Yup, LLMs are rocking for smaller more greenfield stuff like this. As long as you can get your results in 5-10 interactions with the bot then it works really well.
They seem to fall apart (for me, at least) when the projects get larger or have multiple people working on them.
They're also super helpful for analytics projects (I'm a data person) as generally the needed context is much smaller (and because I know exactly how to approach these problems, it's that typing the code/handling API changes takes a bunch of time).
The cultural differences between companies in a country are superficial compared to the cultural differences between countries.
We have strong evidence that deeper cultural, everything from attitudes towards saving, government, and social trust, persists for generations after immigration: https://www.nas.org/academic-questions/36/2/rethinking-immig... (“The authors found that forty-six percent of home-country attitudes toward trust persist in second- and fourth-generation immigrants—in the adults whose parents, grandparents, and great-grandparents were immigrants. People from high-trust societies, like Sørensen, transmit about half of their high-trust attitudes to their descendants, and people from low-trust societies do the same with their low-trust attitudes.”).
You can see this just by going around the country. Scandinavia has much higher social trust than Italy. The upper midwest, where Scandinavian immigration dominated, has higher social trust than NJ/NY, which saw mass immigration from southern Italy.
These deep-seated cultural variations, in turn, have a strong impact on societal prosperity: https://news.harvard.edu/gazette/story/2020/09/joseph-henric... (“One of the points I want to make is a lot of the big institutions we think about, like Western law or representative government, actually flow, in part, from the way people think about the world.”).
> You can see this just by going around the country. Scandinavia has much higher social trust than Italy. The upper midwest, where Scandinavian immigration dominated, has higher social trust than NJ/NY, which saw mass immigration from southern Italy.
OK, that's interesting, I'll have to look into that book.
I can see that (as you said) the Nordic countries have much more trust than Italy, and Italy, Spain and France are similar (along with a similar language and large inter-mixtures over time).
However, look at Ireland vs the UK. Basically the same genetics, an extremely similar culture (particularly given the amount of cross migration back and forth), and yet very divergent amounts of social trust (I'm sceptical of the metric here, would like to see it very density as I suspect that drives a bunch of the results).
> Think about your own life. How important is food to your family and friends as a way of social bonding? Do you think you’d be able to change that easily?
In terms of my parents/culture, not at all. It was much, much, much more about drinking alcohol rather than food. And yet, while that part is still there, there's far more emphasis on food as a socialisation tool in my generation.
Some of that is because of drink-driving laws being enforced, but some of it is definitely a cultural change which would seem to argue against your suggestion of long-term impacts due to culture.
> The cultural differences between companies in a country are superficial compared to the cultural differences between countries.
Again, I'm not convinced this is true. Like, if a company in Ireland has majority European employees but American leadership, what culture will it have?
> You can see this just by going around the country
I think that the particular outcomes of one country, predominantly founded by Europeans, tells us very little about how culture works.
> However, look at Ireland vs the UK. Basically the same genetics, an extremely similar culture (particularly given the amount of cross migration back and forth), and yet very divergent amounts of social trust
Ireland is culturally distinct from the U.K. For example, the U.K. is historically predominantly Protestant, while Ireland is historically strongly Catholic. That manifests in many ways. For example, the Anglosphere tends to have the latest gestational limits on abortion among European and European-derived countries. By contrast, abortion was illegal altogether in Ireland until recently (2018).
There is also the fact that the Irish were brutally colonized by England and Irish society developed a strong cohesiveness from that external pressure. The Bengal Famine of 1943 killed 3 million people out of a population of about 60 million. The Irish Famine, by contrast, killed 1 million people out of a population of only about 8 million. Indeed, the Irish population peaked in 1841, a few years before the start of the Great Famine and never returned to that peak.
> I think that the particular outcomes of one country, predominantly founded by Europeans, tells us very little about how culture works.
According to official stats, 16% of Irish residents are citizens of other countries. Keep in mind that this number will exclude foreign nationals that got Irish citizenship through naturalization (and therefore became Irish citizens).
Most recent numbers from the UK list 16% of the population being "foreign-born". While this number may be similar to Ireland, it still counts someone as foreign born even if they became UK citizens by naturalization.
Also, consider that one of the most prominent migration sources for the UK is of Irish nationals (that can live and work in the UK even after brexit). Irish culture is not too dissimilar to UK culture (especially considering that Northern Ireland is currently part of the UK).
If anything, Ireland experienced more foreign culture immigration than the UK, not less.
Do they count North Ireland as 'foreign born' even though they are notionally Irish and born in Ireland (but not RoI)? Those have got to be one of the major 'immigrants' to RoI.
> The Good Friday Agreement, which was signed between the Irish and British governments in 1998, confirmed that people born in Northern Ireland could choose to be either British or Irish citizens.
> Since 1 January 2005, if you are born in Northern Ireland, you can claim Irish citizenship if your parent (or parents) are either British or Irish citizens, or one of them has lived on the island of Ireland for at least 3 out of the 4 years immediately before your birth.
Are you of Aboriginal extraction? Otherwise, I'm not sure an ethnic homeland for you would be Australia, right?
This stuff is so weird, as basically all humans migrated to wherever they are now. Like, I'm pretty sure that I have Celtic, Norman, Viking and other ancestors, despite my official ancestry being Irish (and all of my last 3-4 generations being born in Ireland).
Is it culture? That would seem to be what people are actually looking for, and I can definitely see the appeal, but culture is something that is generated from interaction with other members of a culture, and isn't dependent on genetics (consider how you or I might behave in OpenAI vs Anthropic vs Goldman Sachs).
Yes it is culture. The desire for an ethnostate is a proxy desire for a monoculture, somewhat easier to implement because it's easier to see someone is white than to see their behavior patterns. There are also studies that show most people have subconscious tension among other races even from the same culture, though.
But the outright desire for an ethnostate or a monoculture are both politically untenable in the West. Never mind that every country we idolize, e.g. Japan, Scandinavia, are essentially that.
America is not a good candidate for that for various reasons, but I see no reason that Denmark or Japan shouldn't be able to codify their ethnic makeup and adjust immigration policy accordingly. As a Korean, it is really nice to visit Korea and feel among 'my people', even though culturally I'm an American. I've heard very similar things from my white friends who move to Utah or visit Scandinavia. It's a feeling that seems to be deeply embedded in the lizard brain, to be among your tribe, identified visually and then culturally. Countries have the right to cultivate this feeling among their citizens.
Okay, but those are very recent developments to which the historical sentiment hasn't fully caught up. To the extent it has, it's via negative memetic sentiments, e.g. "Stockholm is now the rape capital of the world!" Sweden, at least, also, doesn't capture racial demographics, so we don't know the makeup of the foreign born population. Walking around Helsinki, you don't need statistics to notice the homogeny.
Didn't have time to check Norway and Denmark before, but looking now it seems most of the foreign born population is still white/European.
> figures from World Population Review suggesting around 83.2% are Norwegian, and another 8.3% are other Europeans, totaling roughly 91.5% of European descent, though exact "white" percentages vary by source and definition, with estimates often placing the broader "white" or European-origin population well above 90%.
Another source I found puts non-Danes in Denmark at 9%, putting their white/European percentage at 95%.
For approximate parity comparison, Japan is 98% Japanese, UK is 82%, and France is 71%, all falling. Imo Norway and Denmark still qualify as ethnostates, though maybe not for much longer.
Whites visiting scandinavianis a funny one, because today we treat all white people as one race but unless that man was of origin from scandinavia it's most probable its the governance system and culture he likes not a racial kinship.
> Countries have the right to cultivate this feeling among their citizens.
I disagree, because what seems to always be the logical next step is "My monoculture is superior, and deserves dominion above your monoculture, which deserves eradication. And oh by the way, if you don't like our monoculture and try to escape it, we'll invade that place too."
That's kind of what the "mono" in monoculture implies. Multiculturalism isn't an asset because it's the most stable form of soceity, rather it's best we've figured out when the alternative is bloodshed between warring tribes.
I don't see it that way. From my perspective, America would be more stable socially if it balkanized. There is a lot of tension borne of cultural heterogeneity. Coasts vs South, most of Texas vs Houston, etc. Nothing about it is positive, except maybe economically.
I guess this might lead to wars via a more cohesive national identity, like how diversity in the workplace reduces unionization efforts, but I largely doubt it would turn out so poorly.
I don't necessarily think it would. But what makes you think that a hypothetical status quo will go any better than what happened in the Roman empire? There are no great options imo.
What is the Roman Empire worse than in your mind? It lasted centuries, its literally perhaps the longest running in history. If it's the fate of the Roman Empire we have to face, then I can't wish for anything better. You can of course say it slid into dictatorship, now prove it was due to the race admixture. Anyone can read anything into the transition and fall of the Roman Empire.
Suppose a future world where all the world, every country is thoroughly mixed up by today's racial standards. There is no "white" country, no "chinese" country, no "black" country. How do you think race wars would be organized, "Lets kill all the [racist slur] bastards of this country" wait what, there's no easy country to point to, already making race war quite difficult. Or do you like having race wars?
Again, I fail to see what exactly is the "loss" or "gain" of a race existing or not? What has it ever given us? It's like religion and nationalism, very little benefits and pure destruction and waste of human life in the balance of history.
I suppose you will then say, you fear your country would become "Muslim" or insert whatever religion you hate. But Muslim is an idea it's not a race. A religion can convert a country without even a single marriage or mixup because it's an idea, so tell me again how do you feel you are safe from a foreign religion just because your "race" is different? And the nice thing is the arabs or whatever race has you wetting your bed will also have had thoroughly had to mix up with the locals by that point, already rendering them "impure". And I have you covered with a great anti-democratic solution for that: forcible state mandated atheism.
If you do believe in anti-democratic system of governance, then I propose we force every person to marry a person of a different race and forcibly make every locality thoroughly mixed up using information theoretic entropy definitions. Why not, why's mine better or worse than your idea of anti-democratic rule?
Why do you feel among all the cultures there, you know which one among them would will out. I see no point in being among my race if it's a communist or fascist country. The race has already failed me then. Hell give me aliens from mars and I'd gladly live among them if they are liberal democratic and capitalistic. To me these three principles are paramount, the identity of the agents executing them is irrelevant.
I absolutely don't feel comfort based on common ethnicity. I find comfort more or less in any democratic, capitalist liberal society and I absolutely hate life in anything not this even if it may be my race people or anyone else.
> This is an "in distribution" test. There are a lot of C compilers out there, including ones with git history, implemented from scratch. "In distribution" tests do not test generalization.
It's still really, really impressive though.
Like, economics aside this is amazing progress. I remember GPT3 not being able to hold context for more than a paragraph, we've come a long way since then.
Hell, I remember bag of words being state of the art when I started my career. We have come a really, really, really long way since then.
Do we know how many attempts were done to create such compiler before during previous tests? Would Anthropic report on the failed attempt? Can this "really, really impressive" thing be a result of a luck?
Much like quoting Quake code almost verbatim not so long ago.
> Do we know how many attempts were done to create such compiler before during previous tests? Would Anthropic report on the failed attempt? Can this "really, really impressive" thing be a result of a luck?
No we don't and yeah we would expect them to only report positive results (this is both marketing and investigation). That being said, they provide all the code et al for people to review.
I do agree that an out of distribution test would be super helpful, but given that it will almost certainly fail (given what we know about LLMs) I'm not too pushed about that given that it will definitely fail.
Look, I'm pretty sceptical about AI boosting, but this is a much better attempt than the windsurf browser thing from a few months back and it's interesting to know that one can get this work.
I do note that the article doesn't talk much about all the harnesses needed to make this work, which assuming that this approach is plausible, is the kind of thing that will be needed to make demonstrations like this more useful.
> No we don't and yeah we would expect them to only report positive results (this is both marketing and investigation).
This is matter of methodology. If they train models on that task or somewhat score/select models on their progress on that task, then we have test set leakage [1].
> This is matter of methodology. If they train models on that task or somewhat score/select models on their progress on that task, then we have test set leakage [1].
I am quite familiar with leakage, having been building statistical models for maybe 15+ years at this point.
However, that's not really relevant in this particular case given that LLMs are trained on approximately the entire internet, so leakage is not really a concern (as there is no test set, apart from the tasks they get asked to do in post-training).
I think that's its impressive that this even works at all as even if it's just predicting tokens (which is basically what they're trained to), as this is a pointer towards potentially more useful tasks (convert this cobol code base to java, for instance).
I think the missing bit here is that this only works for cases where there's a really large test set (the html spec, the linux kernel). I'm not convinced that the models would be able to maintain coherence without this, so maybe that's what we need to figure out how to build to make this actually works.
> I think the missing bit here is that this only works for cases where there's a really large test set (the html spec, the linux kernel). I'm not convinced that the models would be able to maintain coherence without this, so maybe that's what we need to figure out how to build to make this actually works.
Take any language with compiler and several thousands of users and you have a plenty of tests that approximate spec inward and outward.
The GHDL test suite is sufficient and general enough to develop a pretty capable clone, to my knowledge. To my knowledge, there is only one open source VHDL compiler and it is written in Ada. And, again, expertise to implement another one from scratch to train an LLM on it is very, very scarce - VHDL, being highly parallel variant of Ada, is quirky as hell.
So someone can test your hypothesis on the VHDL - agent-code a VHDL compiler and simulator in Rust so that it passes GHDL test suite. Would it take two weeks and $20,000 as with C? I don't know but I really doubt so.
> Democrats turned on Elon during covid, as he was against lockdowns. This spread into wide distrust and complete sidelining and antagonism towards tech during the Biden administration. You can hear about it from Ben and Marc on their a16z podcast, when they explained why they are endorsing Trump.
Look, a16z basically are talking their book. They went heavily into crypto, and when the Biden administration started taking actions against crypto, they started supporting Republicans.
Who said it was complicated? The biden administration also had plans for the AI industry which would kill them. They also had zero interest in any new tech (smr, vtol). The democrats turned against big tech and new tech.
> The democrats turned against big tech and new tech.
While I didn't agree with a lot of the governmental responses to ChatGPT (the bowlderising of the EU AI act is my personal lowlight), I think that a lot of what the Biden admin was against was monopolies, which I entirely agree with.
It's super dangerous (as an example) that basically 3 companies now control all online advertising, and another 3 have a monopoly on renting servers. This is bad for markets and society so I was mostly onboard with a lot of what they did (and am very disappointed in the results of the trials, particularly the Google ones).
I don't think this is anti-tech, it's anti-monopoly which I think is really good for basically everyone here over the longer term.
That being said, a16z were totally just trying to make sure they had time to unload all their crypto "investments" to another sucker.
It's definitely not a major threat, but many/most finance people are clueless about what is and isn't possible with LLMs.
Again, unless Anthropic are taking on liability for their legal tools, this is not going to impact TR.
That being said, there probably is a potential company here that's gonna be built soon/is currently being built, but it definitely won't just be a wrapper around Claude as the recall will be way too low for these systems unaided.
> mpanies to usually notify you if they receive your data as controller (though there are some exceptions), in reality that's not really happening though (e.g. how many payments processors or acquiring banks have notified you about your credit card payments?).
Depending on why they received your data, they may not be allowed to tell you about this. The Bank Secrecy Act has had a lot of weird downstream consequences.
Sure, but that's in connection with SARs and such (which have legal obligations are around secrecy). What I mean are the "generic" credit card payments where payment processors & banks process the personal data for things like fraud detection. That's perfectly fine legitimate interest, but that doesn't absolve them from article 14 requirements as fraud prevention doesn't have such requirements around secrecy around the fact that it even exists. They can restrict some detailed information e.g. regarding algorithm itself by relying on trade secrets, but that is different from their obligation to inform data subject that they received the information.
So just to clarify there are two different things here:
The information that fraud detection is being performed is something that needs to be disclosed. That's what would be part of the article 13/14 (13 is when controller collects data directly from subject, 14 is when they receive it from anywhere else (including generating it themselves)) notices. It's very rare that any law would forbid giving any kind of article 13 notice, that's why banks do disclose that they process personal data for AML purposes in their privacy policies.
Article 14 itself however does allow omitting the notice in certain circumstances, but those are quite limited. Fraud detection can fit here, but usually only in the context where controllers transmit the information to other controllers regarding risky clients and such. The actual fraud detection itself is a different purpose and it's objectives are not, generally speaking, in risk just because someone knows that certain company ran the fraud detection on this transaction (since fraud detection is run on every single transaction).
The "how" is part of the second thing. That's generally more on article 15 (and 22) territory where controller could omit the information why exactly the transaction was denied (and possibly things like transaction's fraud score). I don't really like the current interpretations either (as it makes it pretty impossible to fix incorrect information) but unless CJEU gives some ruling in the issue it's unlikely that DPAs & EDPB are going to enforce some changes there.
Yeah, but basically all of those are either standard for SMEs or no-ops.
For instance, if I run a bakery and sell baked goods online, I'm probably using Shopify who comply with this with one button.
Even if I built the baking website myself, all I need is email address and physical address to send delicious baked goods to you. I need to keep the payment records for a long time (for dispute prevention if nothing else) but that's it.
Where is the GDPR hassle in this case?
Just stop collecting data you don't need (or make sure it's for a good reason, like fraud prevention) and you'll be fine.
If said bakery creates accounts, it's a little more involved but basically you just need to implement soft delete to comply with your obligations.
I'm not sure this is a massive hit, can you help me understand what SMEs exactly are going to be hit by complex GDPR compliance?
No, a bakery using Shopify will not spare them having these documents. You show a respectable amount of ignorance only to then claim GDPR won't be a hassle in this case. It absolutely is a hassle, which you would know, had you familiarized yourself with the subject.
Even stating "just stop collecting data you don't need" shows, that you did not care to read my response before you replied to it, and how little you generally know about the topic.
Not repeating what I said, I will add this: if you do collect personal data (and you WILL if you do anything online, write invoices or just have a security camera on premises) than you will have to have these documents ready.
Most of the information relates to online marketing, which does tend to come with more GDPR compliance requirements. My wife runs a business through Shopify and the only thing we need to worry about is email addresses.
Can you help me understand what you see as the issues around GDPR compliance here?
reply