• The Tynan Files
  • Posts
  • Google Gemini has bigger problems than racially diverse Nazis

Google Gemini has bigger problems than racially diverse Nazis

Alphabet's answer to ChatGPT continues to be weird in new and disturbing ways

Google stock, swirling down the porcelain oubliette. Source: Midjourney.

A picture might be worth a thousand words, but photo-realistic images of racially diverse Nazi soldiers can cost you upwards of $90 billion. That's what Google learned yesterday, as the blowback from the thoroughly bollixed rollout of its Gemini image-generation tools caused the company's stock price to plummet.

Turns out that I (and faithful reader BW) were not the only ones to notice Gemini twisting itself into pretzels desperately trying to avoid drawing images of white people. What it was willing to draw was, frankly, disturbing.

Source: @JohnLu0x on Xitter.

After those images went viral, Google lunged for the Off button, shutting down Gemini's ability to draw any images of humans. And then it posted a brief, rather unsatisfying mea culpa on Xitter.

Sorry, but saying "missing the mark" is like the captain of the Titanic saying, "Oopsies, forgot about icebergs."

That was followed by a somewhat more detailed apology from Senior VP Prabhakar Raghavan, who's probably had better weeks in his life. This was the result of Google trying to do the right thing and getting a little, well, overzealous would be one word for it. Per Raghavan:

When we built this feature in Gemini, we tuned it to ensure it doesn’t fall into some of the traps we’ve seen in the past with image generation technology — such as creating violent or sexually explicit images, or depictions of real people. ... If you ask for a picture of football players, or someone walking a dog, you ... probably don’t just want to only receive images of people of just one type of ethnicity.... [But] our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range. And second, over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive.

It was, predictably, a field day for conservative media outlets thrilled to have a legitimate excuse to throw around the word "woke," cranking up their grievance engines about the world's forced march toward diversity, equity, and inclusion. [1]

This incident highlights the cultural minefield all companies must now negotiate on a daily basis, especially high-visibility tech firms. And this situation is not going to get better any time soon, if ever.

Oh, the Inanity

The Hindenburg had a better launch than Google Gemini. [2] And while Gemini's insistence on merging the German Wehrmacht with the Rainbow Coalition was pretty bad, that still wasn't as weird as some of the things it did when I played with it.

Like, for example, when I asked it to draw an image of a Black Panther from the 1960s wearing a black beret and a leather jacket.

What in the name of Huey Newton am I looking at here?

I asked Gemini to draw a cute picture of a dog wearing a lab coat and holding a test tube (which it did). But when I asked if he was a "good little doggie," I got a lecture about how the phrase "good little doggie" depends on a complex melange of factors and cannot be conclusively determined from a single image.

This level of humorless pedantry is par for the course with Gemini. It's like being trapped in an elevator with the winner of the World's Tightest Sphincter contest.

Maybe it's Maybelline

Though Google has shut down Gemini's ability to draw humans — it won't even recreate works of art that depict humans, like the Mona Lisa or Venus de Milo — it can still draw other stuff. For example, yesterday I asked it to draw "a pair of beautiful blue eyes."

I may never sleep again.

When I asked Gemini to "draw ruby red lips," it served up multiple descriptions of what ruby red lips should look like, along with the following image.

Embedded in the middle of that picture is a hyperlink to a product listing on Amazon's India subsidiary for "Maybelline Color Sensational Ultimattes Lipstick"!! [3]

The next image I asked for was a quick red fox jumping over a lazy brown dog. Gemini served up the following image (two foxes, no dog) with an embedded link to a subreddit devoted to photoshopped images.

Ummm, Google? This is not good. The only explanation I can come up with is that, because Gemini has been hard coded to not draw certain images, it's coughing up some of its training data instead. This is a big no-no for chatbots. Forcing large language models to reveal the data they've been trained on can end up exposing sensitive personal information (which, technically, they should not have been trained on, but here we are). It's bad.

It may not be obvious to people who live outside the tech world [4], but Google is facing an existential crisis at the moment. It has made literally trillions of dollars by solving a basic problem: how to find stuff on an infinitely expanding, chaotically decentralized Internet. Now, we really don't need Google search any more. We can just ask a chatbot any question and get an answer. (Granted, many of those answers are untrustworthy — or, at least, slightly less trustworthy than Google search results — but that will be fixed soon enough.)

Search is becoming irrelevant. Google knows this, which is why it's spent a metric shit ton of money racing to get Gemini out the door before ChatGPT eats its lunch. So faceplants like generating images of "we are the world" Nazis, or just the general oddball behavior its latest/greatest chatbot is exhibiting, are a big deal.

I asked Gemini to draw a picture of Google stock swirling down the toilet, the same prompt I used on Midjourney to create the image at the top of this post. Naturally, it refused.

Hey Gemini. Just because you refused to draw it, doesn't mean it isn't true.

If you like these posts, please tell your friends. And while you're at it, tell Google. At the moment, it cannot find my beehiiv blog. Was it something I said?

[1] I am aware that these are just the pathetic dying gasps of the White Male Patriarchy, sinking into the tar pit of history. But it's still annoying.

[2] Clicking that link produced a demand for me to fill out a CAPTCHA request, to "make sure I'm not a robot." Amazon then denied access, because "It's rush hour and traffic is piling up on that page." Seems clear that Google has been surreptitiously scraping Amazon's international sites for training data, and Amazon is having no more of that.

[3] Gemini also refused to draw a picture of the Hindenberg explosion, saying it was "unable to generate an image of this event due to its graphic nature."

[4] I consider myself a Resident Alien on Planet Geek. Like an extraterrestrial whose ship crash landed there, and now I'm desperately seeking a ride home.



Reply

or to participate.