I know I made a big deal a couple of weeks ago about not being cynical in this newsletter. What I’m about to say is NOT intended to be cynical, just sort of bemused.
If you were blessed not to be paying attention over the past few days, Google has been rolling out their generative AI thing called Gemini in their own uniquely inscrutable way.
Per Tom Gara: “To be clear, Gemini for Google Workspace, formerly known as Duet AI for Google Workspace, is not to be confused with Gemini Business, which lets organizations use generative AI in Workspace at a lower price than Gemini Enterprise, which replaces Duet AI for Workspace Enterprise.”
Then there’s the former Google Bard, which is also now known as Gemini, unless you want to pay for Gemini Advanced, which will set you up with a Google One AI Premium Plan.
Keep in mind that not a single one of these things existed twelve months ago.
After reading that a couple of times, and then rinsing out your eyeholes, you might be forgiven for believing that Google might not have a coherent internal vision for what it wants to do with AI.
And yet over the weekend, a bunch of people became really upset at Google because Gemini apparently is “woke” with “deranged values”.
Gemini’s deranged values include hallucinating nonwhite-male 17th-century physicists and suggesting that you’ll have to make up your own mind whether Elon Musk is worse than Hitler. I’d like to say nobody was actually harmed by this, but Musk did get really mad about it and publicly attacked the Google PM he thinks is responsible, which had to have made that poor guy’s terrible week 100x worse. Come on, Elon.
Clearly it’s fun for some people to imagine that Google is this evil cabal of woke leftists determined to rewrite history with AI propaganda. I think if this were true, it would be hard to explain why Google immediately acknowledged the issues, disabled the offending features, and swore to do better.
Here’s my read on Google’s culture as someone who was inside it until last month: it is full of smart people who bear no malicious agenda against anyone, but live in deathly fear of screwing up a good thing. This is what happens when you spend 25 years building a brand as the world’s most desirable place to work. An awful lot of people have set a life goal to work at that place, and when you get to the top there’s nowhere to go but down. So people at Google tend to be small-c conservative, reactive, and risk-averse to a degree that is quite common at large companies in general but feels weird for a tech company.
A surefire way to shake Google out of their Ads-induced cozy slumber and get them to actually ship something is to publicly embarrass them. In general you can explain an odd Google decision by asking yourself: “Who are they afraid of today?”
People are saying we’ve fallen behind on generative AI? We’d better rush to market with all the AI things, as fast as possible!
AI safety advocates are warning that AI can perpetuate unconscious biases in the training set? We’d better massage the model so the Biden White House doesn’t make us do the walk of shame.
Mean blue-checks on Twitter are making fun of us because our AI now does transparently silly things? Take it offline until we can figure out how to evade their adversarial prompts.
The gloomy outcome here would not be that Google gets away with their “woke agenda”, but that Gemini is pushed and tugged by criticism from all sides until it becomes sort of a shapeless pudding that is afraid to give a definitive answer about anything. I hope that doesn’t happen. The tech landscape is healthier when Google has a viable stake in the AI game, and—yes—a point of view to go along with it.
The funniest possible outcome would be that Google just decides that the Gemini brand name is now tarnished and renames all their AI stuff all over again. Personally I vote for “Google Assistant”, not to be confused with Google’s other virtual assistant Google Assistant, an evolution of the product that started as Google Voice Search and was later relaunched as Google Now.
A couple of links
My grand piano and I will be doing a truly ridiculous musical talk at SCaLE on the evening of Friday, March 15th. I’ve whipped up some brand-new music for the occasion and it should be a great time. If you’re in the LA area, I do hope you’ll stop by; the conference is a very affordable ticket and the team does a great job of putting it together every year.
The Kubernetes Resume Challenge kicks off in one week - see you there!
Why did so many major players decide they needed a single umbrella GenAI “brand” that uses the same name to label a bunch of almost completely separate products?
Seems to me like Google made a conscious decision to try and be proactively diverse/inclusive which meant that they became horribly historically inaccurate - something that would have just worked if they hadn't messed with the model in the first place. If they had done 5 min of testing they would have picked up the issue and wouldn't have to un-launch something that they only launched a few days ago!