The criticism from the second camp stems from the fact that the WHOLE job is to not drop anything.
A fence with a hole is useless even if it's 99% intact.
A lot of human jobs, especially white collar, are about providing reassurance about the correctness of the results. A system that cannot provide that may be worse than useless since it creates noise, false sense of security and information load.
adding nuance to your fence analogy, most fences are decorative and/or suggestive of a border but can be overcome somewhat easily and are hence not useless because they have a hole somewhere.
The spirit of the analogy holds, as there are plenty of clear alternatives that map verbatim:
- drinking glass that is 99% hole-free
- car that doesn't explode 99% of the time
- bag of candy where 99% of the pieces are not poisonous
In all of these cases, it's more optimal to start from scratch and build something that you know is 100% reliable than to start with whatever already exists and try to fix it after-the-fact.
Personally, I use AI to assist development, especially in unfamiliar stacks, but in the form of a discussion rather than code-vomit. It's primarily synthesizing documentation into more-specific whole answers and providing options and suggestions.
This describes my travel flask when the screw cap is on and the slot in the cap is open; most of my drinking glasses have (and need) a much bigger hole-to-surface ratio to get the fluid in and for me to drink from.
More relevantly for the output of an AI: in cases where testing is easy, a system which has a 99% chance of a producing a saleable drinking glass and just discards the other 1% to recycling isn't unreasonable… provided you can be sufficiently confident about the test.
For AI, the quality of the automated tests of the output is a very solid "it depends on what you're doing".
I beg to differ. True, many fences are easily overcome by adult humans, but most fences are not designed with adult humans in mind. Most fences are intended to keep animals in or out of an area. In more urban areas that may extend to small children. In high security areas, fencing may be just one layer of security, but it is certainly more than just "suggestive". In any of these cases, the fence is useless if it has a gap that can be exploited.
We need to shift from thinking of solar/wind as "electricity sources" to thinking of them as "fuel sources". The marginal cost of producing a transferrable fuel from solar/wind is already lower than current electricity prices. The challenge is the capital expenses on equipment (hydrolyzers, fuel cell etc.)
Of course the article is just a puff public opinion piece. It doesn't change the economics:
Point ONE: nuclear is more expensive CURRENTLY that solar/wind. SOlar/Wind is currently the cheapest by far, and here's the scary thing for gas turbine: solar + storage is very close to druopping under gas combined cycle generation. Look up Lazard 2023 LCOE study. Solar/Wind have dropped to 24$ per MWHr. The cheapest nuclear produces is $141/MWHr. That's right, nuclear is almost 600% more expensive.
Point TWO: Say you miraculously got 100 billion dollars through congress for new plants to start immediate projects. Not a single reactor will come online for 10 years. Now, look at the cost improvement rate of solar, wind, and storage for the last 10 years. 10% year-on-year improvements or more. Even if we had half of that for the next 10 years (and there will be 10 more years of cost improvement), solar / wind very likely will be HALF the current cost.
So you'd have 100s of nuclear plants come online in 10 years generating electricity that costs 10-20x more than electricity from wind/solar. That simply is not a viable national power strategy.
If we had these plants already because construction started in the year 2000, that would probably be a different story.
The only hope for nuclear is probably in about 15-20 years where the cost improvement curves finally stabilize for wind/solar, and then a stable price point can be targeted with new nuclear designs. I personally think that only something like a novel MSR/LFTR which can scale down to mass producable sizes, uses all the fuel, breeds. doesn't have solid fuel rod reprocessing and waste transport/storage, and can use the Brayton cycle for more efficiency has a chance of competing with mature solar/wind.
Puff pieces like this are really about the current nuclear plants and keeping them on funding life support, which I generally support for now. The industry sees that Lazard LCOE curve just like any other person would: do you want to pay solar/wind costs for electricity, or 6x that for nuclear?
The existing nuclear industry can't survive without subsidies.
Your cost comparison does not factor in the grid storage required to convert unreliable solar+wind into 24/7 power. You can only directly compare $/MWHr when the proportion of unreliable energy is small, and offsetting reliable sources (i.e. when the cost of electricity is roughly constant). When all generation is unreliable, there's a huge question mark - how long can the grid weather dark, windless days? What's an appropriate safety margin, and what happens when you breach it?
From your own source, the Lazard 2023 LCOE study: "Most LDES (long duration energy storage) technologies have not yet reached commercialization due to technology immaturity and, with limited deployments, seemingly none of the emerging LDES technologies have achieved the track record for performance required to be fully bankable."
Ideally we'd have a combined approach - nuclear base load, renewables and grid scale storage to reduce the number of expensive nukes we need. That way you'll always have at least some power. But we don't have grid storage yet. And without storage, there's not much point in offsetting the output of nuclear plants, since fuel costs are so minimal - you might as well run all the nukes full tilt, all the time. So really, the only sensible thing to do is build primarily nuclear, and renewables in proportion to our ability to deal with wildly fluctuating energy supply.
Doesn't really make up for the fact that nuclear is 6x more expensive than solar. I really think synthfuels or all the hydrogen stuff going on (or the CO2 --> propane story also on the board today) will far outperform nuclear in grid leveling/storage.
Nuclear simply needs a reset to figure out how to make it economical. I love nuclear power in theory, there is simply so many more orders of magnitude of energy density and long life that it has to be viable somehow.
Again, I'm in favor of keeping the current nuclear plants operational. But build out a substantial new build of nuclear plants? Absolutely not. Research reactors, long term R&D / national laboratory projects on advanced nuclear? 100% support it, to the tune of a billion per year or more.
I would say that 24/7 power you can have using real technology that actually exists today makes up for any claims you might make about "6x more expensive". More expensive than what? Magic non-existent grid storage?
It's all very well saying you think that "synthfuels" and "hydrogen stuff" WILL outperform nuclear, but the practical upshot is that you are arguing in favor of sitting around twiddling our thumbs hoping some new technology will save us, when we could be building nuclear plants. What do you suggest we do now that will substantially eat into fossil consumption?
Go ahead and pretend grid storage is some pretend technology and isn't in actual production use.
That is honestly unhinged, and if nuclear proponents thing grid storage is some far off technology, they are more divorced from th economic reality than I thought.
Nuclear is 141$ per mwhr. Solar +storage is $45. Solar/wind are $24.
This pattern of improvement isn't some new phenomenon, and you can't pretend these are made up figures. Is storage pricing new? Yeah. Is the cost of battery backup going to drop? Virtually guaranteed with new sodium ion techs and the other stuff in the pipeline.
Attitudes like this frankly just validate that the entire current generation of the nuclear industry simply has a flawed and obsolete structure and outlook. They are simply ossified in a huge regulatory capture wall.
Your bet is based off of magical batteries from the future. This is not at all a coherent counterargument to nuclear. Nevermind the very concept of having redundant systems, or admitting the possibility of nuclear getting cheaper.
You can easily be accused of Silicon Valley style innovation delusion. As if something being older or more "regulated" means it is bad. It is the same delusion that got Uber to lose billions until they changed the same per ride as any taxi company. In reality, facts don't care about how you think the world works.
1. "Cost estimates" are more expensive than solar/wind. You know what they are also more expensive than? The historical costs of ACTUAL 50-year-old plants.
2. Average construction time for a nuclear plant is 6-8 years, with many built in half that time. [1] You can build over half a dozen generations of nuclear power plants before my 401(k) matures.
The final price does not depend only on the production price.
You cited Lazard 2023: if you read the firming intermittency price estimations, you figure out that solar and wind in the California grid costs 2/3 times more than when just considering the LCOE. Solar and wind production can be cheaper, but the final price is way higher, and it further increases when renewables take a larger share of the energy production.
How much of the current low cost is due to the currently low storage requirements? As the share of energy production from the unreliable renewables increases, the demand for storage will increase as well.
How do the costs compare when you factor in enough storage and overbuilding to reliably cover the inherently unpredicable generation downtimes?
I had the opposite experience: using taxis/Uber/Lyft was far more expensive than owning a car so after Uber stopped subsidizing rides I had to buy a vehicle.
That was in a city with a metro system, living in one of its main boroughs.
Even today, a cab to anywhere (even 3 miles) would cost me at least $10 each direction, and we have at least 5 such trips weekly. This easily surpasses the cost of the car, including depreciation and fuel.
More than $400/mo wouldn’t be surprising if the true cost is calculated over a year. Insurance, parts, maintenance, gas, putting the winter tires on, etc, etc, etc.
I own a car and I need one where I live. But I question that owning a car can cost much less than $4800 a year given insurance, maintenance, gas, fees, and other costs. As a purely financial decision, I'd trade for a driver if that were my real cost.
"prompt engineering" is a self-destructing field. If you use any rigorous approach to optimizing the prompt, you end up with essentially supervised machine -learning: models can (and do) learn the optimal prompt once there is a yardstick for the goodness of the model's response. That's a classical for a data-scientist, but the skill set has little to do with prompts.
If you are not rigorous, then what you are doing is essentially "black art". It may work for some tasks ad-hoc, but with the rapid pace of model improvement your skill will likely become irrelevant/not needed quickly.
Focusing on specific incantations, yes. Focusing on how to decomposing a problem, probably not, but then you get very close to designing systems / data design and analysis methodologies more than "prompt engineering", so I guess I mostly agree with you in as much as the relevant field is not really about AI as it is about picking up more structured design and analysis practices (again).
I don't have much experience with GPT but with image generation.
You need some amount of experimentation to get the best results but in my experience what works for one model does nothing or worsens the output in others. Adding loras and different types of images into the equation makes this so variable that I would never consider it useful besides keeping a few key words I used to get x good result on y model and experimenting with those when I start a new project.
The greatest technological advancement in recent years critically depends on the hardware from a single company with no competition. yet Nvidia stock is still below its 2021 peak. How so?
It doesn't necessarily depend on Nvidia hardware. Nothing stops you from training an AI on an adequately advanced ASIC or FPGA, in theory. Nvidia does accelerate it though, and they're also offering unparalleled performance-per-dollar to the audience that's in the market.
In a way, it feels like Nvidia is embarrassingly aware of this. They were the reluctant shovel salesman during the cryptocurrency gold rush, and they're rightfully wary of going all-in on AI. If I was an investor, I'd also be quantifying just how much of a "greatest technological advancement" modern machine learning really is.
It's the ecosystem - everyone else is using CUDA, so you need a very good incentive to stray away from that ecosystem. a x2-3 cost of hardware won't justify such move.
The cryptomarket was less favorable to Nvidia because it harmed the loyal customers (gamers, AI) for a temporary market (crypto) that indeed largely declined.
This narrative has been pushed for several years now with the likes of Habana, Cerebras, SambaNova, Graphcore, Tesla Dojo, etc.
And yet none of them seem to have made any dent in Nvidia’s dominance. None of them have any real presence on industry-standard MLPerf benchmarks (not even TPU releases all benchmarks and they started the damn benchmark).
The truth is that making an AI chip isn’t as simple as putting a bunch of matmuls together in a custom ASIC and pointing a driver at it; there’s hard work and optimization the entire stack down, many of which aren’t even focused on the math part.
So while I don’t doubt that some competitors (AMD?) will gain decent market share eventually, Nvidia’s probably not going to be displaced so easily.
Because making decisions on account of an asset's price being higher 2 years ago is just falling victim to price anchoring? Would Nvidia not be worth buying in 2020 because its price was much lower in 2018 and thus must be overvalued in 2020?
Investments should be based on the actual value of the company relative to its price, as well as relative to other investment oppertunities. Trying to making a profit by trading based on historical stock prices will get you whipped by quants who are already doing a much better job of that sort of thing than you could ever hope to do.
But the question isn't "can I do better than teams of quants who do this 100 hrs/wk and are supported by institutions with effectively infinity dollars", but "can I make money on this"? If I buy NVDA at 283, will it go up? There's no guarantee it will, they could lose their edge to AMD and the GPU market could bottom out, but barring some calamity, the answer seems to be yes they well. There maybe other stocks out there that are better buys, but they're part of the SP500 for a reason.
That's a broader question, but in general: it doesn't matter what I think about Nvidia's business. I could be correct all the way, but if other people disagree with me, they won't pay me for the shares.
It's also not necessarily about the 2021 peak but why isn't Nvidia bigger? allegedly it's a necessary component to a technology that can replace hundreds of millions of people (worth trillions in economic output). And unlike OpenAI, Nvidia wins no matter which company wins the model competition.
As far as stock prices, there was a hype cycle paired with government handouts to the people, these combined to push tech stocks to unreasonable valuations.
Why not? They seem to be a lot of leeway before any specific company will find it cheaper to design their own chips, or even to move to AMD (ROCm is not as well supported).
Perhaps someone like OpenAI has both the expertise and incentive to do so, but not many others.
So if you think that maybe OpenAI has other options if NVDA increases their prices why do you think that one of the other big names ( MSFT,GOOG,IBM,TSLA,AMD,INTC,Facebook ) also cannot do the same thing?
I guess for the same reason most of them keep buying from Intel - their market position allows them to pass on the cost to their customers, so it's not worth the distraction.
OpenAI is more of a "one-(very impressive)-trick-pony", so they have a stronger incentive.
A significant part of the 2021 peak may be explained by the crypto craze from which Nvidia benefited greatly and which has almost completely vanished since.
Thinking about it, it’s hard to believe how fast the hype cycle moved on from crypto. Only 1-2 years ago every media person, influencer, YouTuber, tweeter etc. were talking about/selling/shilling some kind of crypto, and now all of it seems to have moved on to AGI doomsaying.
Cryptocurrencies still had high barriers for entry for the public at large - not really a means of payment, and high risk as an investment.
Generative AI is used by millions, has very low barrier for entry (it's even free!) and most importantly does not require a network effect so can be valuable immediately.
Surprisingly, they left that out of their sales pitch.
With LLMs everyone+dog is coming out of the woodwork to let people know that it will lead to the extinction of the species.
Not that I don’t think generative AI is a lot more useful than crypto and deserves (some of) the hype. The problem is the hucksters jumping on the hypetrain to continue their $new_hotness grift.
1. Flexibility. For a rental car you need to plan ahead, pick it up on time and return it on time. Traffic jam delaying you? too bad. You now pay a surcharge.
2. Cost: the rental cost is actually only a small portion of the cost. Extra insurance and other fees end up more than the rent itself.
Your car will depreciate regardless of that road-trip, just by virtue of time passing. Rental companies won't subsidize the wear and tear for you, so you have to pay that anyway.
I think it is absolutely true that people who don't want to be early adopters shouldn't be. At the same time: I had no regrets about being a DVD early adopter, because (hot take incoming) DVDs were in fact better than VHS tapes, and not dealing with VHS's badness was really nice, even if it was also annoying in some ways. Similarly, there are going to be people who are happy being EV early adopters because the (very real) wins of EVs outweigh the early adopter annoyances.
"This technology is still relatively new, and if you buy into it, you're going to have some early adopter pains" is a fine message that will help people sort out what they personally want to do, but isn't the same as a blanket "wait until they improve."
It's a different risk: if your novel DVD player did not work, you could just watch something else. If your novel EV doesn't charge, you may be left stranded, miss a flight or fail to reach medical care. Moreover, no government planned to ban VCRs using those early problematic DVD players as a justification.
As an early adopter, the benefit of less maintenance, home "refueling", and the fact that an EV is more fun to drive offsets the hassle. We've done a few road trips in it, and while road trip charging is annoying (we have a VW ID4, so the CCS network) its still worth dealing with because we enjoy the EV.
Personally I think right now EVs are a great second car for almost everyone because you get the benefits of an EV but can fall back to your gasoline car for trips. We're still in the early adopter phase of "primary" car right now, where you need to be aware of the drawbacks.
That's the thing about cars: their real test (or importance) is not the 90% of the time spent in the "happy path" (short, planned drives after a full night charge). It's to address those edge but critical cases - long trips, unexpected drives, off-the-beaten-path routes.
Another thing rarely raised is the correlated nature of traffic issues. Charging is fine now, mostly because it seems there is a huge over-provisioning of chargers compared to EVs on the road. What happens with busy days? An evacuation order sending 1M people on the same route at the same time? For ICEV there is an easy solution. Not for EVs.
I'd say that it's more like 99%. The vast majority of my time with the car I'm driving around the Boston metro area, where it's an objectively superior experience. The three days a year I'm making a long road trip (0.8% of days), it's mildly more inconvenient.
In June we're driving to northern Maine. This involves a stop halfway through at a CCS charger for 30-40 minutes while we change the baby and eat. Sure it takes longer, but it isn't inconvenient enough.
An evacuation order is an interesting worse case scenario, but not one I'll likely ever encounter living in New England. Our storms tend to be "stay home" affairs instead of "get out".
Gas jerrycans. Either one kept at your garage for emergencies (cost: $5), or the gas station scaling horizontally by using/selling jerrycans.
Haven't seen any solution for instantly increasing charging throughput, and an on-line power is needed, while it is more likely to fail exactly in the kind of events that require mass evacuation.
Toyota/Honda ICE cars my extended family owns usually go many years without more than an oil change every 8-10k miles. Maintenance costs are absolutely trivial if you go Japanese. Heck even the Chevys made it to 200k when we sold them with no major mechanical issues. The costs of ICE maintenance seem to be the most overblown electric selling point. FWIW I dream about ways to rationalize buying a Tesla.
It's not so much cost as in I only bring the car in annually for a cabin air filter, tire rotation, wipers, inspection sticker, etc. Since most of the braking is regenerative the brake pads are essentially unused.
Reducing brake pad wear seems like premature optimization to me. Good pads can last well north of 50k miles, with some claiming 100k. They're also dirt cheap. I haven't changed pads in almost 5 years and my brakes don't squeak yet.
Having both, ICE and EV tech under the hood means more technology that can fail. Heavier than either, less EV range, possibly no subsidy in some jurisdictions, things like that.
I was referring to the reliability and weight claims.
And 72km is still higher than the average US trip distance, and it will kick over to petrol to augment any occasions when you go over that.
Its literally the best of both worlds - plug in charging convenience for the bulk of your driving and fast refuelling for the times when you need to do road trips.
Cities need to become denser to accommodate the growing population and reduce footprint. Efficient public transportation (trams/subway) are no longer regulatory or financially feasible enough to build (how is the second avenue line in Manhattan going?). Robotaxis and ad-hoc buses are a potential solution.
Autonomous cars are a localized regulation, so can be adopted gradually. The cities that will innovate will reap the benefits and those that will still dedicate their human resources to driving will lag.
> Efficient public transportation (trams/subway) are no longer regulatory or financially feasible enough to build (how is the second avenue line in Manhattan going?).
Uh are you sure about that? I don't think self driving cars are that more efficient than just regular human drivers. Giving up and accepting mediocrity (but fully automated!) doesn't seem like a very compelling argument
A fence with a hole is useless even if it's 99% intact.
A lot of human jobs, especially white collar, are about providing reassurance about the correctness of the results. A system that cannot provide that may be worse than useless since it creates noise, false sense of security and information load.