run across them.)


We absolutely need industrial scale farming, even if we take 40 million acres of suburban lawns and convert 10 million to active garden rows (there are many constraints on spacing, I'm dealing with this right now). Because 10 million acres feeds vegetable calories to something in the vicinity of 3-30 million people, at very high labor and logistics costs, for a small portion of the year.

It doesn't have to be 99% industrial scale farming in the current format, is the thing.


> Eventually, you won't have a choice when fertiliser produced from oil runs out, becomes cost prohibitive, or is made illegal due to greenhouse gas problems

You mean, aside from the process of making ammonia using green hydrogen that doesn't use fossil fuel at all? A process that can be sustained indefinitely, using renewable energy?

The single big concern is nitrous oxide emission from bacteria in the soil, but that can be reduced by nitrification inhibitors, some of which can be produced naturally by plant roots (and likely engineered into crop plants.)


Promises of “green hydrogen” and fertiliser made sustainably haven’t panned out. I’ll believe them when I see it, but I’m not a believer in industrial farming, which is more akin to mining.

Green hydrogen can't compete when natural gas-derived hydrogen is allowed to dump its waste CO2 into the atmosphere. That doesn't mean it can't or won't work when natural gas is outlawed. Your evidence shows nothing except that CO2 isn't being controlled.

Fertilizer is mainly made from natural gas, not oil. Accordingly it should last much longer. Worse case scenario when we run out is we switch to less efficient production, for instance splitting water using nuclear power.

Any plan that relies on depopulation isn't going to work and any attempt to force it to work would require crimes against humanity.


Whose time and what inputs are required to make grass and trees? If you simply leave a place alone, it will turn into either a forest, a grassland, or desert (the latter when human activity has thoroughly destroyed it).

I wouldn't replace grass or trees with this, but there's places where not much of anything is growing.

[flagged]


Warning: you may have become cynical.

I didn't read that comment as snarky at all - efficiency comparisons between emerging tech and SOTA (grass, trees) are extremely relevant!

(Warning to welf: you may be naive)


Efficiency is likely much lower than solar panels, however, solar panels are expensive and complicated (chemically) to manufacture. Teaching plans to make stuff for us is a better long term solution as we can just grow the plants.

What? Solar panels are cheap and little to no maintenance. Even though wildly inefficient, I opted to heat water using PV instead of a solar hot water because of how low complexity it is.

Also, nowhere in the article does it mention growing these artificial leaves, they probably need to be manufactured.


I roll my eyes at these "artificial leaf" claims for just the reason you've identified.

Solar panels have a limited lifespan, decrease in efficiency over time, and also get ruined when things like hail happens. This doesn't mean PVs are a bad idea, but it's not accurate to say they have little to no maintenance.

20 year solar panels just loose 5-10% capacity and degradation slows over time the reason most people replace them today is 20 year old panels were 200w where as today panels are 5-600w.

Sadly our solar panels don't self heal or come with 100(0) years warranty.

'Replace every 15-20 years' is not maintenance. Neither is replacement in the case of catastrophic weather events that'll have you replacing all your windows as well. The only 'maintenance' solar panels benefit from (which is still entirely optional) is occasional cleaning.

To be fair your windows are less likely to take the full brunt of an extra large hailstone since they're usually mounted vertically.

Aside from the hail, none of those are maintenance requirements. I've done 0 maintenance on my panels over their lifespan.

Sure but how do these artificial leaves fare when analyzed with the same criteria? Presumably worse, given that solar panels are (roughly speaking) nothing more than a few sheets of material laminated between glass panels.

Artificial leaf is an alternative term for extra complicated solar panel.


If you ignore the nice "Artistic depiction of an artificial tree" it looks like this will also be "few sheets of material laminated between glass panels", but I'm worry it will also need plumbing for the water and output gas.

And pollution down here on earth is way higher priority than light pollution in long exposure photographs of the stars


People have different priorities. Dismissing the needs of the scientific community based on meaningless comparisons for the financial benefit of a few megacorps isn't really a good argument. There are much less disruptive ways to provide good quality internet access for folks even in the most remote places. Considering the US especially, can you honestly say that the telecom companies did everything they could (or even a reasonable bit) to improve the situation, using all the subsidies and grants they took from the government in the name of improving the said connectivity?


Not sure what city you live in but in the big cities I’ve lived in it was always easy to get the level down to 500ppm by opening windows.


Yeah I live in Chicago and it is not hard to keep it at 450ppm. Right now it’s 493ppm in here.


What are you using to measure this?

(I'm sure I could do my own research, but was more curious since it seems like you've already got a setup.)


I mean that’s really low. If it’s cold outside you’re probable just wasting energy keeping your spaces warm with so much ventilation, making the problem worse in the process.


That was a voluntary pledge the company made to the users, right? It wasn't a legally binding commitment that there would never ever be any data sharing.


Dear Users, in our Terms of Service, we tell you that we won't share your data.

Psych, it wasn't legally binding.


Correct. A promise is not legally binding unless there is some sort of payment in return. The exception is if you can prove you suffered monetary damages from relying on that promise, which is basically impossible for data sharing.


> promise is not legally binding unless there is some sort of payment in return

If I recall correctly, I gave them a worldwide, perpetual license to some data.

> if you can prove you suffered monetary damages

This is a separate question (that of calculating damages) from that of whether there was a breach per se.


I can't sell my data to willing buyers for the same price anymore, because Meta illegally shared my data which reduced its value, and that's on top of the lost revenue I could have made selling my data to Meta if I was whatsapp only user.

Oh wait, I forgot those arguments only apply when companies are getting the government to go after people sharing files


Is it not at the very least false advertising?


It came with a threat that you'd lose your account of you didn't approve. That's hardly voluntary.


What I heard was that it was that the poor fellow tried to shoot himself but didn't get a direct hit. Because there was some time passed between the shooting and the death, they didn't find the gun and the body in the same place that you would expect.


You can do that but the amount of incremental data will be negligible compared to the rest of the data. Think of the knowledge cutoff more like a soft value.


Consider applying for YC's Summer 2025 batch! Applications are open till May 13

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: