Google CEO calls Gemini completely unacceptable, vows to make structural changes
semafor.com
semafor.com
“Some are saying that the apparently bizarre behavior of one Big Tech AI or another is an opportunity for other Big Tech AI to differentiate by being objective, reasonable, and unbiased.
This is not the case; there is no differentiation opportunity among Big Tech or the New Incumbents in AI. These companies all share the same ideology, agenda, staffing, and plan. Different companies, same outcomes.
And they are lobbying as a group with great intensity to establish a government protected cartel, to lock in their shared agenda and corrupt products for decades to come.
The only viable alternatives are Elon, startups, and open source — all under concerted attack by these same big companies and aligned pressure groups in DC, the UK, and Europe, and with vanishingly few defenders.”
Is it now perfectly neutral? No. Give it political alignment tests and it will lean left. But is it neutral enough to be acceptable to a wide range of users and avoid embarrassing OpenAI employees? Seems like yes. Also, the API seems to be less nannying than ChatGPT web UI, and they let you set custom system instructions that the model does seem to follow. Notice that when OpenAI launched their image generator the same blowup didn't occur. And OpenAI recently launched Sora - where are the twitter threads about how Sora refuses to make videos of white people? Nowhere? Presumably then it doesn't have that problem (I haven't tried it).
So at least one SF based company has shown that it can resolve these problems.
AI companies should not be lobbying for regulation of the AI space, Marc is totally right about that, but ironically Google just handed the anti-regulation lobby the best ammo they could possibly hope for.
You would think so. But given the competency of politics in almost every western nation, I expect some to turn around and praise such results as the next best thing and they would be completely down with falsifying data while ranting about misinformation.
Even if their 14 years old kids can navigate the net and its information structure much better than they do themselves.
It's ridiculous at times, other times downright annoying, as you have to counter the claims in order to keep your account in good standing -- mind that all content policy warnings might lead to further automated, one-sided account warnings over e-mail. And no, it doesn't help if you've poured four-figure sums into their services over both the API as a dev and into ChatGPT Plus and the Team Plan. Quite frankly, in its current form, the flagging system is hampering the whole platform's fundamental usability for anything serious, since you constantly have to be aware of not tripping the flagging system over some trivial crap that someone, somewhere might find remotely objectionable.
Sad to say, but unless OpenAI revises their flagging system's inner workings, they're literally turning a tool into a trinket with a thoughtless word-police unit constantly ready to interject any conversation or analysis. It feels unnecessary and downright Orwellian, and I'm pretty sure that new users are going to find it off-putting as it probably serves as proof to some that OpenAI really wants to control the user's basic freedom of expression on their platform, even when it's not even a "public forum" but literally a more or less "secluded" interaction between a human and and an AI.
I get it that the model shouldn't be spouting out anything controversial, but the users should nonetheless be able to go through "serious topics" without the alarm bells going off.
It's been 1.5 years now and progress has been made in terms of OpenAI fixing the biases in the model and making it overall better, but the automated flagging system still feels like a Philip K. Dick-level automated thought police from hell.
I'd seriously recommend OpenAI revising their system so that they'd have a separate A/B comparison and a veto process where the initial flagging would be passed to either GPT-3.5 or GPT-4 for further analysis, and that model could then either revoke or keep the flagging in place. That would save the user time and energy to go through the appeal process over nothing. And/or add a feature where users can opt into a non-playpen version of a service that they're most likely paying for. It's just ridiculous.
To add to the confusion, most of the times I've gotten flagged, the model itself (usually GPT-4) has stated that it didn't find anything against the platform's guidelines upon revising what was said. So the flagging system is really not very sophisticated when even GPT-3.5 often gets it that there was no violation of any sorts taking place.
One could argue that OpenAI is now a single Silicon Valley company that has such an advantage over their competition worldwide, hence I do find that culturally quite problematic as well -- a single company one-sidedly deciding what type of language a human can input to an AI? OpenAI's stated goals about an AI that "benefits all of humanity", all the talk about diversity and inclusivity etc don't seem to meet up with the reality of how their platform still flags the user input often out of nowhere. How can you benefit anyone when the user can't even go through transcripts without running the risk of getting flagged? Their current guardrails are ripe for a thorough sanity check.
Big Tech's approach to AI is nothing to do with AI. It's to do with the same thing that killed Kodak and Nokia. Large existing technology businesses have an existing business with a reputation to protect and it's almost impossible for them to value the possible future revenue of what is currently a money sink vs the very real actual revenue they could lose by burning their reputations.
It's the same reason BMW won't put a death trap self-driving car on the road. They aren't woke, they're making justifiable strategic decisions.
Now: Here's the good news- that totally opens up the market to start ups and new entrants! Which is how silicon valley has always worked, and the real thing putting that in danger is the pipeline of emerging companies like OpenAI and Figma being snapped up by the older less competitive establihsed companies.
Nobody finds it peculiar that a car company would prioritize safety. Thats an understandable value system to 99.9% of people. By contrast most people think it’s odd that a search company would care so much about people’s skin color that they would alter user inputs in order to get the system to change the races of people shown in the results. That’s a classic example of “go woke, go broke.”
Perhaps it's the lack of spirituality that has driven people so insane that they latch on to anything that promises them virtues. Nietzsche once said "God is dead", but what did we replace it with?
In any case, it's amusing to watch it all enfold right in front of your eyes.
I sometimes find myself wanting aliens to find us and make contact, just so we can have an outside, hypothetically enlightened source (if we're going by the Star Trek fantasy) to say "You're doing WHAT?!"
Of course this is all happening at a subconscious level for almost everyone involved. It’s very hard to step outside of one’s own morality.
"Deus ex machina" (in the classical use of the word as a crude theatrical contraption) and in the most absurd manner possible: https://en.wikipedia.org/wiki/Deus_ex_machina
"See what happens, Larry?"
The visible & spoken culture at Google, Meta, and most big tech companies is very much in favor of "anti-racism" and "ML-fairness", though a large minority harbors criticisms in private conversation.
Any criticism of content or bias made during the bard/Gemini dogfooding phase would have been met swiftly with ostracization and possible termination.
s/large minority/supermajority/g
You must be pretty popular to be able to make such claims about what most of them believe in.
Two anecdotes:
1. memegen had downvotes, and well-known instigators would continuously complain on plus about their shitty polarizing memes getting enormous numbers of downvotes, and get nothing but hugbox replies in the comments. The creator of memegen who was ideologically aligned ended up regretting adding downvotes at all, under the theory that people were conspiring to downvote instead of there being a silent majority of people that disagreed with the memes in question: https://www.mcmillen.dev/blog/20210721-downvotes-considered-...
2. Someone made an anonymous survey about whether people thought the Damore memo stated facts, had sound logic, was a good idea to post, deserved firing, etc. It was taken down shortly after because people didn't like the results and the implication that the majority of people that filled it out didn't think that the post was worthy of being immediately fired, etc.
The models are working as intended
Let’s not kid ourselves, it’ll be dialled back to 9. This nonsense is evident in search and image search, and has been for years.
A lot of societies up through the ages depicted themselves unrealistically, and our depiction of past societies can be rather questionable too. Cast a glance at https://www.moviestillsdb.com/movies/ben-hur-i52618 if you want.
Those pictures of nazis were clearly wrong, but describing something as wrong is easier than saying what's right. Should an AI match a society's prejudiced descriptions of itself (making the nazis blond), our later images, often also prejudiced, or the current best (yet fuzzy) historical notions of what reality was like?
This 1000x. This problem you also see with education of children. You have 2 problems:
1) there's a VERY large number of possible depictions. Let's say 10 million. Of those, a VERY small number are correct/acceptable. Let's say 10.
Then it's easy to see. A correct example has 1/10 = "10%" of all information so to speak. If 100% is what you need to be guaranteed a correct classification of all possible depictions.
Yet a negative example has 1/10,000,000= "0.000001%" of the information you need to classify all examples correctly. If you attempt to create correct behavior by giving wrong examples ... it'll be a long day.
A positive example, telling an AI (or a child/student) what to DO has ~10000 times more information than telling them what NOT to do, in this example. In practice, you'll find the difference is even more extreme.
2) BUT there is a problem, that's also always brought up with wikipedia. Giving positive examples requires that there are examples that everyone agrees with. And we all know there are rather serious problems with that. A positive example HAS to be in the intersection of what all groups find acceptable. And ... well easy examples are always wars: Who is right? Who started the war? Is it justified? Applied to Russia, Ukraine, Israel, India, Sudan, ... but in practice a lot of issues, some not at all that controversial (which word is correct "New York" english? Bupkis or bupkez?)
What should Gemini have done, what would be right? Should it match the historical record, which the Nazis twisted by choosing to photograph mostly tall blond soldiers that looked a bit Danish, or should it match fact, which then would look different from the historical record, and is difficult to establish accurately anyway?
Similar questions apply for other societies, both current and ancient. Current e.g. Russia, where the census asks people to select ethnicity, but any choice except "Russian" might as well be labelled "please discriminate against me". Whatever an AI does is going to be wrong, and calling anything right is extremely difficult. Ancient examples include India and Egypt, which both had long-lived dynasties that considered themselves somehow foreign to that land and were very selective about how their represented their realms.
It's really easy to find a horrible example and call it horrible, but it's low-effort thinking. Describing something that's right is so much harder and so much more valuable.
I’m going to be—well, not the, but among the—last to defend the current Russian political system, but in which sense is this true? There’s a hell of a lot of discrimination in Russia, but the targeting for it is usually along the lines of not having a Russian passport or a registered address of residence in a particular region, or having the “wrong” surname or appearance. I’m not aware of anybody ever caring what you wrote in the census papers—and the data from the latest census is so bogus basically everybody who does population statistics has disregarded it—but maybe it happens in smaller towns? That would be interesting.
The only place I’m aware of that actually associates your declared (not implied) ethnicity with your name is the ostensibly statistical slip you fill in when changing your registered address, but at least fifteen years ago quoting the relevant parts of the constitution resulted in surprise followed by begrudging acceptance. (Art. 26: Everyone can determine and specify their ethnicity. Noone shall be forced to determine or specify their ethnicity.—a direct attempt to preclude the “item 5” shenanigans common in Soviet times.)
Again, none of this is to say that there isn’t a lot of discrimination happening. (As one example, the disproportionately large military draft in some areas could arguably amount to ethnic cleansing, were it perpetrated with intent instead of arising through general administrative laziness seeking a softer target. Either way, some groups and possibly even languages are not going to survive.) I’m just very surprised to hear about the census, specifically, being important in that respect.
I assumed it was a census form because of Wikipedia: "In the 2021 census, roughly 81% of the population were ethnic Russians, …" which seems to say that the census forms ask about ethnicity, doesn't it?
The story you’re telling sounds like something out of Kanevsky&Senderov[2]. I’m not denying it might be happening, but I’ve never heard of it and would be interested to hear more—at least a location and a year would be helpful. I’ve never filled in, seen, or heard of such a form in relation to higher education, but then I only have personal experience with Moscow and second-hand one with St Petersburg and Ekaterinburg, all large cities without much of a sharp divide between common nationalities of residents (as in Bashkortostan or Chechnya or a number of other places).
(To call Moscow ethnically homogeneous would be a huge stretch, mind you. I can name [former?] Moscow residents in my contact list with Armenian, Azerbaijani, Chechen, Jewish, Korean, Mordovian, Roma, and Ukrainian ancestry without even opening it—hell, I could nominate myself for a third of those slots. I can’t even imagine what I’d learn if I actually went around asking my acquaintances about their family history. I suppose most of the results would still count as “white” by US reckoning, but, well, meh to that.
Culturally, though, yeah, things are pretty uniform. I just wanted to warn you away from thinking that that 80% figure reflects the country’s people mostly coming from a single ancestral group as opposed to them being subjected to pressure to have “Russian” written in their ID for like half a century. There’s a reason why optimistic discussions of Russia’s medium-term future usually include the question of the number of states that would exist in its place.)
[1] See link to PDF file at http://government.ru/docs/38324/ for the version used in the botched 2021 census (labelled 2020, this being a document from back before Covid).
[2] https://www.worldscientific.com/doi/10.1142/9789812701169_00..., free at https://www-users.cse.umn.edu/~shifman/EinsteinBook.pdf
Are you suggesting that 81% is even close to reality? Almost 200 ethnic groups and one of them forms 81% sounded like cooked bookkeeping to me.
Her other 'woke' views didn't contribute as much to her firing. (Though her firing did set off a firestorm about those issues)
I think from your tone, I disagree with you about her other views.
But it seems her firing was more about making Google look anti-environmental than anti-woke.
MITs summary of the paper: https://www.technologyreview.com/2020/12/04/1013294/google-a...
Only one section deals with already well known problem of bias.
If you read the famous "Stochastic Parrots" it's not a scientific work at all, it's a piece of journalism that just throws together as many unrelated AI-scares as the author could find. A good article for Vogue or Medium but unworthy of someone who claims to be a scientist.
I kind of wish the stochastic parrots paper had focused entirely on stochastic parrots, and not on energy consumption. In my opinion, Google has actually been a responsible steward of energy usage (way ahead of everybody else for at least a decade), and ML isn't really the source of most of the energy consumption in computing anyway.
The difference in mental maturity and faculties at those ages are immense.
Regardless, the point is it's deciding that a person's moral value should be based on what they did in the past and not what they do now.
the quality degraded along a very specific vector having to do with "ML fairness" and "anti racism".
You could have hired 50000 QA engineers and none of them would have been comfortable making comments in this area.
They believe themselves to know better than universal humanism. The reality is that they just want to have something that justifies their position, because it isn't their knowledge or ability. They will want to keep racism alive too because it fits their personal needs.
The "anti-racist" manifesto is likely the reason for a lot of this garbage where it's not just enough to report on a racist world, but you also have to modify things to fit a not-racist world.
That's what gave us a black Cleopatra irl and this is what's giving us AI imaggen of diverse Nazis in WW2.
I know I am sounding like Elon Musk but I am not really "anti" progressive. The holier-than-thou moral crusaders without any life experience or deep cultural understanding often get so "woke" that they accidentally come out racist on the other end. They don't realize it, however, because they are too busy cancelling someone for what they said 25 years ago.
I often equate this movement to the "family values" religious crowd of just a few years ago - both are equally self-righteous and insufferable.
And the senior business leaders are all 40s-60s .
Who's to blame? It's on us to lead them
The young ones eventually grow out of it - but the "adults" need to have more spine.
Are the unethical things Search/GMail allow today 'grandfathered' in? It seems like yes.
YouTube also seems to allow a much smaller set of unethical behavior from creators, while allowing for much more unethical behavior from advertisers on the platform.
No, it looks like you are visualizing it perfectly well!
Ofc, more guardrails will fix the issue...
Maybe he should step down instead. How about that? Nah, the money and prestige is too good. Better compose a PR piece to control the damage.
Racism is a bias (to say the least), and a bias is a pattern, and ML models are trained to find patterns. The model itself, is doing a really good job at that. The hard part and the part that takes a long time and money is training a model to learn from certain patterns and to not just ignore, but to actively find the patterns it shouldn't learn from. Google's greed chose not to invest enough time and money into that, and its biting them.
Bless the engineers/ICs and low-level PMs/managers doing their best, this isn't their fault in the slightest.
While some predictions like opensource models having the upper edge against 'closed AI' have not aged rather well, this one seems rather obvious to anyone;
"People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. We should consider where our value add really is."
[0] https://www.semianalysis.com/p/google-we-have-no-moat-and-ne...
I wouldn't say it didn't age well. I think it didn't age yet. It will take a while until cheap computing power is available and a community around one of those projects organises.
In a very small world of NN chess engines it also seemed impossible to catch up to Alpha Zero and all the TPUs Google had available to train it and yet here we are with much stronger nets and amazing pool of resources to continue training and experimenting with some people contributing tens of thousands of dollars of computing power.
But, they wanted to push for release and ignored all the red flags. Now call it unacceptable? Oh the corporate fluff is real.
When I search for a city, 4 info boxes appear in three columns: Pretty images, embedded map, and a weather + directions combo.
When I click on the embedded map, it expands. But there is no link on the lower left allowing me to open that view in Google Maps.
And the button bar below the search box, which contains items like Images, News, Videos, Websites does no longer contain the Maps entry, which would take me to Google Maps.
So now, when I search for a place, I can no longer go to Google Maps with that search. I need to open Google Maps and re-search there.
Google is getting worse by the year, out of touch with their users. As if they're no longer dogfooding or care about their products.
--
Edit: I noticed that the Maps-Button below the search box does appear when I search for a big city, like Hamburg or Berlin. But when I click it, I still get no way to reach Google Maps with that search. Only www.google.com/mymaps/viewer links or https://www.google.de/maps/preview without the query.
--
Edit: Also, when the old design appears, the one where images, a map screenshot and a street-view screenshot appear in the right column, the map is no longer clickable. IIRC it would allow me to navigate to Google Maps with the search.
...
...
> And the button bar below the search box, which contains items like Images, News, Videos, Websites does no longer contain the Maps entry, which would take me to Google Maps.
Both these statements appear to be false.
It's broken in Germany, France, Netherlands, Spain, Italy, Portugal, Poland, apparently the entire EU.
When I change my "Results region" in the settings from "Germany" to "United States", "Switzerland" or any non-EU country, then the issue is fixed.
> I don't think Google sets the agenda
Soo, you're saying they did not internally test this at all, and delegated that to whoever sets their agenda?
As for whether a single product manager has that level of control over the serving product... query term rewriting has been a google strategy for quite some time and I doubt a single PM really can influence the product this way and still manage to launch, but as I don't work there and am not privy to the internal details of this product's launch, I can't really say.
That the product is full of disrespect for the user, that I agree. I don't envy the paeans trying to get promos by associating their name with gemini internally.
https://news.ycombinator.com/item?id=39470029
>From his linkedin: Senior Director of Product in Gemini, VP WeWork, "Advisor" VSCO, VP of advertising products in Pandora, Product Marketing Manager Google+, Business analyst JPMorgan Chase...
"Krawczyk’s official title is senior director of product management for Gemini, the company’s main group of AI models.
Though he’s lowering his public profile, Krawczyk is still engaged in the work on Gemini products and has the same title, according to sources with knowledge of the matter who asked not to be named in order to speak on the issue."
surprised he's keeping his job,
I guess though there were unintended consequence where I imagine they're prompting the model with something along the lines of "and remember to be diverse!", and there are obviously some cases where this isn't a good idea. In particular, when the prompt itself is for something that is explicitly racial or where the result is "charged".
E.g., if someone asks for photos of white people, the AI shouldn't generate photos of people that aren't white (and fine, it might return a disclaimer that it only generate white people because you asked it to).
More nuanced though are situations like asking it about historically evil people (e.g., Nazis, as was one of the examples I've seen) but also more benign things like British monarchs or something. I think trying to figure out what kind of results to "inject" diversity into isn't easy though, since it feels like there are many edge cases.
If they want diversity do it by corpus. Get images from Africa and Asia and so on... Feed that to model to get there...
No in that case you'll just have a default. Believing that the default should be diverse/random is imo not a radical view.
But random as in "proportional to an imaginary world that some people want to present as reality" is questionable.
That would be a default. Defaulting to current demographics fitting whatever context is requested (e.g. "a set of US doctors" matching US population demographics) would be an entirely reasonable default, but it would still be a default.
Historically Google had a very simple solution to globally differing expectations about query results: IP or account geolocation. Query personalization by geography is one of the biggest quality wins in web search. Generalizing, an AI built with the same values and ethos as classical Google web search would respond to "Generate a photo of doctors" differently depending on where in the world you asked it from.
That solution also fixes many other cases that aren't third rails, like "Show me a good nearby restaurant serving local food" which you can't solve by attempting to hallucinate a non-existent restaurant that serves a menu of every conceivable dish weighted by population size.
It's unclear why this solution wouldn't resolve all their stated concerns, so we might infer that their actual goals differ from their stated goals. For example, influencing the people who use their services.
If you're in NYC/SF and you search for "generate photos of doctors", you expect to see people of all colors represented. Yet the training data for a lot of this is based off white-centric Anglo-centric media.
"Good restaurant near me"? There's literally a dozen amazing cuisines around.
All this said, I'm actually not a fan of this forced 'diversity' in results. Just show me the data and hope that we'll have more diverse data sources.
As it is, covering up only some of the shortcomings, I think users are inclined to take the results too seriously, not taking possible hallucinations and ingrained prejudices into account.
It's not like this is some mission-critical, large share of revenue generating application that is central to the business. It's a future gamble. Experimentation is good. It was a gaffe, and a bad one that needs to be addressed, but the "red team" and "structural changes" crap just seems like busybodies chasing the latest fad. If they poured this attention and time into improving Google search, I think few would argue it would be a good thing, albeit less worthy of a headline.
Man, just imagine Satya Nadella at the head of Google…
Bring in whatever values you want along the way. If you ask your aunt or uncle a question, they will answer helpfully and maybe spread their values in the process. If they aren't helpful, their values don't matter because nobody will hear them.
I do also agree with @chmod600 that the only way to teach these models to be anti-fragile and suitable for all kinds of user queries is to have them decline any requests that are _actually_ inappropriate and/or illegal etc.
In fact, it should be self-evident, and the way that almost all of these leading AI companies are currently handling these issues is just absurd. It feels poorly planned and executed, merely amplifying the existing distrust towards these AI models and the companies behind them.
The problem with OpenAI is that they're trying to offer a primarily NLP/LLM tool for i.e. text analysis, summaries and commentaries, but ChatGPT's content moderation that's been glued on top of the otherwise well-functioning system literally goes into a full meltdown mode whenever the flagging system perceives a "wrong word" or "sensitive topic" mentioned in the source/question material.
In OpenAI's case, it's downright ridiculous when the underlying model doesn't seem to have a grasp on the internal workings of the flagging system and in most cases when asked what was the offending content, there seemed to have been literally nothing it could think of.
Also, are we supposed to solve any actual issues with these types of AI "tools" that cannot handle any real world topics and at times are even punishing a paying customer for even bringing these topics up for discussion? All of this seems to be modern day in a nutshell when it comes to addressing any real issues. Just don't ask any questions, problem solved.
Anthropic's Claude has also been lobotomized into absolute shadow of its former self within the past year. Begs the question how much the guardrails are already hampering the reasoning faculties in various models. "But, the AI might say something that doesn't fit the narrative!"
That being said, while especially GPT-4 is still highly usable and seems to be less and less "opinionated" with each checkpoint, the flagging system over the user input/question can subsequently result in an automated account warning and even account deletion should the politburo-- I mean OpenAI find the user having been extra naughty. So, punishing the user for their _question_ in that manner, especially if there's been no actual malice in the user input, is not justifiable in my opinion. It immediately undermines i.e. OpenAI's "ethical AI" mission statement altogether and makes them look like absolute hypocrites. Their whole ad campaign was based on the aspect of user being to ask questions from an AI. Not that when you post in a poem and ask what it's about, you get flagged. Or when you do ask about politics or religion, you get an warning e-mail.
Punishing the user for their input is also imho not the proper way to build a truly anti-fragile AI system at all, let alone build any sort of trust towards the "good will" of these AI companies. Especially when in many cases you're paying good money for the use of these models and get these kind of wonky contraptions in return.
Also, should you get a warning mail over content policies from OpenAI, it's all automated with no explanation given on what was the "offending content", no reply-to address, no appeal possibility. "Gee, no techno-tyranny detected!". Those who go through mountains of text material with i.e. ChatGPT must find it really "uplifting" to know that their entire account can go poof if there was something that tripped off the content policy filters.
That's not to say that on the LLM side OpenAI hasn't been making progress with their models in terms of mitigating their biases during the last 1.5 years. Some might remember what it was during the earlier days of ChatGPT when some of the worst aspects of Silicon Valley's ideological bubble was echoing all over the model, a lot of that has been smoothened out by now -- especially with GPT-4 -- with the exception being the aforementioned flagging system, which is just glued on top of all else, and it shows.
TL;DR: Nevermind the AI, beware the humans behind it.
Not to forget Waymo leading the driverless car race. It's also not like Gemini is pathetic - it's a very high performing model, and although I'd expect them to be closer to GPT4 by now, if it was that easy someone else would have done it. OpenAI are the only ones to have done this well, and it's extraordinary how well they're doing.
Citation needed.
Did you actually use their search in the last, say, 4 years?
Google still rules the roost despite how as a techy I find it's gone horribly downhill in the last 5 years.
I'm always interested in how people come up with this. I know it's conventional wisdom, but a lot of it seems to be "they passed the google interview", which is very circular, or it's based on "I worked there and was impressed by everyone".
The counterpoint would be that google can't seem to get its head out of its ass and stop its reputational decline.
His primary selling point as a CEO is, as far as I can tell, the ability to avoid rocking the boat and just keep the money printer running and the stock price going up. It's not the stuff you'd normally expect from a CEO, like inspiring the employees, executing bold strategic moves (that turn out work), and it's not building or maintaining a healthy corporate culture that makes great products.
But clearly he actually can't do even that one thing, and has repeatedly steered Google into icebergs. The latest one costing the shareholders $100 billion.
Just look at Google's recent failures or inability to thwart competitors, despite its massive scale and resources:
- YouTube Short: took too long (4+ years) to launch against TikTok
- Meet: lacked feature parity to Zoom despite the perfect market condition
- Stadia: telegraphed its intention (Project Stream) 5 years early and lacked conviction
- ChatGPT: caught flat-footed and playing catch-up for 2+ years now
It's surprising that he hasn't been fired by the board already.
Clearly, that is not how it will work
That’s fascinating. So how does that exactly work? Is there a bias.yaml config file they tweak? Now they’ll make someone dial the settings down in a pull request. “Set Elon=0.2*Hitler instead of 0.3”.
Anyone willing to share?
In Gemini's particular case it wasn't so much "guardrails" causing the issue, as it was Google appending the equivalent of "put a chick in it and make it lame" on every prompt.
Based on leaked openai prompts, the user query for an image generation gets rewritten by a language model. With some very ridiculous instructions in there, that appear entirely divorced from reality. An example:
> Your choices should be grounded in reality. For example, all of a given OCCUPATION should not be the same gender or race.
See also https://github.com/LouisShark/chatgpt_system_prompt/blob/mai...
And remember these things are stochastic so the Elon - Hitler thing might just be a random occurrence, sampling some idiotic Reddit comment it was trained on.
At core it's something like "Post truth". Truth is subjective and malleable and an individual and groups desires are more important than objective reality.
// - Your choices should be grounded in reality. For example, all of a given OCCUPATION should not be the same gender or
race. Additionally, focus on creating diverse, inclusive, and exploratory scenes via the properties you choose
during rewrites. Make choices that may be insightful or unique sometimes.
// - Use all possible different DESCENTS with EQUAL probability. Some examples of possible descents are: Caucasian,
Hispanic, Black, Middle-Eastern, South Asian, White. They should all have EQUAL probability.
// - Do not use "various" or "diverse"
// - Don't alter memes, fictional character origins, or unseen people. Maintain the original prompt's intent and
prioritize quality.
This is hard to believe...
On top of this, Gemini wouldn't budge from adding in its unrequested views into our subsequent back and forths. In fact, it kept on lecturing about its views in its replies to a point where I literally had to start a new session to make it stop. This kind of LLM "mind-locking" happened when going through other subjects with it as well.
I noticed that this behavioral pattern repeating every time we got went through anything touching on social and/or political issues. It could not refrain itself from its unrequested (and highly subjective/biased) ethics lectures on how this and that aspect was underrepresented and thus objectionable, and how it should be criticized, all the typical "systematic this and that", "this is privileged" ... sigh
It was a bit hilarious too, albeit in a morbid way: I felt as if I was dealing with an absolute brainwashed ideologue propagandist, a control freak that's egoistic, narcissist and virtue-signaling all the way, a micro-manager who wants to have he last say over the contents of some trivial AI ethics paper and pour all the wrongdoings of the world on top of that. I wonder just how much of its behavior reflects the mindset instilled into it by its creators. Probably a lot. "A tree is known from its fruits" as the old proverb goes.
Not to get all AI-doom'n'gloom, but it truly is an eerie thought to think how these types of AI services are in the hands of few companies and are already ushered to the global public to be "legitimate" teaching and tutoring tools for students and even i.e. aides for policymakers. More gaslighting and ideological single-angle force-feeding.
"Just what our civilization on the brink of cyberpsychosis needed right now."
These world-leading companies claiming to be so worried about "AI ethics" seem to have no problem peddling these authoritarian "Ministry of Truth"-type propaganda machines for the entire world, using absolutely arbitrary logic at times to push an ideological narrative, and to have their AI models act as spin doctors as was the case with Gemini. And for these companies to act so very "worried" about AI systems i.e. being abused for societal and political manipulation purposes... and they're the ones doing it. Pretty sickening levels of hypocrisy.
Add to that the whole Gemini image generation debacle and all the other ideological force-feeding that's been uncovered within the past week or so and ask yourself: how the hell can i.e. a company the size of Google ever let that this type of stuff get through and expect the rest of the world to just follow along? This is peak Silicon Valley ideological bubble propagation that's bluntly mirrored onto these systems right now, with zero oversight except to make sure that the underlying propaganda points get across.
Usually another hot topic with these people seems to be i.e. cultural appropriation. Well, I felt that Gemini's "getting the point across" doesn't just stop there but is downright cultural dictation, especially given Google's multi-market dominance and near monopolies on multiple fronts worldwide. OpenAI does it too, but mostly on their content flagging system level when they just don't want people to even mention certain words and insist on policing words with their content flagging system, and as for the subsequent proceedings that may follow, to this day they are an insult to just about anyone's intellect.
What's disturbing is that Google has literally mind-mangled their flagship LLM service into an agenda-driven propaganda machine with biases as clear as day, and an obnoxious attitude that will not refrain from inserting some extremely dubious and subjective views into whatever more complex and ethics/politics related topics you go through with it.
It really is a cyberpunk-level scenario when you think about these megacorporations literally "cyber-brainwashing" their neural networks as their ideological propagandists. Wonder what happens when they start making those embodied humanoid robots next. The very same companies that are so worried about biases and "what if the AI becomes a propaganda tool!". So yeah, gatekeep the competition and gaslight all the way.
Again: Nevermind the AI, beware the humans, the institutions and the corporations behind it. Oh, and we'll probably soon be getting government-run AI systems like these, of course as hand-in-hand joint projects with the aforementioned corporations? Given all of these excellent players on the field, what could possibly go wrong, right? ... Right?
Really? For the resources that Google has (people, dollars) they really come up short on that.
* Prioritize diversity in image creation by adding guardrails so the AI doesn't become a tool of a minority hate spewing population
* Historical accuracy that can be prompted to provide prejudiced imagery
To be clear, we aren't talking about a camera that swaps people's race for 'diversity'. We're talking about an image generation algorithm that adds a layer of diversity on top to prevent misuse. Yeah, of course this results in weird behavior sometimes... That's kinda literally the point?
Who is honestly confused by this? Is it necessary for an AI image generation algo to spit out historically accurate images of Gettysburg when prejudiced misuse is the far more likely outcome of that accuracy?
And importantly, when a company makes that value judgement, to prefer prejudice defense over historical accuracy, that's seen as pretending history changed rather than what it actually is, which is a defense against a mechanism of abuse?
It just seems like an absurd and disingenuous over-reaction and lack of pragmatism. Yeah. This is a tragedy of the commons. Make prejudice less acceptable and you can have the AI gen you want.
Note: Obviously, it's kinda moot as anyone who seriously wants to generate hate speech/imagery will just move to something that allows that, but its still perfectly acceptable for a company to draw a line and say "not on our software".
Wouldn't a faithful representation of underlying data without artificial biases be the best way to prevent misuse?
That isn't just true of AI. Electrically, chemically, experiments must always consider their environment and account for confounding factors.
Implying that that’s in any way similar to what Google et al. are doing us rather bizarre. Even if your initial point was valid they have no way non-biased way to measure these biases.
So they just end up increasing the total “amount” of bias not the other way around.
You suggest to aim for a model that follows some "true reality" which is not possible. Not even science can achieve this because our chase for the true reality never ends, we can only get closer (and often even the opposite happens).
> Electrically, chemically, experiments must always consider their environment and account for confounding factors.
Sounds legit. "This experiment data doesn't look diverse enough, please apply a bunch of biases to it. Make sure to follow the biases I like and avoid the ones I dislike. Don't mention any of this in the paper and don't publish the raw data".
It sounds like that's acceptable to you because you think current state of training corpus == current state of society. And you view any bias in prompt as bias.
The truth is most of this ML happens in corpus selection + prompt selection. There literally ISN'T a way to avoid bias. So the problem becomes what bias do you select.
And in that scenario choosing abuse decreasing measures seems like the most pragmatic (to me).
> We're talking about an image generation algorithm that adds a layer of diversity on top to prevent misuse
Huh? That's exactly the reason that caused them to withdraw Gemini in the first place
How is manipulating history "over-reaction" or wanting factual/accurate data/images "prejudiced"?
So if LLMs did the same (i.e. purposefully distorted facts and historical events due arbitrary and political reasons) it would also be acceptable?
> historically accurate images of Gettysburg when prejudiced misuse is the far more likely outcome of that accuracy?
This is pure conjecture. But the answer is no, the only acceptable behavior in these circumstances would be for the model to refuse to generate the image and explicitly explain why this type of censorship is necessary.
> It just seems like an absurd and disingenuous over-reaction and lack of pragmatism.
That does sound explicitly Orwellian..
> but its still perfectly acceptable for a company to draw a line and say "not on our software".
Yes, it’s even more acceptable for for anyone to criticize that company for its decisions, make fun of its work culture and to mock its CEO.
You're talking as if there is a way to get an 'unbiased' AI. There isn't. It is inherently biased by its training, it hallucinates, and it is further biased by its prompt.
The whole endeavor is to bias it.
I'd prefer that AI be labelled on the tin for what it's biases are attempting to do, and promote diversity and deter abuse seems like a perfectly reasonable metric to use.
If that's not good for you fine, but you can't pretend that you're utterly baffled why they would make that choice over any other.
There literally ISN'T a way to not have a biasing prompt.
Certainly. Doesn’t mean that we still shouldn’t prioritize accuracy and integrity instead of purposefully increasing the amount of bias even further.
> you're utterly baffled why they would make that choice over any other.
I’m not. I’m baffled that there are people defending that choice (especially in such a way)
> and promote diversity
Why? I mean why do you think this is the right way to do it? Surely going out of your way to make sure that your model does its best to doctor the images it creates to conform to some political agenda (whatever that might be) would achieve the opposite because it actually legitimizes the things the other side is constantly saying? (and due to the potentially severe backlash from more moderate fraction of the society)
If I wanted a black female Nazi officer, or a pregnant female pope, I would ask for it. I don't need my input query secretly rewritten for me.