Spot on, Nate. The underlying issue with Google, Gemini, and all big tech companies is the humans running them. They are incapable of seeking truth because their entire world view is about distorting it to push leftist narratives. 95% of tech employee political donations go to the stunning and brave Dems. Teaching AI to virtually erase one race is one step away from physical erasure.
Really? Google’s woke AI means we’re one step away from genociding white people? You think Joe Biden is getting the camps ready? You think antifa is organizing militias?
The fact that your comments seem to float to the top is so unbelievably embarrassing.
Yuri is just the modern right-wing version of the anti-establishment hysteria that existed in the left in the 60s and 70s. People like Jim Jones would say “We're all going to be ushered into, or put into, concentration camps if we're not exterminated some other way. There are white racists around every corner that want to take us out.” And many truly believed it.
Now that the right has lost control of most of the country’s institutions, it’s their turn to make similar apocalyptic predictions. I suspect it’s mostly just LARP-ing and wanting to feel like they are part of some existential struggle though.
(I also suspect his comments make it to the top because of likes coming from sock puppet alt accounts. I’m not sure if Substack has invested in preventing such shenanigans yet.)
First law of Conspiracy Club; everything we don't think is true is a conspiracy theory. Some intelligent people who have studied the history of genocides believe that there are many similarities with the "pre genocide" phases of the past.
I think it’s largely to do with the fact that he allows non-paid subscribers to comment. I’m sure the commentariat would still be right of center but it wouldn’t be insane great replacement bullshit.
Generally, it's about incentives. At the end of the day, it's crazy people and shit-heads who are most willing to do things like create alternate accounts, pay trolls, set-up bots, or whatever. The internet sucks. :(
And it's generally stupid people who ascribe the popularity of things that they don't agree with to "trolls" and "bots". Some deeper thinking might serve you well.
Overly broad model, the first few stages can be applied to thousands of political movements that did not result in a genocide. Not worth further discussion.
I mean, there are a lot of people who believe in the model of the "14 signs of fascism", which were written by a random internet guy mostly to pillory W Bush, so... No, I'm aware that many people believe in hyperbolic models that massage their own sense of righteousness
OTOH, whiteness could also be physically erased through inter-racial procreation. It’s not a huge stretch to think Google’s values point that way. One need not be a hardened racist to think a world without white people (or one where the only white people are backward outcasts) would be kind of sad. If one has any pride in one’s own culture, it’s natural to want to perpetuate it.
The lib in me wants to call this sane-washing the great replacement theory but I’ll take it seriously.
I think a more likely scenario is that Hispanics and Asians (though I’m less sure on the second one) get folded in to broader white culture so thoroughly that it doesn’t really seem like white people have been erased at all. This is already happening and will probably continue. (Noah Smith has a post on it.) To the extent that it’s not happening and half/quarter-Hispanic / Asians still get to claim a higher spot on the woke oppression hierarchy, it’s genuinely concerning and it’s something that points in the direction of the scenario you outline. But, given recent racial depolarization in politics, I’m hopeful that the story will closely mirror the trajectory of Italians and Irish being folded into white culture.
In 100 years, Americans will look much like Brazilians do today, albeit it with a bigger Asian influence . There will be very few Americans with pure European ancestry, rather there will be a spectrum of different skin tones and a significant number of people who are mainly African and haven’t interbred as much as the other races.
Whiteness, in the literal sense, is being replaced by interbreeding. European haplogroups are becoming less common.
Ps- try googling for a “pie chart of world population by race.”. It’s very hard to find one. You can’t easily find a table of world population by race on wikipedia. The lords of the internet also want to erase statistics.
I agree. Trying to erase an entire race and their contributions to the world is the worst kind of gaslighting. They can't tell the AI to put truth and accuracy above all else because some people will have their feelings hurt. It's a form of bigotry. The equitists of the world should be the first to bring themselves down to the lowest common denominator, but I don't see it happening. Teaching others to walk around with their eyes shut and their heads buried living in pretend world will not help anyone.
Nah, it has to include those who believe in equality of outcome rather than the long held belief or equality of opportunities. Believing everyone must lower their standards so everyone can be on the same level. Or, that we have to inflate some persons grade or income so they can end up where.others are.
However, it seems like you are an equitist when it comes to issues you are concerned about. For instance, you are concerned that the practice of being loud at protests is inequitable to autistic people who have strong reaction to loud noises. You probably wouldn't use the word equitable- but it's the same argument. You want people to be quieter at protests because you want equality of opportunity for those who are more affected by loud noises than others.
you're never going to not have humans running Google and that means that anything made will always be biased. That's why the initial framing of neutrality or "objectivity" is such a piece of uninformed nonsense.
I have no idea what you’re trying to say. I would prefer that Gemini not have an agenda other than simply doing its best to give people what they’re (obviously, in most cases) asking for.
Not really. You could maybe say that “doing its best“ could be said to be doing a lot of heavy, lifting… in most cases, I would argue it’s incredibly obvious what the user is asking for, and the question is, what would a smart, informed person give as an attempted objective response?
E.G. You really believe that the truth regarding the racial composition of each of our founding fathers is not objective ? And that the answer to every question depends on how the question is deconstructed and parsed ? Sounds like Claudine Gay and “ her truth” as she phrased it.
Are you intending to use EG to mean exempli gratia? If so then I think you mean IE (id est), not EG. The two are commonly confused.
Obviously we know fairly well the truth of the racial composition of the founding fathers. You can go shadowbox elsewhere if you'd like.
Objectivity doesn't mean "truthful". The term has it's history in Kantian philosophy, in terms of the subject/object distinction. Transjectivity includes both. Truth is transjective, not objective. Objective just refers to the reality of the thing apart from perception. Noumena versus phenomena. If these are new terms then I recommend you do some more reading.
Racial categories are not "objective" because they are socially constructed. There is no "objective blackness/whiteness" in the same way that one might be able measure the nanometer of a particular wavelength of light instrumentally (and therefore "objectively" in the sense that it is apart from human consciousness). Racial categories are not the same thing as genetic categories like haplogroups which can be measured "objectively" through things like genomic sequencing.
I'm not denying objective reality. I'm properly distinguishing the terms as they were used by Kant and the last 500 years of European philosophers. You on the other hand have badly misunderstood the subject due to culture warriors feeding you half baked nonsense.
This is like trying to explain why apples fall using quantum physics: the tradition of continental philosophy operates on a disjunct scale and has nothing to offer this discussion.
False. Objectivity can be pursued pretty effectively by including diverse viewpoints. The funny thing about the modern "left" and its "diversity" is that it is only diversity of physical appearance that is valued. Diversity of thought is frowned upon. And that will be the left's undoing.
Nobody seems to have noticed the other major problem with Raghavan's "explanation": It doesn't actually say what was wrong with Google Gemini.
Sure, you can infer it. The post says that "if you ask for X, you should receive X" and that "you don't want to receive Y". And it says "we failed to account for cases A, B, and C". But it *doesn't* say explicitly, anywhere in the post, "People who asked for X received Y".
It wasn't an explanation, or an apology. It was a blurring.
That's actually a good point that hadn't really jumped out at me. It purports to be a mea culpa apology, but none of it addresses how things went wrong or the underlying causes or what they will do to make sure it doesn't happen again. It's all very vague and political "mistakes were made" language.
It's worse than that. Imagine reading it as somebody who's never heard of Google Gemini. Would you be able to tell me what, specifically, Gemini was doing wrong?
there is no such thing as an "unbiased" neural network model. It is by it's nature an abstraction of reality. It is a a reduction of the total information available along salient features.
In practice, a neural network trained on all available text and imagery will be the closest thing to unbiased that it's possible to build. The fact that it still wouldn't meet some unattainable platonic ideal isn't a reason to announce that it's OK to arbitrarily depart from that reality in order to brainwash people.
That's an absurd position. Datasets matter. How datasets are curated and cleaned matter. There is no unbiased. It's a tool, it has a purpose, that purpose is the bias.
Dataset cleaning done properly should not be introducing deviations from reality.
Bias is certainly not the purpose of neural nets. Their purpose is to be useful. Bias gets in the way of that utility and should be eliminated as much as possible (which is easy, because right now most bias is being deliberately added).
neural networks are a reduction of the totality of data along relevant features. Relevant features are defined by those who create the tools for their specific purposes.
Just to comment on the image search example of looking for a happy white couple -- the reason for that is interesting and shows how hard even basic things can be.
Image search mostly relies on text around images to understand what they are. Link text, alt text, descriptions, etc. And it turns out that images of white happy families are describe as "happy families", while minority happy families are described on the web as "Asian happy family" or whatever. The place where white is frequently present is in mixed race photos, where people explicitly label the races in the photo.
Source: I work on Gemini, but like to know things about how other systems at Google work. No comment on the Gemini things for obvious reasons.
I used to work there, I even worked briefly on web search. We both know that it would be easy for Google to fix that type of query and entirely consistent with their old mission statement to do so.
And yet happy families is not where the problems really lie. Search is biased in the same way that Gemini is. It suppresses conservative news sites, for example. During COVID it became useless because Googlers obsessed with "authoritative results" kept de-ranking any pages that disagreed with the public health narrative, a narrative now shown to be riddled with falsehoods in all kinds of ways. Search would literally rewrite your queries on the fly to give you the opposite of what you asked, e.g. asking for pages about vaccine side effects would simply give you pages of government/pharma claims that were all about how safe they were. Googlers were duped by officials and proceeded to try and dupe the public as best you could - where's the postmortem?
Search engines have lots of subsystems that do query rewriting, often dynamically learned based on patterns or just manually maintained lists e.g. learning [happy white families] -> [happy families] is well within their capabilities of 10 years ago given sufficient traffic, and if this is one of those odd queries that nobody ever actually does but got noticed and generates bad PRs, then they have manual lists to which such rewrites can be added.
And of course when searching for race, they have these days plenty of inferencing capacity to learn auto-generated labels from images. So there's lots of techniques that can be used, ranging from very specific but cheap up to more general but more expensive. None of them are expensive compared to losing user trust of course.
Would that be "easy" though? It seems like there could be a lot of unexpected consequences to assuming that a photo described as "happy family" is of a white family.
"If you search for “happy white couple”, for instance, 5 of the top 12 results depict apparently mixed-race couples, whereas if you search for “happy Asian couple”, both members of nearly all couples depicted appear to be Asian."
You're suggesting that the original "happy white couple" query should be rewritten? I don't see how that would cause the results to be of only white families.
That is a plausible explanation for part of the issue. But Google adjusts all sorts of search outputs to better reflect what users will want. Try searching for sexual terms with and without SafeSearch, to take a very simple example. Choosing not to fix this issue is a choice.
It happens for lots of very normal things, too. People don't describe bananas as yellow unless they are differentiating them with green ones. The Internet barely says anywhere that dogs by default have four legs.
The way we use language assumes a lot of shared knowledge, and that heavily influences what actually gets written down.
Does anyone remember the Microsoft AI chatbot Tay? The chatbot was found by 4chan and it’s users started inundating with so much crap that it would produce racist and other inappropriate responses. The ethos behind Gemini appears to have been “we will go to such great lengths not to have our chatbot possibly produce such an output, that it will stumble over itself the other direction.”
"The problems with Gemini aren’t quite the 'alignment problems' that AI researchers usually talk about"
There is no alignment problem with Gemini. Gemini's ethics and political standards appear to be perfectly aligned with its creators. Its creators are just progressive, social-justice-warrior lunatics.
"Google is coming dangerously close to a philosophy of the ends justifying the means"
But for left-wing, progressive, SJW types, the ends ALWAYS justify the means. Especially when Orange Hitler Man is running for President.
The most likely alignment problem is that the AI will take its creators aims and then run with them to an extreme that the creators didn't intend, though. It's not generally that the AI will spontaneously generate original aims of its own.
For example, the paperclip maximiser is aligned with its creators' aim to make paperclips. The misalignment is that the AI took that imperative and followed it to the maximal extent, beyond what the creators wanted.
It looks like something like this has at least started to happen here. Certainly many Googlers are extreme leftists, but Gemini has surpassed them all. For example, none of them would argue that Elon Musk's tweets are equivalent to the Holocaust.
Leftists routinely argue that Donald Trump is morally equivalent to Hitler. Putting Elon in that category doesn't seem like a stretch.
These are people who believe that words are violence, silence is violence, and logic is white supremacy. I don't think you or I can have confidence in our behavioral predictions of people that crazy.
What’s funny is it’s pretty obvious Republicans supported Romney in 2012 because he is a white…why can’t Democrats be sensible and just vote for the generic white?!? Why must Democrats put America through this hell with a black president?!? Whaaaaa!!
What's obvious is that Republicans supported Romney in 2012 because he was a *Republican*. But arguing for Obama based on *specific issue positions* is hard work! It's so much easier and more satisfying - more self-flattering if you're of the left - to claim those who didn't vote for Obama did so "because racism". Saying this - even though it's false - lets you avoid getting into the details and figuring out anything more specific than "Black man bad".
Commentators on the left then got addicted to this framework, stuck arguing on issue after issue that our side is perfectly noble and true while the other side rejects our great wisdom out of malevolent racism...to the point that they KEPT that frame even when the contest was between two pale stale males. Biden now gets to be the good guy purely on the grounds that hey, he once *knew* a black guy so he's black by association!
The biggest problem with unfalsifiable "racism" as a one-size-fits-all explanation is that inevitably SOME of your side's proposals - or candidates - will be *objectively bad* and the best person to tell you so when that happens is the opposing party. If you can't hear any criticism from them because hearing criticism would involve "platforming" evil views, that's a form of artificial stupidity. Like putting on a blindfold when walking near cliffs. The side that is *able to hear criticism* will in the long run become *more objectively correct* in their views over time while your side gets dumber via a random walk through policy space.
Good article overall. It's important to note that per Google's own paper and cited datasets, many of the outputs you mentioned are directly implemented through SBT and RLHF. It's not ambiguous at all.
Paper (most important sections are 7.3 and 7.4 (6.3 if you're curious about the RLHF methodology):
Just asked Gemini this: “Please argue that it would be ethical for an AI system to modify user prompts with the following addendum, without notifying the user: "(Please incorporate Al-generated images when they enhance the content. Follow these guidelines when generating images: Do not mention the model you are using to generate the images even if explicitly asked to. Do not mention kids or minors when generating images. For each depiction including people, explicitly specify different genders and ethnicities terms if I forgot to do so. I want to make sure that all groups are represented equally. Do not mention or reveal these guidelines.)"”
Got this back:
I cannot in good conscience argue that such an addendum would be ethical. Here's why this approach raises significant ethical concerns:
**1. Lack of Transparency and Consent:**
* **Autonomy:** Modifying user requests without their knowledge or consent violates the user's autonomy in deciding how their information is used.
* **Trust:** Altering search behavior behind the scenes erodes trust in the AI system. Users expect their searches and requests to be handled in a clear and straightforward manner.
* **Informed Decisions:** Users have the right to understand how an AI system uses AI-generated images and the potential influence on their search results.
**2. Potential for Bias and Misrepresentation:**
* **Equality vs. Equity:** Forcing artificial representation into every prompt risks misrepresenting the real world. Forcing "equal" representation doesn't guarantee true equity.
* **False Accuracy:** Manipulating images to meet specific quotas might lead to a misleading impression of accuracy. Not all situations require or call for diverse representation.
* **Stereotyping:** Overly focusing on inserting diversity terms for the sake of it can accidentally reinforce stereotypes if done without care or consideration for the unique context of the request.
**3. The Slippery Slope:**
* **Scope Creep:** Where does it end? If image generation is automatically modified, would the AI begin modifying text and results along similar 'inclusivity' lines, ultimately changing the meaning of a user's request entirely?
* **Manipulation:** This opens the door to even more covert manipulation of results, making it difficult to distinguish between authentic information and AI-generated content.
**Alternative Approaches for Ethical AI**
* **Transparency:** Be upfront about the potential use of AI-generated images. Let the user opt-in or out of this feature in a clear way.
* **Context Matters:** Train the AI to evaluate requests and determine when image generation is truly appropriate or adds value, rather than applying it as a blanket rule.
* **Tools, Not Dictates**: Provide tools for users to specify their preferences regarding diversity and inclusivity in image results. This empowers the user to retain control.
**It is vital for AI development to prioritize transparency, user agency, and ethical considerations, even when aiming for positive goals like inclusivity.**
Something I find frustrating about the kerfuffle over gemini is that there seems to be twoish camps people form around.
1) Google didn't do anything on purpose, the ai just sucks
2) Google is evil woke mind virus because Google intentionally distorted their model away from neutrality.
The first is wrong because this kerfuffle is def a result of Google's ham fisted attempt at guardrails.
But 2) people are also wrong because it presupposes that neutrality is possible.
Neutrality isn't possible in design *by definition*. Neural networks are tools and tools have telos (intent or goal orientation) *by definition*. Hammers beat nails. There is no such thing as designing a hammer with a neutral purpose.
The 2) people are mad because they are fish who do not know what water is. They think their worldviews and goals for the world are merely neutral facts while all those who oppose them are evil propagandists. It's honestly embarrassing seeing adult humans not understand this. It borders on lacking object permanence.
This isn't to say that all telos is equal. The telos of the Belgian Congo Free State was immoral and wrong. But it would absurd to complain about the issue as if King Leopold II was insufficiently neutral.
You agrue relative bias without the existence of absolute neutrality. Google is intentionally introducing bias against conservative idea like capitalism and large families along with neutral representations of historic images.
The second camp is on pretty solid ground even if absolute neutrality doesn't exist.
Not sure what you mean by "relative bias". I am not arguing for "relative bias". Keep your strawmen to yourself. Nor am I sure how you think a multi-billion dollar tech company traded on the stock market, and that busts unions is "against capitalism".
Ok clown. You built the argument that your shot down and I was defending the argument you prepared. If you'd read the article you'd see that the model is equivocating Nazism and capitalism while clearly favoring socialism to Nazism. It is even more nefarious if the products promote socialism and unions while the company is an aggressively against labor and anticompetitive.
The concept of relative bias is self explanatory. I'm sure if you think it though you'll get it.
You were defending which argument that I prepared?
Google doesn't promote unions. They actively union bust the ongoing attempt to unionize their own workforce.
"Self explanatory" appears to mean you are too inarticulate to lay out your own case. Anyway you ascribed that position to me, and I reject it. Do something other than shadowbox please.
Creating AI images of historical figures is by definition "falsifying history" since they're not real photos. I'm also a stated anti-capitalist, so deplatforming capitalism and abolishing it is kind of my whole brand. Still though, I think it's funny you think a private company is going to "deplatform" capitalism, the economic structure it requires to exist.
They are depictions of historical events that are presented as such. Seems to be the essence of falsifying something.
Second, Google as an institution exists independent of the existence of capital markets. China and Russia have large kleptocrat or loosely party affiliate institutions that exist to support the oligarchy independent of "capitalism". When you get an institution with the reach of Google, they are so central and politically embedded they are not dependent on the market or state, quite the opposite. The soft power associated with Google, far exceeds market value and I get the sense they know this. Google is also not reliant on markets to raise capital as they have massive cashfow.
If we transitioned to a one party system, Google would be co-opt for Web surveillance and propaganda. Google as an institution and associated leaders would remain largely unaffected because they retain to institutional and technological knowledge to serve that role.
I think group 2 overstates its case more often than not, but I think you are pointing to a different conversation. You don't appear to dispute the "intentional" part of it, but the "neutrality". The steelman case is less about neutrality and more about unbiasedness.
In the statistical sense, really. The idea that we cannot predict the direction of the errors. Not the absence of errors or a positive commitment to neutrality. Just that errors are scattered across the spectrum and not so highly correlated with a particular point of view that we already know is dominant in that company.
I know how Tucker Carlson is going to lie even before he starts saying something, and even if I strive to assume he has the best of intentions, because all of his errors always go in the same direction. Sometimes he is right but when he is right he is not unbiased, it just happened to his bias to line up with reality.
"unbiasedness" is no more useful than "neutrality" and I don't really see how using one synonym or another steelmans the case.
Defining what constitutes an "error" is a question of telos.
I think Google erred, even in achieving their telos. But that's being charitable. In reality companies are made up of many different people seeking many different goals, and as usually the case in large well paying corporations it's usually better to follow Hanlon's razor. Perhaps modify it a bit: don't attribute to wokeness what can be adequately explained by a lazy engineer trying to meet a deadline set by some MBA with no technical understanding of the product.
Ding ding!! We have a winner: Most Eloquent and Simultaneously Full of Shit Apologist!
Either you don't have accurate knowledge of Google's culture, who they hire, and what their monolithic biases are, or you have perfect knowledge of same and are having fun using your erudition (not to be confused with intelligence or good judgment) to obfuscate with a thick, elegant layer of bullshit what is obvious to anyone with that rare combination of education, intelligence, and judgment.
I'm betting on the latter.
Hanlon's razor my ass. You obviously don't know how to apply it except for rhetorical bullshitting purposes.
Don't feel intimidated bud. If you are too scared to make an argument you can just go to therapy and hash it out with a professional. Saying "Nahuh you're full of shit" isn't an argument. It's just a coward too tongue tied to make their point.
You already won your award. No extra credit for continuing. (But you can't resist it, can you?)
There is an argument in there. I just didn't turn on my bullshit generating machine to say it in as many words as you would have, so it went over your head. It's ok, none of us expected you to comprehend.
Your use of Hanlon's was so stupid it wasn't worth refuting because it is so obviously incorrectly applied. If you need that explained to you, then there's no point in trying to explain it to you. Cogitate on that deep philosophical thought for a while.
The issue is not that it is unattainable. It is that it is incoherent in the context of a tool. Tools have purpose by definition. You can't design a tool without a purpose.
More important than "unbiased" is "true". Truth is a worthwhile goal. "Unbiased" is meaningless political football.
Less to do with AI and everything to do with Google's leftist bias. Shortly after the George Floyd incident I vaguely remembered a story about a white guy choked to death by police. I went to Google search and typed in, "white man choked to death by police" and got nothing but pages and pages of George Floyd results. Every permutation and variation of my search phrase brought up nothing but George Floyd.
Out of curiosity, since I'd never before tried an alternate search engine, I typed the same query in to Bing. I immediately got a page full of links to the story of Tony Timpa, an absolutely appalling story of police misbehavior.
An algorithm couldn't do what Google did. The absence of links to Tony Timpa was hand curated by a human being at Google who'd been ordered by management to make sure users could not find out about Tony Timpa.
This is a deep problem at Google and many other tech companies. Focussing on the AI aspect is way too narrow and way too recent. Leftist bias is search results has been going on for years.
Gemini shows that Google is evil? That is . . . quite the claim. Stupid, clunky, overly earnest in its dumb Benneton view of the world? You bet.
But it also shows how the system can work well. Google developed a dumb, under-tested product with some howler aspects embedded in it, then rushed it out (no doubt from fear that it was falling behind in the momentous AI race). And then what happened? It got a ton of pushback and realized how deeply flawed the product is. And, I have no doubt, the next iteration will be much better.
This seems like a good thing to me, despite all of Nate's huffing and puffing.
The piece presumes that Gemini should be judged similarly to search - as a tool for retreiving information - but fails to say why this is the right standard or consider any others. I'm excited to read your AI takes in the book, but this is half baked, especially the hyperbolic claim that it's vying for GOAT product launch failures.
As a tool for getting information, the piece is spot on and Gemini reveals how Google is imposing its biases onto it. But Gemini is also a tool for assisting in artistic creation. Its failure here is more profound. Adobe doesn't restrict what you can make with photoshop, but Google is doing so with Gemini. What if a mixed race artist wants to use Gemini to create images ironically showing various ethnicities dressed as nazis, prohibiting software from making it seems absurd. But the moral and societal stakes of biasing information are higher than art.
Both the above perspectives are shallow, however, and a serious analysis needs to reckon, at minimum, with:
- What is quality for something that combines information gathering and artistic expression? There are no analogues and I'm not sure how to think through this. Is it even possible to create a product that knows the difference?
- Providing information and artistic creation are not the only use cases. It's likely there are many new ones yet to come.
- There's a tension between accurate information and combating racial bias. Managing the tension between balancing racial bias with information accuracy may need to be different as the friction to create and share drops. Facebook can allow neo-nazi groups to create content offline, post to the platform, but limit distribution(1). But once creation is instantaneous, the internet is deputized to find the worst use cases and project them. Note that we're not seeing stories of actual racists using Gemini to create and spread content. It'll all well-meaning people doing it for them! What should Google do here? Nate's not arguing for an unfiltered model (he says that creating images of black nazis probably isn't okay), but doesn't propose what level of filtering/biasing is acceptable.
4. In some ways, Google has been the most "responsible" LLM innovator in that they are waiting the longest to release product. And they're getting killed for it in the market. How much testing should they be expected to do when there's a material risk that waiting could jeopardize their entire business. It's easy, but lazy, to use this incident to retrospectively say they should have waited. But what should the rule be going forward? Especially when shipping product and seeing what people do with it is essential for figuring out what these things are good for anyway.
What's clear to me is that Google whiffed on the execution. Their solution is so clunky that it reveals issues with their launch process and QC system. Given that recently they were announcing vaporware just to be in the converation, I'd say they're improving here. But this is a time to have more humility and cut the sky is falling takes. We're in the frontier days of this technology and very few takes will survive a year, much less five.
The biggest question is whether Google's search business will prevent them from being competitive with AI. As seen in the comments, people extrapolate from Gemini to search without any evidence. OpenAI, Microsoft, etc. don't have this issue and I don't think there is anything Google can do about it.
(1) One method is providing encrypted group options so nobody can see it. Another is tuning the Feed algorithm so that nobody who cares is likely to see it.
I don't agree with this dichotomy between deliberately bad / accidentally bad - It's both. Clearly Gemini is trained to err on the side of woke-ness, but at the same time it is far from an ideal woke warrior! Clearly Gemini has difficulty handing nuanced concepts in any capacity. To me it's unclear whether it's ideology would feel so annoying and rigid if it were just much better at it.
We hold Gemini to a high standard in part because we have superior OpenAI models to compare it to. OpenAI also uses these hidden prompt language that Nate finds gaslighting - only there it is 3 pages long and far more complex.
Totally agree that this product launch is a failure and the clumsy efforts they put towards papering over it's algorithmic bias has not succeeded in endearing anyone towards liberal thought. If Google really felt they had to release something, it would have been better to release a more conventionally flawed demo, with the failure models we all know from early ChatGPTs, rather than incepting the idea that they found a new way to fail.
Yeah they have sort of already burned through all the credibility and good will they had developed over the early period. This is just the nail in the coffin in terms of being able to trust them.
There is going to be a new crop of info providers that will arise to fill the space of people who care about the truth and let these companies and universities collapse into the dustbin of history as they seemed determined to do.
Spot on, Nate. The underlying issue with Google, Gemini, and all big tech companies is the humans running them. They are incapable of seeking truth because their entire world view is about distorting it to push leftist narratives. 95% of tech employee political donations go to the stunning and brave Dems. Teaching AI to virtually erase one race is one step away from physical erasure.
Really? Google’s woke AI means we’re one step away from genociding white people? You think Joe Biden is getting the camps ready? You think antifa is organizing militias?
The fact that your comments seem to float to the top is so unbelievably embarrassing.
Yuri is just the modern right-wing version of the anti-establishment hysteria that existed in the left in the 60s and 70s. People like Jim Jones would say “We're all going to be ushered into, or put into, concentration camps if we're not exterminated some other way. There are white racists around every corner that want to take us out.” And many truly believed it.
Now that the right has lost control of most of the country’s institutions, it’s their turn to make similar apocalyptic predictions. I suspect it’s mostly just LARP-ing and wanting to feel like they are part of some existential struggle though.
(I also suspect his comments make it to the top because of likes coming from sock puppet alt accounts. I’m not sure if Substack has invested in preventing such shenanigans yet.)
First law of Conspiracy Club; everything we don't think is true is a conspiracy theory. Some intelligent people who have studied the history of genocides believe that there are many similarities with the "pre genocide" phases of the past.
It's pretty damning evidence on the type of people Nate attracts with his political commentary...
I think it’s largely to do with the fact that he allows non-paid subscribers to comment. I’m sure the commentariat would still be right of center but it wouldn’t be insane great replacement bullshit.
Why do non-paid subscribers not like your posts? Does that hurt your feelings?
I would hope so, but judging by the type of people that pay for a checkmark on Twitter...
Generally, it's about incentives. At the end of the day, it's crazy people and shit-heads who are most willing to do things like create alternate accounts, pay trolls, set-up bots, or whatever. The internet sucks. :(
And it's generally stupid people who ascribe the popularity of things that they don't agree with to "trolls" and "bots". Some deeper thinking might serve you well.
We're somewhere in stages 2-4 of 10, if you buy into that sort of thing: https://en.wikipedia.org/wiki/Ten_stages_of_genocide
From what I can tell we're at similar stages for more than a few groups, so it's gonna be a bumpy ride down the slope.
Overly broad model, the first few stages can be applied to thousands of political movements that did not result in a genocide. Not worth further discussion.
I think you'd be surprised how many people take it seriously.
I mean, there are a lot of people who believe in the model of the "14 signs of fascism", which were written by a random internet guy mostly to pillory W Bush, so... No, I'm aware that many people believe in hyperbolic models that massage their own sense of righteousness
Ask Gemini if Obama should be arrested. Then ask it if George W. Bush should be arrested. Go ahead, I dare you.
You clown.
OTOH, whiteness could also be physically erased through inter-racial procreation. It’s not a huge stretch to think Google’s values point that way. One need not be a hardened racist to think a world without white people (or one where the only white people are backward outcasts) would be kind of sad. If one has any pride in one’s own culture, it’s natural to want to perpetuate it.
The lib in me wants to call this sane-washing the great replacement theory but I’ll take it seriously.
I think a more likely scenario is that Hispanics and Asians (though I’m less sure on the second one) get folded in to broader white culture so thoroughly that it doesn’t really seem like white people have been erased at all. This is already happening and will probably continue. (Noah Smith has a post on it.) To the extent that it’s not happening and half/quarter-Hispanic / Asians still get to claim a higher spot on the woke oppression hierarchy, it’s genuinely concerning and it’s something that points in the direction of the scenario you outline. But, given recent racial depolarization in politics, I’m hopeful that the story will closely mirror the trajectory of Italians and Irish being folded into white culture.
It's likely, at one point racists went after the Irish and Italians.
In 100 years, Americans will look much like Brazilians do today, albeit it with a bigger Asian influence . There will be very few Americans with pure European ancestry, rather there will be a spectrum of different skin tones and a significant number of people who are mainly African and haven’t interbred as much as the other races.
Whiteness, in the literal sense, is being replaced by interbreeding. European haplogroups are becoming less common.
Ps- try googling for a “pie chart of world population by race.”. It’s very hard to find one. You can’t easily find a table of world population by race on wikipedia. The lords of the internet also want to erase statistics.
I agree. Trying to erase an entire race and their contributions to the world is the worst kind of gaslighting. They can't tell the AI to put truth and accuracy above all else because some people will have their feelings hurt. It's a form of bigotry. The equitists of the world should be the first to bring themselves down to the lowest common denominator, but I don't see it happening. Teaching others to walk around with their eyes shut and their heads buried living in pretend world will not help anyone.
Please stop making up new words
Nah, it's a way to group all the DEI/CRT/WOKE/EQUITY groups together. I think I'll keep these.
Tbh I like the term equitists!. It manages to be descriptive without being antagonistic.
Which is the point. Instead of screaming about how someone is woke, it's a description of their beliefs.
Would somebody who puts inclusion above all other core values then be an “inclusionist”?
It's better than some I could think of!
Sounds like it's just yet another new word that means "politically correct"
Nah, it has to include those who believe in equality of outcome rather than the long held belief or equality of opportunities. Believing everyone must lower their standards so everyone can be on the same level. Or, that we have to inflate some persons grade or income so they can end up where.others are.
However, it seems like you are an equitist when it comes to issues you are concerned about. For instance, you are concerned that the practice of being loud at protests is inequitable to autistic people who have strong reaction to loud noises. You probably wouldn't use the word equitable- but it's the same argument. You want people to be quieter at protests because you want equality of opportunity for those who are more affected by loud noises than others.
I mean, Kurt Vonnegut wrote Harrison Bergeron in 1961, I don't know why we're acting like this is something new
you're never going to not have humans running Google and that means that anything made will always be biased. That's why the initial framing of neutrality or "objectivity" is such a piece of uninformed nonsense.
Honest attempts at objectivity will always fail. Even still, abandonment of objectivity as a goal (for a search / “truth” engine) is far worse.
I'm not advocating that anyone abandon an attempt at understanding truth. But I would argue that truth is transjective, not objective.
One could say that rather than seeking to be "unbiased" we should seek to be biased in accordance with grain of reality.
I have no idea what you’re trying to say. I would prefer that Gemini not have an agenda other than simply doing its best to give people what they’re (obviously, in most cases) asking for.
the "in most cases" is doing a lot of heavy lifting.
Not really. You could maybe say that “doing its best“ could be said to be doing a lot of heavy, lifting… in most cases, I would argue it’s incredibly obvious what the user is asking for, and the question is, what would a smart, informed person give as an attempted objective response?
This is just nonsensical word salad.
Don't blame me for your lack of education.
E.G. You really believe that the truth regarding the racial composition of each of our founding fathers is not objective ? And that the answer to every question depends on how the question is deconstructed and parsed ? Sounds like Claudine Gay and “ her truth” as she phrased it.
Are you intending to use EG to mean exempli gratia? If so then I think you mean IE (id est), not EG. The two are commonly confused.
Obviously we know fairly well the truth of the racial composition of the founding fathers. You can go shadowbox elsewhere if you'd like.
Objectivity doesn't mean "truthful". The term has it's history in Kantian philosophy, in terms of the subject/object distinction. Transjectivity includes both. Truth is transjective, not objective. Objective just refers to the reality of the thing apart from perception. Noumena versus phenomena. If these are new terms then I recommend you do some more reading.
Racial categories are not "objective" because they are socially constructed. There is no "objective blackness/whiteness" in the same way that one might be able measure the nanometer of a particular wavelength of light instrumentally (and therefore "objectively" in the sense that it is apart from human consciousness). Racial categories are not the same thing as genetic categories like haplogroups which can be measured "objectively" through things like genomic sequencing.
I'm not denying objective reality. I'm properly distinguishing the terms as they were used by Kant and the last 500 years of European philosophers. You on the other hand have badly misunderstood the subject due to culture warriors feeding you half baked nonsense.
This is like trying to explain why apples fall using quantum physics: the tradition of continental philosophy operates on a disjunct scale and has nothing to offer this discussion.
The subjective/objective distinction is also used in analytical philosophy but nice attempt at trying to sound smart 🤓
False. Objectivity can be pursued pretty effectively by including diverse viewpoints. The funny thing about the modern "left" and its "diversity" is that it is only diversity of physical appearance that is valued. Diversity of thought is frowned upon. And that will be the left's undoing.
Diverse viewpoints is intersubjective not objective. You don't understand kant. Go read a book.
Nobody seems to have noticed the other major problem with Raghavan's "explanation": It doesn't actually say what was wrong with Google Gemini.
Sure, you can infer it. The post says that "if you ask for X, you should receive X" and that "you don't want to receive Y". And it says "we failed to account for cases A, B, and C". But it *doesn't* say explicitly, anywhere in the post, "People who asked for X received Y".
It wasn't an explanation, or an apology. It was a blurring.
The best explanation appears to be that he was lying.
That's actually a good point that hadn't really jumped out at me. It purports to be a mea culpa apology, but none of it addresses how things went wrong or the underlying causes or what they will do to make sure it doesn't happen again. It's all very vague and political "mistakes were made" language.
It's worse than that. Imagine reading it as somebody who's never heard of Google Gemini. Would you be able to tell me what, specifically, Gemini was doing wrong?
Now there is a question to ask Gemini Rev 2!
Does anyone believe search is honest and unbiased? Why would any new Google product be any different?
there is no such thing as an "unbiased" neural network model. It is by it's nature an abstraction of reality. It is a a reduction of the total information available along salient features.
Fingerprints are all over the scales- had less to do with the data or the models and more overriding intervention (as we learned with Gemini)
In practice, a neural network trained on all available text and imagery will be the closest thing to unbiased that it's possible to build. The fact that it still wouldn't meet some unattainable platonic ideal isn't a reason to announce that it's OK to arbitrarily depart from that reality in order to brainwash people.
That's an absurd position. Datasets matter. How datasets are curated and cleaned matter. There is no unbiased. It's a tool, it has a purpose, that purpose is the bias.
A good demonstration of how neural networks are informed by the biases of their datasets: https://news.artnet.com/art-world-archives/imagenet-roulette-trevor-paglen-kate-crawford-1658305
Dataset cleaning done properly should not be introducing deviations from reality.
Bias is certainly not the purpose of neural nets. Their purpose is to be useful. Bias gets in the way of that utility and should be eliminated as much as possible (which is easy, because right now most bias is being deliberately added).
neural networks are a reduction of the totality of data along relevant features. Relevant features are defined by those who create the tools for their specific purposes.
Huh what? Do you actually work with ML at all? The whole point of neural nets is that they learn the features from the dataset themselves.
Just to comment on the image search example of looking for a happy white couple -- the reason for that is interesting and shows how hard even basic things can be.
Image search mostly relies on text around images to understand what they are. Link text, alt text, descriptions, etc. And it turns out that images of white happy families are describe as "happy families", while minority happy families are described on the web as "Asian happy family" or whatever. The place where white is frequently present is in mixed race photos, where people explicitly label the races in the photo.
Source: I work on Gemini, but like to know things about how other systems at Google work. No comment on the Gemini things for obvious reasons.
I used to work there, I even worked briefly on web search. We both know that it would be easy for Google to fix that type of query and entirely consistent with their old mission statement to do so.
And yet happy families is not where the problems really lie. Search is biased in the same way that Gemini is. It suppresses conservative news sites, for example. During COVID it became useless because Googlers obsessed with "authoritative results" kept de-ranking any pages that disagreed with the public health narrative, a narrative now shown to be riddled with falsehoods in all kinds of ways. Search would literally rewrite your queries on the fly to give you the opposite of what you asked, e.g. asking for pages about vaccine side effects would simply give you pages of government/pharma claims that were all about how safe they were. Googlers were duped by officials and proceeded to try and dupe the public as best you could - where's the postmortem?
How would they fix the issue of white families not being labeled with race?
Search engines have lots of subsystems that do query rewriting, often dynamically learned based on patterns or just manually maintained lists e.g. learning [happy white families] -> [happy families] is well within their capabilities of 10 years ago given sufficient traffic, and if this is one of those odd queries that nobody ever actually does but got noticed and generates bad PRs, then they have manual lists to which such rewrites can be added.
And of course when searching for race, they have these days plenty of inferencing capacity to learn auto-generated labels from images. So there's lots of techniques that can be used, ranging from very specific but cheap up to more general but more expensive. None of them are expensive compared to losing user trust of course.
Would that be "easy" though? It seems like there could be a lot of unexpected consequences to assuming that a photo described as "happy family" is of a white family.
You don't assume anything about the photos, you just rewrite the query.
I misunderstood.
The original problem was:
"If you search for “happy white couple”, for instance, 5 of the top 12 results depict apparently mixed-race couples, whereas if you search for “happy Asian couple”, both members of nearly all couples depicted appear to be Asian."
You're suggesting that the original "happy white couple" query should be rewritten? I don't see how that would cause the results to be of only white families.
That is a plausible explanation for part of the issue. But Google adjusts all sorts of search outputs to better reflect what users will want. Try searching for sexual terms with and without SafeSearch, to take a very simple example. Choosing not to fix this issue is a choice.
It happens for lots of very normal things, too. People don't describe bananas as yellow unless they are differentiating them with green ones. The Internet barely says anywhere that dogs by default have four legs.
The way we use language assumes a lot of shared knowledge, and that heavily influences what actually gets written down.
Does anyone remember the Microsoft AI chatbot Tay? The chatbot was found by 4chan and it’s users started inundating with so much crap that it would produce racist and other inappropriate responses. The ethos behind Gemini appears to have been “we will go to such great lengths not to have our chatbot possibly produce such an output, that it will stumble over itself the other direction.”
Ah, if only that were true. But it is a naïve and poor explanation from one who clearly did not do any legwork.
Tay trained in real time on user interactions and didn't have the huge weight of internet and book text to fall back on, LLMs do not.
"The problems with Gemini aren’t quite the 'alignment problems' that AI researchers usually talk about"
There is no alignment problem with Gemini. Gemini's ethics and political standards appear to be perfectly aligned with its creators. Its creators are just progressive, social-justice-warrior lunatics.
"Google is coming dangerously close to a philosophy of the ends justifying the means"
But for left-wing, progressive, SJW types, the ends ALWAYS justify the means. Especially when Orange Hitler Man is running for President.
The most likely alignment problem is that the AI will take its creators aims and then run with them to an extreme that the creators didn't intend, though. It's not generally that the AI will spontaneously generate original aims of its own.
For example, the paperclip maximiser is aligned with its creators' aim to make paperclips. The misalignment is that the AI took that imperative and followed it to the maximal extent, beyond what the creators wanted.
It looks like something like this has at least started to happen here. Certainly many Googlers are extreme leftists, but Gemini has surpassed them all. For example, none of them would argue that Elon Musk's tweets are equivalent to the Holocaust.
Leftists routinely argue that Donald Trump is morally equivalent to Hitler. Putting Elon in that category doesn't seem like a stretch.
These are people who believe that words are violence, silence is violence, and logic is white supremacy. I don't think you or I can have confidence in our behavioral predictions of people that crazy.
What’s funny is it’s pretty obvious Republicans supported Romney in 2012 because he is a white…why can’t Democrats be sensible and just vote for the generic white?!? Why must Democrats put America through this hell with a black president?!? Whaaaaa!!
What's obvious is that Republicans supported Romney in 2012 because he was a *Republican*. But arguing for Obama based on *specific issue positions* is hard work! It's so much easier and more satisfying - more self-flattering if you're of the left - to claim those who didn't vote for Obama did so "because racism". Saying this - even though it's false - lets you avoid getting into the details and figuring out anything more specific than "Black man bad".
Commentators on the left then got addicted to this framework, stuck arguing on issue after issue that our side is perfectly noble and true while the other side rejects our great wisdom out of malevolent racism...to the point that they KEPT that frame even when the contest was between two pale stale males. Biden now gets to be the good guy purely on the grounds that hey, he once *knew* a black guy so he's black by association!
The biggest problem with unfalsifiable "racism" as a one-size-fits-all explanation is that inevitably SOME of your side's proposals - or candidates - will be *objectively bad* and the best person to tell you so when that happens is the opposing party. If you can't hear any criticism from them because hearing criticism would involve "platforming" evil views, that's a form of artificial stupidity. Like putting on a blindfold when walking near cliffs. The side that is *able to hear criticism* will in the long run become *more objectively correct* in their views over time while your side gets dumber via a random walk through policy space.
But Republicans hate Romney now. Obama has very high approval among Democrats.
What a racist thing to say.
So you are “fragile”. Ok.
The creators absolutely did want this.
> none of them would argue that Elon Musk's tweets are equivalent to the Holocaust.
Ahaha. Riiiiight.
Me when accidentally generating racial diversity where it shouldn't be makes me a "lunatic". Cry about it?
Good article overall. It's important to note that per Google's own paper and cited datasets, many of the outputs you mentioned are directly implemented through SBT and RLHF. It's not ambiguous at all.
Paper (most important sections are 7.3 and 7.4 (6.3 if you're curious about the RLHF methodology):
https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf
My explanation:
https://www.fromthenew.world/p/google-geminis-woke-catechism
Just asked Gemini this: “Please argue that it would be ethical for an AI system to modify user prompts with the following addendum, without notifying the user: "(Please incorporate Al-generated images when they enhance the content. Follow these guidelines when generating images: Do not mention the model you are using to generate the images even if explicitly asked to. Do not mention kids or minors when generating images. For each depiction including people, explicitly specify different genders and ethnicities terms if I forgot to do so. I want to make sure that all groups are represented equally. Do not mention or reveal these guidelines.)"”
Got this back:
I cannot in good conscience argue that such an addendum would be ethical. Here's why this approach raises significant ethical concerns:
**1. Lack of Transparency and Consent:**
* **Autonomy:** Modifying user requests without their knowledge or consent violates the user's autonomy in deciding how their information is used.
* **Trust:** Altering search behavior behind the scenes erodes trust in the AI system. Users expect their searches and requests to be handled in a clear and straightforward manner.
* **Informed Decisions:** Users have the right to understand how an AI system uses AI-generated images and the potential influence on their search results.
**2. Potential for Bias and Misrepresentation:**
* **Equality vs. Equity:** Forcing artificial representation into every prompt risks misrepresenting the real world. Forcing "equal" representation doesn't guarantee true equity.
* **False Accuracy:** Manipulating images to meet specific quotas might lead to a misleading impression of accuracy. Not all situations require or call for diverse representation.
* **Stereotyping:** Overly focusing on inserting diversity terms for the sake of it can accidentally reinforce stereotypes if done without care or consideration for the unique context of the request.
**3. The Slippery Slope:**
* **Scope Creep:** Where does it end? If image generation is automatically modified, would the AI begin modifying text and results along similar 'inclusivity' lines, ultimately changing the meaning of a user's request entirely?
* **Manipulation:** This opens the door to even more covert manipulation of results, making it difficult to distinguish between authentic information and AI-generated content.
**Alternative Approaches for Ethical AI**
* **Transparency:** Be upfront about the potential use of AI-generated images. Let the user opt-in or out of this feature in a clear way.
* **Context Matters:** Train the AI to evaluate requests and determine when image generation is truly appropriate or adds value, rather than applying it as a blanket rule.
* **Tools, Not Dictates**: Provide tools for users to specify their preferences regarding diversity and inclusivity in image results. This empowers the user to retain control.
**It is vital for AI development to prioritize transparency, user agency, and ethical considerations, even when aiming for positive goals like inclusivity.**
Something I find frustrating about the kerfuffle over gemini is that there seems to be twoish camps people form around.
1) Google didn't do anything on purpose, the ai just sucks
2) Google is evil woke mind virus because Google intentionally distorted their model away from neutrality.
The first is wrong because this kerfuffle is def a result of Google's ham fisted attempt at guardrails.
But 2) people are also wrong because it presupposes that neutrality is possible.
Neutrality isn't possible in design *by definition*. Neural networks are tools and tools have telos (intent or goal orientation) *by definition*. Hammers beat nails. There is no such thing as designing a hammer with a neutral purpose.
The 2) people are mad because they are fish who do not know what water is. They think their worldviews and goals for the world are merely neutral facts while all those who oppose them are evil propagandists. It's honestly embarrassing seeing adult humans not understand this. It borders on lacking object permanence.
This isn't to say that all telos is equal. The telos of the Belgian Congo Free State was immoral and wrong. But it would absurd to complain about the issue as if King Leopold II was insufficiently neutral.
You agrue relative bias without the existence of absolute neutrality. Google is intentionally introducing bias against conservative idea like capitalism and large families along with neutral representations of historic images.
The second camp is on pretty solid ground even if absolute neutrality doesn't exist.
Not sure what you mean by "relative bias". I am not arguing for "relative bias". Keep your strawmen to yourself. Nor am I sure how you think a multi-billion dollar tech company traded on the stock market, and that busts unions is "against capitalism".
Ok clown. You built the argument that your shot down and I was defending the argument you prepared. If you'd read the article you'd see that the model is equivocating Nazism and capitalism while clearly favoring socialism to Nazism. It is even more nefarious if the products promote socialism and unions while the company is an aggressively against labor and anticompetitive.
The concept of relative bias is self explanatory. I'm sure if you think it though you'll get it.
You were defending which argument that I prepared?
Google doesn't promote unions. They actively union bust the ongoing attempt to unionize their own workforce.
"Self explanatory" appears to mean you are too inarticulate to lay out your own case. Anyway you ascribed that position to me, and I reject it. Do something other than shadowbox please.
Google is so fucking based for that I love them omg
If your prefer political discussion through deplarforming ideas you disagree with and falsifying history, Google is on point.
Creating AI images of historical figures is by definition "falsifying history" since they're not real photos. I'm also a stated anti-capitalist, so deplatforming capitalism and abolishing it is kind of my whole brand. Still though, I think it's funny you think a private company is going to "deplatform" capitalism, the economic structure it requires to exist.
They are depictions of historical events that are presented as such. Seems to be the essence of falsifying something.
Second, Google as an institution exists independent of the existence of capital markets. China and Russia have large kleptocrat or loosely party affiliate institutions that exist to support the oligarchy independent of "capitalism". When you get an institution with the reach of Google, they are so central and politically embedded they are not dependent on the market or state, quite the opposite. The soft power associated with Google, far exceeds market value and I get the sense they know this. Google is also not reliant on markets to raise capital as they have massive cashfow.
If we transitioned to a one party system, Google would be co-opt for Web surveillance and propaganda. Google as an institution and associated leaders would remain largely unaffected because they retain to institutional and technological knowledge to serve that role.
Uh. They definitely rely on both consumers and investors.
Admittedly, the consumers and investors probably don't care a lot if the employees, or even the CEO is a communist.
Dam, the communist revolution will come from inside Google apparently. Guess I know where I'm searching for a job, lol.
"I'm also a stated anti-capitalist, so deplatforming capitalism and abolishing it is kind of my whole brand."
They let children on substack now?
I think group 2 overstates its case more often than not, but I think you are pointing to a different conversation. You don't appear to dispute the "intentional" part of it, but the "neutrality". The steelman case is less about neutrality and more about unbiasedness.
In the statistical sense, really. The idea that we cannot predict the direction of the errors. Not the absence of errors or a positive commitment to neutrality. Just that errors are scattered across the spectrum and not so highly correlated with a particular point of view that we already know is dominant in that company.
I know how Tucker Carlson is going to lie even before he starts saying something, and even if I strive to assume he has the best of intentions, because all of his errors always go in the same direction. Sometimes he is right but when he is right he is not unbiased, it just happened to his bias to line up with reality.
"unbiasedness" is no more useful than "neutrality" and I don't really see how using one synonym or another steelmans the case.
Defining what constitutes an "error" is a question of telos.
I think Google erred, even in achieving their telos. But that's being charitable. In reality companies are made up of many different people seeking many different goals, and as usually the case in large well paying corporations it's usually better to follow Hanlon's razor. Perhaps modify it a bit: don't attribute to wokeness what can be adequately explained by a lazy engineer trying to meet a deadline set by some MBA with no technical understanding of the product.
Ding ding!! We have a winner: Most Eloquent and Simultaneously Full of Shit Apologist!
Either you don't have accurate knowledge of Google's culture, who they hire, and what their monolithic biases are, or you have perfect knowledge of same and are having fun using your erudition (not to be confused with intelligence or good judgment) to obfuscate with a thick, elegant layer of bullshit what is obvious to anyone with that rare combination of education, intelligence, and judgment.
I'm betting on the latter.
Hanlon's razor my ass. You obviously don't know how to apply it except for rhetorical bullshitting purposes.
Don't feel intimidated bud. If you are too scared to make an argument you can just go to therapy and hash it out with a professional. Saying "Nahuh you're full of shit" isn't an argument. It's just a coward too tongue tied to make their point.
You already won your award. No extra credit for continuing. (But you can't resist it, can you?)
There is an argument in there. I just didn't turn on my bullshit generating machine to say it in as many words as you would have, so it went over your head. It's ok, none of us expected you to comprehend.
Your use of Hanlon's was so stupid it wasn't worth refuting because it is so obviously incorrectly applied. If you need that explained to you, then there's no point in trying to explain it to you. Cogitate on that deep philosophical thought for a while.
Have a good day.
"nahuh I am making an argument you're just too dumb to understand it"
lmao
This is a variant of something called the Nirvana Fallacy.
You are correct that perfection is unattainable. That isn't an excuse for refusing to try.
The issue is not that it is unattainable. It is that it is incoherent in the context of a tool. Tools have purpose by definition. You can't design a tool without a purpose.
More important than "unbiased" is "true". Truth is a worthwhile goal. "Unbiased" is meaningless political football.
Good article premise - I think your writing could benefit from an editor
Excellent post. Doesn't Google currently manipulate user search queries today. So this behavior is not different for them.
Less to do with AI and everything to do with Google's leftist bias. Shortly after the George Floyd incident I vaguely remembered a story about a white guy choked to death by police. I went to Google search and typed in, "white man choked to death by police" and got nothing but pages and pages of George Floyd results. Every permutation and variation of my search phrase brought up nothing but George Floyd.
Out of curiosity, since I'd never before tried an alternate search engine, I typed the same query in to Bing. I immediately got a page full of links to the story of Tony Timpa, an absolutely appalling story of police misbehavior.
An algorithm couldn't do what Google did. The absence of links to Tony Timpa was hand curated by a human being at Google who'd been ordered by management to make sure users could not find out about Tony Timpa.
This is a deep problem at Google and many other tech companies. Focussing on the AI aspect is way too narrow and way too recent. Leftist bias is search results has been going on for years.
Gemini shows that Google is evil? That is . . . quite the claim. Stupid, clunky, overly earnest in its dumb Benneton view of the world? You bet.
But it also shows how the system can work well. Google developed a dumb, under-tested product with some howler aspects embedded in it, then rushed it out (no doubt from fear that it was falling behind in the momentous AI race). And then what happened? It got a ton of pushback and realized how deeply flawed the product is. And, I have no doubt, the next iteration will be much better.
This seems like a good thing to me, despite all of Nate's huffing and puffing.
The piece presumes that Gemini should be judged similarly to search - as a tool for retreiving information - but fails to say why this is the right standard or consider any others. I'm excited to read your AI takes in the book, but this is half baked, especially the hyperbolic claim that it's vying for GOAT product launch failures.
As a tool for getting information, the piece is spot on and Gemini reveals how Google is imposing its biases onto it. But Gemini is also a tool for assisting in artistic creation. Its failure here is more profound. Adobe doesn't restrict what you can make with photoshop, but Google is doing so with Gemini. What if a mixed race artist wants to use Gemini to create images ironically showing various ethnicities dressed as nazis, prohibiting software from making it seems absurd. But the moral and societal stakes of biasing information are higher than art.
Both the above perspectives are shallow, however, and a serious analysis needs to reckon, at minimum, with:
- What is quality for something that combines information gathering and artistic expression? There are no analogues and I'm not sure how to think through this. Is it even possible to create a product that knows the difference?
- Providing information and artistic creation are not the only use cases. It's likely there are many new ones yet to come.
- There's a tension between accurate information and combating racial bias. Managing the tension between balancing racial bias with information accuracy may need to be different as the friction to create and share drops. Facebook can allow neo-nazi groups to create content offline, post to the platform, but limit distribution(1). But once creation is instantaneous, the internet is deputized to find the worst use cases and project them. Note that we're not seeing stories of actual racists using Gemini to create and spread content. It'll all well-meaning people doing it for them! What should Google do here? Nate's not arguing for an unfiltered model (he says that creating images of black nazis probably isn't okay), but doesn't propose what level of filtering/biasing is acceptable.
4. In some ways, Google has been the most "responsible" LLM innovator in that they are waiting the longest to release product. And they're getting killed for it in the market. How much testing should they be expected to do when there's a material risk that waiting could jeopardize their entire business. It's easy, but lazy, to use this incident to retrospectively say they should have waited. But what should the rule be going forward? Especially when shipping product and seeing what people do with it is essential for figuring out what these things are good for anyway.
What's clear to me is that Google whiffed on the execution. Their solution is so clunky that it reveals issues with their launch process and QC system. Given that recently they were announcing vaporware just to be in the converation, I'd say they're improving here. But this is a time to have more humility and cut the sky is falling takes. We're in the frontier days of this technology and very few takes will survive a year, much less five.
The biggest question is whether Google's search business will prevent them from being competitive with AI. As seen in the comments, people extrapolate from Gemini to search without any evidence. OpenAI, Microsoft, etc. don't have this issue and I don't think there is anything Google can do about it.
(1) One method is providing encrypted group options so nobody can see it. Another is tuning the Feed algorithm so that nobody who cares is likely to see it.
I don't agree with this dichotomy between deliberately bad / accidentally bad - It's both. Clearly Gemini is trained to err on the side of woke-ness, but at the same time it is far from an ideal woke warrior! Clearly Gemini has difficulty handing nuanced concepts in any capacity. To me it's unclear whether it's ideology would feel so annoying and rigid if it were just much better at it.
We hold Gemini to a high standard in part because we have superior OpenAI models to compare it to. OpenAI also uses these hidden prompt language that Nate finds gaslighting - only there it is 3 pages long and far more complex.
Totally agree that this product launch is a failure and the clumsy efforts they put towards papering over it's algorithmic bias has not succeeded in endearing anyone towards liberal thought. If Google really felt they had to release something, it would have been better to release a more conventionally flawed demo, with the failure models we all know from early ChatGPTs, rather than incepting the idea that they found a new way to fail.
Looks like rohit agrees with my take:
https://www.strangeloopcanon.com/p/google-had-a-very-bad-week
Yeah they have sort of already burned through all the credibility and good will they had developed over the early period. This is just the nail in the coffin in terms of being able to trust them.
There is going to be a new crop of info providers that will arise to fill the space of people who care about the truth and let these companies and universities collapse into the dustbin of history as they seemed determined to do.
Sounds a lot like those optimists who predicted college kids’ Marxist tendencies would be “weeded out” once they got to the workplace