Google rules the AI roost
Welcome to Cautious Optimism, a newsletter on tech, business and power.
Monday. Good morning from the chilly Northeast, where we’re reprising the fall we didn’t get and delaying the winter we’re supposed to receive.
📈 Trending Up: Memory prices … the Indian IPO market … crypto taxes … the end of Russian aggression? … good news … the Chinese property crunch … layoffs
📉 Trending Down: Russian oil prices … DOGE … GLP-1s as the everything drug … Apple News … Russian oil revenue
Things That Matter
How about some good news? I try not to harp on about it, but my status as a recovering addict (curse you, alcohol) makes me one of the few people in my social group who has gone to rehab. I am, therefore, often uniquely able to comment on the experience: it is dull.
Rehab is one of the very few options people are offered when an addiction takes over their lives, however. It’s mostly AA in a skinsuit, if you were curious, though there are some variations on the theme if you hunt around. It doesn’t work very well, data indicates, and I can attest to that after seeing people cycle right back into rehab during my two-week stint.
So, what’s the good news? GLP-1s could turn the entire world of addiction care on its head. This is the headline that I was waiting for:
Friends with dependencies and addictions of various types have told me when they started using GLP-1s for weight loss, they found that whatever was crouched on their shoulder started weighing less; drinking lost some of its zing; insert drug of choice here became less interesting.
If you have spent any time in AA, you will understand that the idea of using medication to help defeat the addiction monster sounds nigh-sacrilegious. So what? We can’t let the religious fervor of the AA Kingdom hold back real progress; hell, the old-school AA types are still opposed to standard psychiatric care.
The data is starting to accumulate on how impactful GLP-1s can be at helping curb addiction. Is it a panacea? Doubtful. Can it be a useful tool in our sparse anti-addition quiver? Heck yes.
What’s up with AI content licensing? Despite various wide-ranging efforts, getting AI companies to pay for the data they need to power their tools and services remains difficult. I hope that Really Simple Licensing, Cloudflare’s Pay Per Crawl and other offerings succeed because I think it will lead to better AI.
So when Microsoft started working on what it calls the ‘Publisher Content Marketplace,’ my ears perked up. Microsoft owns a piece of OpenAI and Anthropic, and its cloud services underpin a good chunk of the digital economy. It has heft, in other words.
Critically, Axios reported that “Microsoft’s Copilot assistant will serve as the first AI buyer within the marketplace for publishers to sell their content.” That means Business Insider is now picking up nickels and dimes from Redmond. It’s good to see public note of progress on this front. More, please.
Google is the global AI leader
Last week’s deluge of new AI models demonstrated that improvements to model quality are still possible. What’s more, several AI labs can push the boundary forward simultaneously.
While xAI, Google and OpenAI released new models in recent days (Grok 4.1, Grok 4.1 Fast; Gemini 3 Pro preview; and GPT-5.1 Pro and GPT-5.1-Codex-Max, respectively), the Mountain View crew came out on top.
Gemini 3 Pro is crushing across the normal run of online benchmarks we lean on to determine model performance:
#1 overall on LMArena, coming first in text, webdev, vision, text-to-image (via Gemini 3 Pro Image preview, aka Nano Banana Pro), and image editing.
#1 on the Artificial Analysis Intelligence Index. Two of OpenAI’s models came in second and third; Moonshot AI’s Kimi K2 Thinking secured #4, and Grok 4 and 4.1 Fast rounded out the group at fifth and sixth.
#1 on Humanity’s Last Exam, a “multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage.”
Perhaps even more importantly, Gemini 3 seems to have made people excited about AI again. This tells us that some of the previous AI hype was predicated on future intelligence gains. Examples abound:
Reddit users on AI-focused forums went nuts, calling Gemini 3 “mindblowing,” citing its progress as evidence that the ‘AI wall’ doesn’t exist, and sharing its thinking traces for fun.
X users are sharing a never-ending scroll of projects and ideas they’ve made with Gemini 3, including websites, apps, games and other arcana like blueprints-to-3D-model converters.
Business leaders are crowing in public. Companies like Figma are impressed, and Box is singing praises.
But the following comment from Salesforce CEO Marc Benioff is probably the single loudest piece of commentary from the entire weekend:
You can read the linked WSJ story here, in case you haven’t yet. It’s an overview of the positive sentiment surrounding Gemini 3, so sharing the piece while praising the model while comparing it with ChatGPT is akin to kicking sand into Sam Altman’s cereal.
The fact that Benioff is willing to do so indicates that OpenAI’s grip on the AI market, weakening in recent years as other companies caught up, has reached a new low in terms of forearm grit.
I’m happy. Not that I want OpenAI to lose; I don’t care who wins and therefore do not care who loses the AI rush. What’s great news for you and I is that Google just pantsed OpenAI in public at its own game.
Google runs the default search engine, has the bigger compute footprint, makes the most-used browser, a very popular email service, and is now developing the best AI model (at least for now). That leaves OpenAI leading what, precisely? Sure it owns the consumer AI market today, but even that is at risk, per the WSJ:
[I]n August, the debut of Nano Banana launched Gemini usage on its fastest trajectory to date. Gemini’s 650 million monthly users jumped from 450 million in July.
I wonder if OpenAI would rather occupy the #2 spot in an AI market that has regained some of its mojo, or first place in an AI market that was teetering under hype? No matter what Altman would choose, we’re all living in option one land.