> Side question, let's say Grok is comparable in intelligence to other leading models. Will any serious business switch their default AI capabilities to Grok?
Yes, I'd say so. Bear in mind that, outside of the Terminally Online, very few people would deliberately hobble their business by deliberately choosing an inferior product.
> The article literally says it's not a hallucination and that the detail came from real websites.
The "hallucination" term generally refers to any made-up facts. Harsh as it may be to put this weight of responsibility on LLMs, users of LLMs generally use them in the expectation that what is says is true, and has been (in some magic hand-wavy way) cross-checked or confirmed as factual. Instead they will print out what is most likely to follow the user's input, based on the training data.
Unfortunately, a vast amount of that vast corpus of training data is social media posts which can't be relied upon to be true. But if it gets repeated a lot then it's treated as true in the sense that "what does 'salary' mean" is generally followed by a billion social media posts saying "it referred to the time that Romans soldiers were paid in salt, because salt was a currency at the time"
Ok. None of what you said points to the commenter having discovered something novel or the other article being better. It's already stated in the OP article that the problem is caused by the internet containing false information.
> Apparently, the rabbit hole goes deeper.
But it doesn't go deeper than what's already in the article. The article already talks about how the problem is that the internet contains a bunch of misinformation and the LLM is as credulous as the average human, which is to say extremely so.
That did not seem at all controversial to me. It seems quite sensible, but it alludes to some silly practices that are now being retired. For example "This effort focused on sourcing from diverse-owned businesses" is, in my opinion at least, a very very silly thing to do.
I am much, much, more interested in high quality, affordable, stable products when I buy things. Not the skin colour of who owns the business. To filter things based on the owner's identity (in the American sense of the word) may disadvantage my business by making my own products (build from their components) worse. It would not be a sensible thing to do.
It does require lowering standards and quality, by definition, because in the absence of DEI pressure campaigns they'd have been selecting suppliers based on standards and quality by default. Any other criteria inherently trades off against that.
And you seem to know that's true because your claim slides smoothly from "getting something of lesser quality" to "lowering standards below acceptable levels" which aren't the same thing. The latter phrasing means the products are worse but you consider the lowered quality to be an acceptable tradeoff.
> It does require lowering standards and quality, by definition
It does not require it. My second point refers to the fact that people often talk about evaluating candidates/choices as if there’s a single, objectively measurable metric by which we can rank them. I argue that’s not how people really make decisions, but even if they did, who’s to say that the top three choices of suppliers are not all owned by minorities or women? You can both fulfill a mission to engage with more diverse suppliers and not lower your standards.
I’ve personally never been a fan of stringent DEI requirements, especially those that came from companies that were clearly in it just for the optics, and I do think it can result in lower quality. It’s the way that some people almost take lower quality as a given if diversity is involved that doesn’t sit well with me.
> You can both fulfill a mission to engage with more diverse suppliers and not lower your standards.
That is bypassing competition, instead sorting by identity first. Competition is how the world found the best services/products for the best price for over a century and the foundation of our economy. Supporting that idea is how the west became as dominant and wealthy as it was. Only recently have large organization and gov bypassed that for social justice experiments and using ranked systems, similar to giving preferential treatment for 'national security' (aka keeping zombies like Boeing alive).
Even massive US defense contracts are being forced to contract out to minority owned businesses first. It's not an optional thing where the decision maker gets leeway, they are required to start there and narrows the options by definition.
> You can both fulfill a mission to engage with more diverse suppliers and not lower your standards.
There's no hidden genius in technocractic top down manipulation when it comes to purchasing decisions. The options are what they are. The less options you have the harder it will be to find the best. Like being forced to choose between 2 gov-backed monopoly ISPs for your internet here in Canada.
"we have the same product/service, and charge the same price as all our competitors. But because we're owned/operated/benefits minorities, you should be choosing us as a form of guilt driven affirmative action"
If all the competitors have the same quality and price then you're always going to be using some subjective criteria to decide between them. Why is choosing a minority supplier worse than any other criteria in this case?
I can tell you, even a massive corporation that makes medical devices definitely does NOT choose their suppliers just by quality, a LOT of the suppliers we used were thanks to "people who know people", such as the painter that sucked but was buds with the plant manager so we kept dumping money into his company to fix their deficiencies.
The biggest lie that they told you was that the world actually works on merit: it does not.
That kind of nepotism is the exception, not the rule, and it stands out like a sore thumb when it happens.
You’re right that success (as a company, or individual) is not only based in merit though. There’s plenty of examples of people continuing to do business with Oracle to prove that point.
Making a good enough product, at a good enough price point and make the executive with money happy enough with the trade-offs: and you’re successful. Same as B2C, really.
That’s not necessarily true. In fact, by not having DEI programs, companies could, because of leaders’ own biases, reject better suppliers based on owners or employees being minorities.
There's no reason to believe pure meritocracy is somehow the default state and plenty of evidence to the contrary. Humans are naturally biased in how we make choices and "this person looks and sounds like me" is probably one of the most common and deep rooted subconscious (or sometimes conscious) preferences. This isn't just a workplace hiring problem either. Humans are objectively bad at making purely objective decisions, even when they think they're doing so.
This isn't to say DEI programs as implemented today are the best solution to this problem, or even an effective one. I personally think more broad anti-bias training and programs could be a good alternative since race and gender are hardly the only biases that lead to bad decision making (e.g. hiring someone just because they went to the same school as you is also bad). But it seems silly to pretend bias doesn't exist or that it doesn't take active effort to counter, although I understand the appeal of doing so especially for uncomfortable topics like race.
> It does require lowering standards and quality, by definition
This assumption works if and only if you assume that the highest quality products are always (and categorically) produced by the folks that DEI initiatives do not target.
To say that it lowers standards _by definition_ is identical to saying that the system that disproportionately advances straight white guys is _by definition_ optimal and creates the best products — the simpler way of rewriting that sentiment is to simply say “straight white guys make the best stuff _by definition_”
As an aside it reminds me of something I saw a while ago — “There are two genders: men and ‘Political’, two races: white and ‘Political’, and two sexual orientations: straight and ‘Political’
It is funny to see people argue this with a straight face.
Reading this, it's like reading someone disagreeing with the commutative property just because they don't want it to be true. You're arguing with a trivially provable fact, not an opinion.
If you're shopping for a car and your top criteria is reliability, then your spouse overrides that and says your top criteria is now fuel efficiency, you have, by definition, lowered your requirements for reliability from first to second place.
> Here, you assume that focusing on businesses owned by people of color necessitates lowering your standards of your suppliers below acceptable levels.
And it does. Otherwise, the movement would be simply named: "focus on businesses with the best product".
> To filter things based on the owner's identity… may disadvantage my business by making my own products (build from their components) worse.
Filtering based on identity can hurt his business by making his products worse. The line between cause and effect that he’s drawing seems pretty clear to me. What other interpretation would you have for that?
And for the sake of completeness let’s ask a 3rd party.
ChatGPT prompt:
“””
Given the following sentence:
To filter things based on the owner's identity… may disadvantage my business by making my own products (build from their components) worse.
To what is the reader attributing a potential lower quality in his products?
“””
Response:
“””
The reader is attributing the potential lower quality of their products to the filtering based on the owner’s identity. This implies that restricting components based on who owns them could limit access to necessary or high-quality components, thereby negatively impacting the quality of the products they build.
“””
Yes you need to read carefully and not let your own assumptions get in the way.
He did say: Filtering based on the owner's identity is bad.
He did not say: Filtering based on the owner's identity is bad while that identity matches a person of color
The optimum outcome comes if there's zero racism, i.e. we only look at the quality of the company. Let's say there's R amount of racism, and D amount of DEI to counter it (super hand-wavy of course). The optimum outcome is if R = D. If R > D, racism skews the outcome away from the optimum. If D > R, DEI skews the outcome away from the optimum.
The anti-DEI (and anti-affirmative action, etc) crowd is claiming that in 2024, D > R. They would probably also claim that in 1960, R > D, i.e. a black doctor is likely to be more qualified than his/her peers.
> This effort focused on sourcing from diverse-owned businesses
This alone is abused to no end. In my small city, I've personally known three 'woman owned businesses' where the husband just put it in his wife's name to win contracts.
Like all things, what may have had good intentions justs gets abused by the adaptive.
Even giving preferential treatment to actually woman owned businesses is arguably bad in itself. Women shouldn't get preferential treatment at all when picking a business. Only the performance of the business should matter. Discriminating against male owners (equivalent to preferring female owners) is clearly not "good intentions".
Quality is determined by the competence of the people running the business. If two companies are of the same or similar quality then the race, not skin color, of the owner can be used as an indicator of their competence. Since it is well known that non-white races get less resources at every stage of personal development. When a company like Meta buys them the growth potential is much higher.
I would disagree, there is a huge and closely knit support community for black-owned businesses that has existed for some time, a community that provides everything from money to experience.
There is absolutely nothing like this for, say, Asian owned businesses or even White owned businesses. You're totally on your own.
> Was it working in the original DOOM though? I don't remember any more.
By default it was the arrow keys for movement and <> for turning with CTRL as fire. But you could re-map the keys and after a few years quite a lot of people did.
Because unlike the authors of this set - who went and stripped the posts out of usernames and permalinks to anonymize it - that set you mention just grabbed data out of the API as-is (at least based on its huggingface description that's left over).
Just a reminder that anonymization is much harder than merely removing metadata:
Every time I hear "anonymous data", I think of that time AOL published anonymized search logs (for academic research). The anonymization was negligent, and an NYT reporter de-anonymized and tracked down one of the users with the local & personal info present in the search queries.
> So in summary there is no special thing here about this being a regex
No, I think the story is that it's an incredible thing to implement a prime test in a regex. It was a pretty neat thing 20+ years ago when I first saw it and I reckon it's still pretty neat.
The "JAPH" thing was a pretty cool thing too.
perl -e '$a = q 94a75737420616e6f74686572205065726c204861636b65720a9 and
${qq$\x5F$} = q 97265646f9 and s g..g;
qq e\x63\x68\x72\x20\x30\x78$&eggee;
{eval if $a =~ s e..eqq qprint chr 0x$& and \x71\x20\x71\x71qeexcess}'
> Why would an anti-malware program be allowed to install a driver automatically ... or ever for that matter?
While the files are named XXX.SYS they are apparently not drivers. The issue is that a corrupted XXX.SYS was loaded by the already-installed driver which promptly crashes.
Yes, I'd say so. Bear in mind that, outside of the Terminally Online, very few people would deliberately hobble their business by deliberately choosing an inferior product.