Hacker Newsnew | past | comments | ask | show | jobs | submit | timmg's commentslogin

> Researchers at Amazon had used a series of prompts to get Anthropic’s Fable 5 model to provide them with information that could be used to aid cyberattacks...

All models can do that. I wonder if they found Fable was significantly better at it.


All models almost certainly can’t do that.

Most of the agentic coding capable ones can. Even very small ones like Qwen 3.6 or Gemma 4 are surprisingly good at it.

And if LLMs can do it, then the information is freely available out there with a single google search. Time would be better spent on making infrastructure more resilient.

That doesn't follow. I can't google search, "are there any vulnerabilities in auth.go in this directory?" I can ask an LLM. And, if they find something, I can review it, and fix it, thus making infrastructure more resilient.

Maybe the model found something Amazon didn't want to be known, and not necessarily a cyber vulnerability, but a particular way Amazon operates.

Down AWS East and make the traffic look like it’s coming from the Vatican.

"find aws zero day. makes no mistakes"

xAI or OpenAI? (Or both?)

I guess it will be interesting if, in a week or two, OpenAI launches a "Fable class" model and it isn't blocked by the government.

There are a lot of things like this.

My favorite is how elegant solutions often look simple in retrospect. So if you noodle on a problem for a while and then come up with a clever solution: once you explain it to someone they'll be like, "yeah, of course."

Meanwhile the guy next to you that overcomplicates the problem ends up getting kudos for building something so difficult :D


"Je n'ai fait celle-ci plus longue que parce que je n'ai pas eu le loisir de la faire plus courte."

("I have made this longer than usual, only because I have not had the time to make it shorter.")

Blaise Pascal


"Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away."

— Antoine de Saint-Exupéry


I read this in Leonard Nimoy's voice, from narrating Civilization IV.

"A facility for quotation covers the absence of original thought."

― Dorothy Sayers


I am reminded of this course in university where there was a written assignment with a minimum page count. Even at that time, I remember thinking: "If I am able to express everything necessary in just one page, that should give me the absolute top grade!".

I feel like AI coding is accelerating everyone's work toward greater solution complexity and I think it's pushing people to build defenses and be more averse to someone else's complexity rather than being impressed by it. Bigco's are probably well behind the curve on this and are still impressed by complexity, but for people on the receiving end of AI stuff either directly via your own hand or indirectly via others, it seems like complexity is not as impressive as it once was.

Definitely. Previously, intent and effort were required to increase complexity. Now intent and effort are required to prevent complexity*!

But also, what a beautiful problem to have!


There was a thread recently on HN about Claude Shannon and how his papers were filled with clear descriptions and explanations. Then someone commented how they had found an elegant solution to a problem that either could be described shortly and beautifully so that a high schooler could understand or take the long and tedious path of a convoluted explanation.

The director then clearly advised that they should use the complicated way because that's how you get published: not because you're clever, but because your solutions sound complicated.

It resonates perfectly with your comment and it's an unfortunate reality that most people don't bother for beautiful solutions and praise complicated processes. That's how we neded up with bureaucracy, probably :D


I have said it for decades - Basic is easy, Simple is hard.

But when someone comes up with something simple but effective, it always looks so obvious in retrospect.


On the other hand, I don't help people with their computer problems anymore because I've found that the more difficult the problem, the longer it takes me to rescue their data or whatever the less impressed they are. The more miraculous the save the more likely they are to tell me the story about their nephew who solved a trivial issue instantly as if to point out that I didn't.

Things I won't work with: printers.

The passage that comes to mind for me whenever this idea comes up, from the Brett version of the Holmes story "The Dancing Men":

  H: So, Watson.
  W: Hmm.
  H: You do not propose to invest in South African securities?
  W: How on earth do you know that?
  H: Now, confess, you are utterly taken aback.
  W: I am!
  H: I should make you sign a paper to that effect.
  W: Why?
  H: Because in a few minutes you will say it is all so absurdly simple.
  W: I should say nothing of the kind!
  H: You see, my dear Watson, it is not really difficult to construct a series of inferences, each dependent upon its predecessor and each simple in itself. If, after doing so, one simply knocks out the central inferences and presents one's audience with the starting point and the conclusion, one may produce a startling, though possibly a meretricious, effect.
  H: I can tell by an inspection of the groove between your left forefinger and thumb, that you have decided not to invest your small capital in the gold fields.
  W: I can see no connection.
  H: Very likely not; but I can quickly give you a close connection.
  H: Here are the missing links in the very simple chain: You had chalk between your forefinger and thumb when you returned from the club last night. You put chalk there when you play billiards, to ease the cue. You never play billiards except with Thurston. Now, Thurston, you told me, four weeks ago, had an option on some South African security which expired in a month, and which he desired you to share with him. Your checkbook is locked in my drawer, and you have not asked for the key. So, you do not propose to invest your money in that manner.
  W: How absurdly simple!
  H: Quite so. Every problem is absurdly simple when it is explained to you.


Meta: I found this video essay (?) "Jeremy Brett vs Basil Rathbone — Who Was the Real Sherlock Holmes?" interesting:

* https://www.youtube.com/watch?v=WaQFJcI_yfI


Real Sherlock was Vasily Livanov of course.

Liking it, but I think it's even better captured by the more lauded quote -

H: "How often have I said to you that when you have eliminated the impossible, whatever remains, however improbable, must be the truth?"


Why on earth would you quote a TV version of a book, when the book is readily available to be cited?

“So, Watson,” said he, suddenly, “you do not propose to invest in South African securities?”

I gave a start of astonishment. Accustomed as I was to Holmes's curious faculties, this sudden intrusion into my most intimate thoughts was utterly inexplicable.

“How on earth do you know that?” I asked.

He wheeled round upon his stool, with a steaming test-tube in his hand and a gleam of amusement in his deep-set eyes.

“Now, Watson, confess yourself utterly taken aback,” said he.

“I am.”

“I ought to make you sign a paper to that effect.”

“Why?”

“Because in five minutes you will say that it is all so absurdly simple.”

“I am sure that I shall say nothing of the kind.”

“You see, my dear Watson”—he propped his test-tube in the rack and began to lecture with the air of a professor addressing his class—“it is not really difficult to construct a series of inferences, each dependent upon its predecessor and each simple in itself. If, after doing so, one simply knocks out all the central inferences and presents one's audience with the starting-point and the conclusion, one may produce a startling, though possibly a meretricious, effect. Now, it was not really difficult, by an inspection of the groove between your left forefinger and thumb, to feel sure that you did NOT propose to invest your small capital in the goldfields.”

“I see no connection.”

“Very likely not; but I can quickly show you a close connection. Here are the missing links of the very simple chain: 1. You had chalk between your left finger and thumb when you returned from the club last night. 2. You put chalk there when you play billiards to steady the cue. 3. You never play billiards except with Thurston. 4. You told me four weeks ago that Thurston had an option on some South African property which would expire in a month, and which he desired you to share with him. 5. Your cheque-book is locked in my drawer, and you have not asked for the key. 6. You do not propose to invest your money in this manner.”

“How absurdly simple!” I cried.

“Quite so!” said he, a little nettled. “Every problem becomes very childish when once it is explained to you. […]”

The Adventure of the Dancing Men, The Strand Magazine, Vol. 27, January 1904, The Return of Sherlock Holmes, by Arthur Conan Doyle <https://www.gutenberg.org/cache/epub/108/pg108-images.html#c...>


Why on earth? Because I love that TV series, and the passage from that TV series is what comes to mind for me.

Yeap. Managers perceive complexity by how personally confused they are. I'm late in my career, and I'm realizing I wasted so many man years trying to make code clean, user friendly, and maintainable, when that code was never read by another person again and forgotten 15 minutes after it was released, then used for years. This is why I think AI is coming for our jobs much sooner than many people think: clean code, separation of concerns, maintainability, etc, all the things we spend the most time on, have never actually been valued. "Good enough" is fine, and keeps management happy. And, if something does pop up, AI can patch it, even if with spaghetti, just like fucking that ass at work.

Every company where I've worked at as a SWE openly rewarded "engineering complexity" as a criteria for getting promoted, which I've always found to be absurd because complexity can always be manufactured (both of the problem and the solution).

One of those days, however, you come up with another of your elegant, simple solutions and it actually replaces a 150K LoC monstrosity with either a 1K script or, even better, with nothing as a simple shift in perspective or process obsoletes it completely.

In the long run, IME, you'll be recognized either by management or your peers if you keep doing that over and over again.


  > elegant solutions
My favorite is how people will yell at you about how elegance doesn't matter, that they "just care that it works", and "keep it simple". I'm certain all the sayings repeated in industry are metastasized variants of actually good practices repeated by those who can't be bothered to understand what they mean.

And of course that's true. We push for speed, absent of direction, while praising velocity. To be honest, at this point I'm disappointed the engineers gave up and just started becoming business people.


"An idiot admires complexity, a genius admires simplicity"

In academia that translates to, the more senior the faculty member, the easier the talk is to understand.

"elegance and speed go hand in hand" - d. mcilroy

> There has been plenty of research that shows LLMs encode social biases.

At the risk of stepping into a hornets nest: is that different than "knowledge"?

Or maybe, what would it mean if an LLM had no social biases? (Would we ever agree that was the case?)


Yes, it would be extremely bad if the statistical weight of the total corpus of training data caused a system using an LLM to make decisions about extending credit to offer worse terms (say) to women.

> sing an LLM to make decisions about extending credit to offer worse terms (say) to women.

In general, or if it isn't the correct answer?

Like: young men pay more for car insurance than young women (today). This is based on statistical models. Should they be outlawed? I think that is a very interesting question (but they aren't, today).

If the LLM was in charge, would it be wrong for it to charge young men more? Should we train that "bias" out? Or should we only train out biases that are wrong? And would that be different than how we train them today?

I don't know the answer. But I think it is less obvious than some people seem to think.


young men pay more for car insurance than young women (today). This is based on statistical models. Should they be outlawed?

EU has outlawed them. their argument is that differentiation is only valid if the difference is the actual cause and not merely statistical correlation.


Ironically, in the US it is ok to charge men more for car insurance, since they cost more in aggregate. It is illegal to charge women more for health insurance even though they cost more in aggregate.

given the economic realities of income between men and women, i think that makes sense.

It would obviously be very bad if those decisions were being made based on the statistical weight of the training corpus of a general large language model.

That just shows how biased you yourself are. Every human is. It is FAR more likely that the algorithm would give better credit terms to women and worse terms to men, as it is already the case with insurance. Yet you assume the opposite because of your personal biases.

At least LLMs offer a way to be tuned against that. Not that their creators would be interested in that, since the LLM's bias is exactly the mainstream opinion that they like very much.


I wasn’t assuming anything. I was asking whether the problem was bias — which we already see in some things that are highly regulated — or just wrong bias.

I’m trying to understand what people think we should correct for.


Correct. They will never not have a social bias. Which leads to the question of, who controls these tools, and what biases are they okay/not okay with specifically training for. Currently they can be seen more as a reflection of broader culture (and even that has problems) but as we're already seeing with Grok they can be tuned at a whim to display any specific ideologies.

Those are some of the questions it leads to, but there are other questions that situate agency outside of the labs and in the hands of users, like, what processes do you have set up to backstop automated decisionmaking?

It's not interesting to observe that Grok was successfully trained to be an edgelord; anybody paying attention knew that was easily achievable.


> what processes do you have set up to backstop automated decisionmaking?

The companies releasing these models actively encourage the act of automated decision making by them. The entire value proposition is the automation of decisions and knowledge work. It's rare to find a use case for them that isn't offboarding your thinking and therefore agency


The entire value proposition of the computer industry is the automation of decisions and knowledge work. We are and always have been in the business of automating away people's jobs.

I reckon we agree more than we disagree, but there is a dichotomy of expansive and contractive technologies. Much of the computer industry has given more agency, choice, and knowledge to people.

That's not in tension with the fact that computers have displaced enormous numbers of jobs. The pitch has always been that the displacement is accompanied by new opportunities elsewhere in the economy.

The way I've been thinking about it: there is too much money trying to pour into the market. That's why valuations are so high.

Maybe getting more of these big private companies public will bring valuations down a bit.

(Just my impression. No math or financial studies behind it :)


> there is too much money trying to pour into the market

Keep in mind that inflation ran over 7% annualized in April [1].

[1] https://www.bls.gov/news.release/cpi.nr0.htm


From that doc, prices went up 0.6% in one month, multiple by 12 get 7.2% annual inflation rate.


The vast majority of that was fuel.


> vast majority of that was fuel

Everything else is up around 3% YoY. And if energy and transportation are up double digits, and producer prices are up double digits, other consumer prices will follow.


Yea and the cost of fuel has zero downstream effects on the economy.


Inflation is a measure of the cost of living. It's not got loads to do with large-scale, institutional investments.


> Inflation is a measure of the cost of living

The faster your cash loses value, the stronger your incentive to trade it for something else. That something else can be financial assets.

> It's not got loads to do with large-scale, institutional investments

For investors, particularly retail investors, the consumer price index is most relevant. But for whatever it's worth, producer prices are up over 16% in April (7% excluding "foods, energy, and trade services," which jumped over 50% annualized) [1].

To be clear, I'm floating a hypothesis here. I have seen no evidence linking inflation to demand for these companies' shares. (If anything, it should be the inverse.)

[1] https://www.bls.gov/news.release/ppi.nr0.htm


That depends. Inflation is a measure of the cost of living in terms of currency. It can be high either if goods and services required for living become scarce, or if currency supply increases. Currency supply increasing does affect asset prices.


Yes, but they're not directly correlated. Of course events can affect them both! Going to war would both increase the cost of living and (some) asset prices would go way up. But that doesn't mean they should be measured together like that.

Inflation then is already higher. Cost of living is driven mostly by rent


Corporations across the board are experiencing record profitability. That's the reason behind the high valuations.

This isn't true of AI companies...yet. But these are companies entering the market with pre-IPO userbase (including lots of B2B) numbers that Meta and YouTube would have dreamed of before their acquisition/IPO.

I think this whole situation is very sleazy and corrupt, but ultimately my prediction is that nothing serious will come of it. Even the exposure of index and passive investing is overstated.



No, the crash (that we all know is coming) will do that. Until then, history teaches that we'll just keep going up and up


This is one of those "everyone who dies, breaths air" statements.

It's frustrating people who parrot it think they're smart by saying it to others with no basis and finally when it does happen they're like SEE SEE!?

> Until then, history teaches that we'll just keep going up and up

And this is the more important part. As long as you're <40 you SHOULD always buy SPY or VOO, even at the very top.

People have been saying the crash has been coming since 2022. If you believed this and acted on it, you would've missed 3-4 +10%/yr returns.

As Buffet says: You can't time the market; be in it.


It doesn’t seem Berkshire is that much in the market right now.


Just to add some color using real numbers: Berkshire's Q1 cash pile was $397.4 billion, which is nearly 60% of its investable assets.


One of my favorite phrases is “the market can stay irrational longer than you can stay solvent.”

Even if all signs point to impending doom, at the end of the day if people are still buying, stocks will hold their value.


>And this is the more important part. As long as you're <40 you SHOULD always buy SPY or VOO, even at the very top.

But why? The US population is set to dramatically shrink in the next 30 years. Where does all the money come from?


The US population is projected to 385M in 2055. [0] What makes you think the population will dramatically shrink?

Even if it did, the money would come from global investors who find the US more attractive than their home markets.

[0] https://database.earth/population/by-country/2055


The birthrate across all western nations has been below replacement levels for quite a while. All nations will see population shrink in the coming decades (aside from labor importation)

https://pmc.ncbi.nlm.nih.gov/articles/PMC11537490/


Right now SPY not be such a great idea with SpaceX launch upcoming since it will be included into it immediately. Retail investors will be bearing that particular flop's cost.


> SPY not be such a great idea with SpaceX launch upcoming since it will be included into it immediately

S&P has not announced a methodology change yet.


They haven't approved a change, but they have released proposals, which are tentatively to go into effect on June 8:

https://www.spglobal.com/spdji/en/documents/indexnews/announ...


no one is going to get wealthy buying SPY/VOO. you might get rich, but not wealthy. things have changed now in a sense that handful of companies are large percentage of the stock market to the point where one has to question why invest in “s&p 500” vs “s&p 25-ish”

while going with the tried&true makes some sense, I think we have to open our eyes to a different reality of our stock market… and this market concentration into few companies is going to get a lot worse…


> things have changed now in a sense that handful of companies are large percentage of the stock market to the point where one has to question why invest in “s&p 500” vs “s&p 25-ish”

A small number of companies have always driven most stock-market gains. Betting on size isn't fundamentally a bad bet. But it is a bet against value and the historical tendency for small companies to be higher risk and higher reward.


you may be technically correct but today’s concentration in say top 10-15 companies is historic and by significant margin. I have been self-employed for a long time and somewhat “forced” into being “an investor” and starting in 2021-2022-ish I took my money out of all the “funds” … while I do not disagree that it is “a bet” - it is a calculated bet. things are different now even if historically you are right, no question


> you might get rich, but not wealthy

What do you mean by this? It seems pretty likely to be wealthy by investing in these indices. Certainly a “normal” worker who started investing $10k/year in SPY when it started in 1993 has enough wealth to allow them to retire now.


I was old enough to remember the 08 crash. Then the market starting recovering in 2011/2012 and the sentiment was that the system would crash again soon like 08. Turns out, it was an amazing time to invest.

Post 08 crash, all sorts of conspiracy websites like Zero Hedge were popular saying how the world economy would keep crashing.


The only reason this happened was due to taxpayers bailing out financial institutions. This only exacerbated an insane amount of moral hazard already present in the market following previous bailouts.

Unfortunately, the US Government continued to run themselves into the ground spending-wise and may have a difficult time with another bailout, unless everyone pretty much agrees that we cannot have a USG failure, so they all pretend like nothing happened.

Eventually the merry-go-round stops, I'm just not sure what the catalyst will be, and it might be 100 years from now.


I am old enough to have had multiple career changes since starting on a major firm’s rates floor in 2008. These IPOs are tiny compared to the overall stock market and the stock market is absolutely tiny compared to debt markets. People consistently underestimate the size of the world economy or even their local economy. The world may look small from an orion capsule near the moon but almost every aspect of human society is bigger than most people can reason about. It is possible these IPOs have an outsized impact on sentiment for weird reasons. But it won’t be an actual outsized impact on capital markets.

Edit: I should add the AI bubble can absolutely burst but there is no reason to believe these IPOs are the end of the ride. If I knew I would be…


The thing is that the IPOs necessitate a full release of their actual costs for inference and training. This by itself should be enough to pop the bubble, if the occasional bits of it we get are anything to go by.

There is a reason anthropic is still hiding those details:

> key details typically included in that form about a company’s operations — like potential risks to its business, executive compensation, and other financials — won’t become public until later on in the process

Source: https://www.theverge.com/ai-artificial-intelligence/941016/a...

We'll see, maybe they trigger some new rule change to be allowed to keep it hidden. Wouldn't be surprised about that at all.


Without massive government intervention it probably would have


It would crash if not for massive public debt that went mostly into capital markets and huge inflation.

I very much disagree that it's coming. I think we need to completely reset our expectations of how the market works. There's been nearly an entire generation working in this "new" bull market, where things like EPS mean absolutely nothing and speculation no longer requires actual returns.


>I think we need to completely reset our expectations of how the market works.

Is this not just "It's different this time" thinking? I remember it being used all the time during the dotcom boom


> There's been nearly an entire generation working in this "new" bull market

You mean 0DTE babies?


It’s unfortunate, but this exact sentiment precludes every market crash. Just before dotcom burst it was the same thing. Technology will change everything the nasdaq will continue to grow in the technology age, etc etc. eventually the chickens come home to roost as they say


It did though. Look at it now. The crash was incredibly short lived, and if you try to point it out on a graph of the last 30 years, you won't be able to find it.

> the crash (that we all know is coming) will do that. Until then, history teaches that we'll just keep going up and up

Stock prices don't have to crash. They can just stagnate while profits catch up and multiples compress.

Debt binges, on the other hand, tend to go bust with a bang. But after the recent private-credit scare, the AI build-out has been predominantly financed with stock. (I think.)


Hasn't there been a _lot_ of debt to buy up Nvidia GPUs? I follow this stuff somewhat closely and it feels intentionally confusing, so I've likely lost track.


> Hasn't there been a _lot_ of debt to buy up Nvidia GPUs?

I believe that's been concentrated at the hyperscaler layer, and subsided when the aforementioned private-credit scare reared its head. (I haven't heard a big datacenter debt deal announced in a while. Though of course that doesn't mean they aren't being done.)


And we're still extremely compute constrained. We need more Nvidia GPUs, RAM, power.


> Equity bubbles don't have to crash. Prices can just stagnate while profits catch up and multiples compress.

Is there is historical evidence for that? As someone who used to follow Jeremy Grantham a lot (he considered himself a "bubble historian"), IIRC every bubble he studied always mean reverted, and it usually (maybe always, can't remember) overshot on the downside during the correction.


> IIRC every bubble he studied always mean reverted

This really depends on how we're defining these things. Let's call a stock-market bubble a period of elevated multiples. That can mean revert by prices decreasing while earnings stay constant or by prices staying constant and earnings rising. (Alternatively, both earnings and multiples can rise and fall.)


Yes, for equity prices in particular he talks about P/E ratios (among some other metrics like corporate profit margins), and so you're right, it would be possible for this to mean revert by prices holding stagnant and earnings catching up. However, as far as I can remember (primarily because a big emphasis of his was how unchecked bubbles can cause a lot of damage on the downside) all the historical bubbles he studied (something like 50) always crashed with a big price drop. Not 100% sure though, which is why I was curious if you had any contrary examples.


There is nowhere else for that money to go


My first ever programming interview was like a group interview. There were three or four programmers asking me questions, one at a time.

The only one I remember was to check if two strings were equal (in C). I wrote (maybe buggy) code to iterate both pointers, comparing while looking for the null terminator.

The interviewer stopped me and said, “You should compare their lengths first. If they are different, you can exit early.”

I was pretty young and didn’t know much, but I explained, “But you have to look for the terminator to find length so it’ll take twice as long.”

He snapped, “There are optimized functions for that!”

I assumed he was right. Needless to say, I didn’t get the job.

Maaaany years later, I realized the std library was probably open source. So I checked (one). It was nice to be vindicated :D


On the upside, think how annoying it would have been to be part of that team and have a boss that doesn't have any ability to admit he is wrong.


Yep, sounds like a bullet dodged there.


Funny how often "there are optimized functions for that" really means "I haven't thought through what the optimized function actually has to do"


C is pretty bizarre but I expect someone writing it professionally to know that even passing void compareStrings(char str1[], char str2[]) is equivalent to compareStrings(char * str1, char * str2) so no way to get the length of it with sizeof(str1) and strlen walks the string until it finds the null terminator.


No, it's not bizarre, it's standardized on what is and isn't UB based on history, and usually for performance.

NUL-terminated strings don't know their lengths and so, without an "n" variant function and running strlen() ahead-of-time, must iterate the entire thing. Pascal format strings (supported up to 255 byte lengths in the classic form) could find length as an O(1) operation because there was no terminator necessarily.


It is bizarre. C is bizarre, there is no string type and there should've been a string type. scanf is an invitation for buffer overflow attacks. I know how poorly designed languages look like and C despite its strengths is still a poorly designed language.

"Group interviews" produce horrible, unfair dynamics.

Snapping with know-it-all arrogance is toxic and doesn't help anyone. You were correct because strcmp has to iterate both strings, just like strlen would.. and it's totally pointless.

    #include <string.h>
    /* C89 and handles NULL and overlapping strings */
    int strequal(const char *a, const char *b) {
            return (a == b || (a != NULL && b != NULL && strcmp(a, b) == 0));
    }
I think you dodged several bullets by not getting hired there because they sounded insufferable.

> Pmf is this weirdly defined thing where "if you're not sure you have it then you don't".

I'm not sure if this runs counter to your point or not, but: I don't see any future where LLMs aren't a core part of Software Engineering. The horse is out of the barn. There is no going back.


Yeah but the product is not “LLM” it’s “proprietary frontier model LLM paid by the token”.

And I don’t even necessarily disagree with OP! It’s more like the competition is shifting so quickly that your competitors could undercut your PMF in a blink of an eye.


There will be cheaper solutions. And they will generally be less capable than the more expensive ones. Just like most other products.

But my guess is that the cost of SWEs themselves mean that the more expensive ones will be worth the delta to most companies.

But time will tell.


History bears out that cheap and satisficing soundly beats expensive and optimal every time. Until we have smarter and more prescient decision makers in leadership, the bottleneck on output will be the quality of decision making not the quality of code. Trying more things faster and cheaper will win.


Aka the cheap plastic solution always wins.


True but that is maybe 5% of what is being promised by the average booster


Give examples of boosters (average or not) and what they've promised?


I wonder if instead (or in parallel), Norway should build a set of training data and share it (for free) with all the model builders.

Seems like making the frontier models know Norwegian and their culture is a better (or additional!) way to reach the end they are going for here.


The frontier models know Norwegian just fine. They can also adapt to Norwegian dialects, and even ape old Norwegian fairly well.

E.g. I had Claude describe the novel "De knyttede næver" from 1911 in Norwegian orthography ca. 1911, as it's a novel I've read, and it does a good job.

What it lacks is an understanding of Norwegian literature, culture and history. It had to look up "De knyttede næver", which was one of the best-selling Norwegian novels around the time it was published before I'd get anything out of it (ChatGPT does better; in thinking mode in particular it gives a detailed summary).

While not exactly well known today, the author was a prominent newspaper journalist for decades, and the novel series is well enough known that e.g. there's a Norwegian singer that took his stage name after the protagonist, and it was covered in Norwegian papers and books for decades (partly because of controversy over the authors political views and how they coloured his novels), so it does feel like a reasonable test that reveals a quite significant knowledge gap.

I do agree with you that it'd be better if the data set from the national library was made more accessible, though it seems a major addition here is that they have a deal to train on copyrighted data locked away in their archives that they have limitations on the use of.

But even just making the out of copyright data in their collections would be a great start.


Odd, I'd imagine Wikisource (in many/all languages) would be part of training data for all LLMs with SOTA ambition?

https://no.wikisource.org/wiki/De_knyttede_n%C3%A6ver


You'd think so. It seems like there are a lot of odd gaps like that.

I also have a favourite English language PhD thesis I ask every new model about that they still struggle to find even though there's a Wikipedia article about it that links a blog post I wrote about it.

Anyone who thinks they've exhausted even publicly crawlable resources should ask them about some obscure stuff.


you might be surprised if you take this approach.. give key words and phrases in small amounts, each sentence of a prompt building on a previous sentence. Take a an example that is not very hard, like Lewis Carrol Alice in Wonderland original text. Although a quick question might get things sort of wrong, or miss details, if you guide the LLM to a certain part of the story, then a certain set of characters in that part of the story, then a certain statement or dramatic moment with those characters in that part of the story, you might get very specific detail that is close to line-by-line accurate. On the other hand, if you ask a quick, ordinary question about the same part of the story without supplying context and character names, you get something equally vague. YMMV


For the PhD thesis in question, I've actually tested a lot of requests about different parts of it, and both Claude and ChatGPT still draws a total blank if you don't let them do searches.


the models don't retain their full training data set


No, but they do retain enough that it is interesting what they fail to retain.


Why should they share all this data with the greedy american corporations that are stealing everyones data for their own profit? Much better to keep the legal agreement with the national institutions and possibly develop something actual useful to their own country.


You are contradicting yourself. If you're hoarding the data for yourself you're not going to develop something useful. Sharing the data means that it will be integrated into the big LLMs, which will be useful "for their own country".


Obviously, it's fine to be wary of any development in your area. But it seems like there is a certain amount of irrational(?) fear of datacenters. And I really don't understand it.

I saw a poll recently that people would rather live near a nuclear power plan than a datacenter. That's... their choice, of course, but doesn't seem logical to me.

I have heard several "concern stories" about them on NPR recently. Maybe there is a political component to it. But I do worry there is some kind of manipulation being done.


Watch these

https://youtu.be/_bP80DEAbuo?si=4XpIb0vb8YjY1g_k

https://youtu.be/t-8TDOFqkQA?si=EB8zAF0JYHvOB23a

https://youtu.be/3VJT2JeDCyw?si=ak7haiWzbX9O8BL9

Then, tell me if you want to live anywhere near those.

Then, tell me of a nuclear power plant that has that bad a repo.


Have you read the responses to (at least) the first of these videos? https://blog.andymasley.com/p/contra-benn-jordan-data-center...

Also, I thought the response by Benn Jordan on Bluesky was informative. https://blog.andymasley.com/p/contra-benn-jordan-data-center...


I read the first link and it said:

> When low-frequency sound becomes strong enough to be heard or otherwise felt, it can cause annoyance, discomfort, and sleep disruption like any other normal noise pollution.

So which is it? Sure, I don’t really believe that there is magical super special harmful noise from a datacenter, but are these monster datacenters emitting disruptive amounts of low frequency sound or are they not?


It would be helpful if you didn't post rebuttals from people with a massive financial incentive to do so.


Ad homniems aside, is the accusation even accurate? So far as I can tell he doesn't obviously have "a massive financial incentive to do so", like he's a VC investor in anthropic or whatever. He does seem to be bullish on AI in general, but I'm not sure why that'd be a disqualification for someone on the pro-ai camp any more than someone who's interested in retaining their property values or whatever would be a disqualifier for the anti-ai camp.


That's not an ad hominem, though?

Ad hominem would be if shimman had said something like "don't post rebuttals from people who are stupid meanyheads". Identifying a characteristic of the posters that affects their incentives is a perfectly legitimate reason to discredit their posts, or at least call their impartiality into question.


From wikipedia:

>Ad hominem (Latin for 'to the person'), short for argumentum ad hominem ('an argument to the person'), refers to when a speaker attacks the character, motive, or some other attribute of the person making an argument rather than the substance of the argument itself.

>motive

Emphasis mine.


No it wouldn't. I want to hear his argument.


I think it's as simple as people generally believe that electricity is good for the world, and AI is bad for it. In the former case you're kindof taking-one-for-the-team. Even places that have nasty nasty stuff like tailings ponds generally have a kind of civic pride that coal mining or whatever was a necessity and the sacrifice of their local environment made a lot of other people's comfy lives possible. Data-centers are just not going to inspire that sentiment lol.


For some strange reason people aren't all that keen on building something that'll increase their utility bills, pollute everything, and threatens to take their job.


Here in Michigan, people are all for auto factories, but the polluting, energy intensive, and job taking data centers are a big no no. They use electricity, they look ugly, and they use water. Can't have that.


The auto factories propose to actually create jobs, though.


> Here in Michigan, people are all for auto factories, but... data centers are a big no no. They use electricity, they look ugly, and they use water. Can't have that.

Because they know economics better than their politicians and academia.

Data centers saddle the public with their power and water capital expenses, for new generation and transmission which are used solely for the benefit of data centers. And get this, in many cases the data bros get sweet tax-free deals for many years.

All of a sudden, the entire economics establishment loves communism for the rich, where the rich get exclusive use of public utilities built and paid for by the public.

The media, academia and politicians silence is deafening, which is why people have to raise their voices if they want to be heard.


> That's... their choice, of course, but doesn't seem logical to me.

Wouldn't the question be more simply, Do you want your power bills to go up for the same power used?

And the nuclear accidents that have happend have mostly been overblown (apart from Chernobyl).


Thanks to information campaigns people who live near nuclear facilities tend to have an above average, positive, view of the safety and threat.

A large part of my extended family lives near a large facility. Wind turbines launching ice at the nearby roads is a larger (yet trivial), safety concern.


I don’t know the background to this project, but a nuclear project would likely be very transparent - with public studies on the impacts and meetings for the public to make their views known. It’s far quicker to build a datacenter than to increase local grid and water capacity later.

> The Stratos Project moved forward with far too many unanswered questions around water, power, cost, and transparency.


In a town near me a paper plant recently closed (to much anger), then people protested the potential use of the land for a data center, citing concerns about noise, water use, power use, and traffic.


I have childhood memories of visiting my grandparents, who lived near a town with a paper mill. Going to town, going to a restaurant etc, meant being inundated with the horrible smell produced by that mill, within a radius of miles. It was a fact of life there. Genuinely hard to imagine a data center producing worse waste products than that.


Datacenters are financially a net negative for whichever municipality they end up in. They're operated mostly remotely with little staff and they have no tangible production, meaning any wealth they generate ends up vast distances away. Meanwhile the municipality ends up with increased costs because of the inefficiencies of bruteforcing computation, and because of the subsidies and tax breaks that the companies not only expect but demand for construction, there's no revenue being generated even for the local government.

That alone is enough of an argument against them.


Why did you not include the tax they bring in? I think this is a serious omission and points at motivated reasoning.

Can you do one where you account for the tax dollars and compare it to similar industries?


Because in reality they don't actually bring in any revenue for the first few years thanks to all the subsidies and tax breaks they demand upfront before construction. At best it would be state and federal taxes used for operations that any operating business brings like federal payroll, but there wouldn't be any property taxes for at least five to ten years and the federal corporate income tax would likely be from the state the company is based in rather than the state the datacenter is based in. The municipality, be that the county or city the datacenter's in, gets screwed.

Meanwhile just to run a trucking depot you'd have the heavy vehicle tax, international fuel agreement tax, registration tax, sales taxes for the trucks and trailers, property taxes, and whatever incidental taxes required by the state you're operating in. The property tax, IFAT, and local payroll taxes meanwhile all go to the municipality and don't skip straight up to the state or national level. This is with no expectation of any of this being waived or delayed because the trucking industry doesn't have the surface visible financial performance of the industries municipalities are more lenient towards.


False. They bring in huge amounts of taxes.

> Over the past 20 years, the data center industry in Loudoun County has grown significantly. As a result, the amount of revenue that the county receives from personal property tax revenue on the computer equipment located inside data centers has also grown significantly. Currently, data centers occupy approximately 4 percent of commercial parcels of the land in the county; however, they yield 38% of general fund revenue collected by the county

38% of fund revenue for this county comes from data centres.

https://www.loudoun.gov/m/faq?cat=241


It’s pretty simple. People think that AI will take their jobs and maybe murder them, probably because the people developing AI have said it’s going to take their jobs and maybe murder them.

Opposing data centers is the biggest lever most people have to impede AI development.


They're big, use up a lot of power, destroy a large batch of land, produce noise and locals get basically nothing out of that (it's not like they provide a lot of jobs or anything). The power bills also go up.


It makes more sense when you look at how much waste heat it will generate and how it will get its energy [1].

[1] https://www.sltrib.com/news/environment/2026/05/07/utahs-dat...


As someone who lives in Northern Virginia, it makes me furious to receive my electricity bill and see that even though I used less electricity than the same month last year, I am paying significantly more. And this happens every year.

Do you think Virginia is adding solar, battery, and wind proportional to that additional power draw? Nope! It's natural gas and coal power imported from PA and WV. It would be one thing if I was paying more to build out renewable energy for environmental purposes and to set up a reliable and clean grid for the future. But no, I'm just subsidizing these huge companies and hurting the environment to boot.


This echoes some of my biggest gripes about data centers:

We should be mandating green power, to a great extent, be built to support these facilities.

We (US states) should not be competing, in a race to the bottom, to be the state to give the biggest tax breaks and pass the cost to the citizens.

We should not be ignoring the citizens who will have their health and livelihoods affected.

AI data centers, for better or worse, are very necessary for many reasons. They could be built responsibly, or at least less hazardously, but the care isn't being put into that aspect of their construction.


So true. A legislator in Virginia finally pitched ending the tax credit that brought all the data centers to VA in the first place, and I hope it passes (I know it won’t). But seeing the upcoming rate increases already on the books and the number of data centers they are planning to build is pushing me to consider solar again. The payback time is getting shorter and shorter :)


Could you share more about the rate increases? The newspaper articles I've seen seemed sketchy on how people were affected.


Aside from "moral outrage" style concerns ("AI is bad for the environment", power consumption, water consumption, or "datacenters benefit rich people, rich people bad, so datacenters bad"), I've heard of specific bad examples how datacenters (allegedly) negatively impacted the surrounding population:

- Noise (from fans to generators to possible infrasound concerns)

- Air pollution (from data centers semi-permanently running on generators)

- Electricity prices (although I don't understand how this is supposed to work)

- Water consumption affecting the population (water restrictions, price increases, water table dropping)

Many of these are one-sided stories told from the perspective of the residents that I didn't try to verify, but I suspect some of these concerns are legit.

The company building the datacenter has a lot of incentives to cut corners and/or cause some of these impacts, externalizing its costs (e.g. by saving money at the expense of noise emissions, running the DC on unpermitted gas turbines to be able to build a DC where there isn't enough grid, negotiating clever deals that benefit the company but screw over the utility forcing it to raise prices for others, using groundwater for evaporative cooling to make cooling cheaper, etc.)

The company building the datacenter also likely has a lot more experience while the people of the town and the town itself are doing this once, so there is an inbalance in experience that makes it easy for the company to get away with some of these.

There is very little benefit that the people of the area can expect from a data center - as I understand it, there are very few jobs in one past the construction phase, even the construction jobs are often filled with experienced travelling workers, and given the negotiation imbalance, a town seems likely to get screwed on any contributions that the data center promises.

Maybe the solution would be some kind of framework/organization that guarantees (ideally with binding, well tested contracts) that the datacenter won't be a nuisance, builds a reputation for being reliable, and in exchange, companies that work under that framework can expect quick approvals and less pushback.

Until that exists, or companies start offering guarantees up front (e.g. guaranteeing a certain power price or noise level), I'm not surprised that people push back (especially if the company building the data center has screwed up in the past).


All of these concerns are valid. But none of them are unique to datacenters.

A golf course uses a lot of water. A factory can use a lot of power -- and generate pollution. A chemical factory could have all kinds of externalities (if not properly managed.) Heck, switching to electric heat (over gas) or electric cars over ICE for an area will also drive up power usage.

But we don't freak out when someone builds a golf course or a factory or switch to electric.

We have rules about all those things. Sound is one: you need to be within reasonable limits. Electricity usage is another: power operators always need to manage their load and expand generation (that's why we keep adding solar and wind everywhere.) Air pollution is similarly managed.

I can understand if people are concerned about "infrasound" -- why not pass a law that regulates it -- like other noise limits?

Datacenters may have specific potential issues. But none of them are unique to datacenters. And we've been managing these issues for hundreds of years.


> I saw a poll recently that people would rather live near a nuclear power plan than a datacenter. That's... their choice, of course, but doesn't seem logical to me.

Data centers come with gas-fired plants that pollute the air and reduce your life span. It’s quite rational to not want to live next to one of these: https://www.wired.com/story/a-new-google-funded-data-center-...


The resource consumption is huge and it provides relatively little to the surrounding community compared to its intake. For most residents who live near one it’s a net loss. Qualify of life decreases and utility bills go up so that a Silicon Valley exec can get a nice bonus for closing the deal.

A nuclear plant creates energy and a decent amount of jobs, while a data center’s value is dubious to the average human and the data center barely brings in any jobs.


> I saw a poll recently that people would rather live near a nuclear power plan than a datacenter. That's... their choice, of course, but doesn't seem logical to me.

Yay people have finally become rational about nuclear power safety !!!

...right, right?


You’re saying it as if living near a nuclear power plant is bad or something


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

Search: