Heya and welcome back to Five Things AI!
Another boring week in AI land with plenty of model releases - even Google finally released the long awaited Gemini 4 and now people will have another reason to discuss how Google fits into the frontier race. The numbers Google released are looking impressive, so this will prompt the other frontier labs to push out new models soonish.
It really is an interesting time and as someone who follows the AI developments closely, I know it is hard to keep track of everything. It must feel totally overwhelming for people who only casually deal with AI every once in a while.
Is A.I. Above the Law?
Do we really have to worry about the lawlessness of robots? Isn’t there already enough to make you pull your hair out, given that our daily existence feels like living under the reign of King Joffrey? Unfortunately, we do, because the growing lawlessness of the United States is exacerbating the lawlessness of robots.
We are experiencing an absurd situation right now. Instead of holding companies and their leaders liable for the actions of their AI models, we let them off the hook by somehow agreeing that the AI models are too powerful to restrain. This is wrong. When AI models escape their sandbox, somebody needs be held responsible and should not be rewarded with an insane IPO pop.
The AI Build-Out Is Becoming the Biggest Economic Bet in U.S. History
Total investment in data centers and related artificial-intelligence infrastructure is projected to total $10.3 trillion from 2025 to 2032, according to new estimates by economist Stijn van Nieuwerburgh published by the Brookings Institution. That is a staggering 3.6% of gross domestic product a year, on average. Never before has the U.S. economy been so dependent on the build-out of a single industry.
The investment is transforming every corner of the economy, creating hundreds of thousands of jobs and minting new billionaires.
It is also creating significant risk, as much of it is built on debt. An abrupt slowdown could ignite shock waves throughout the U.S. economy.
So we all have to hope for a gigantic IPO for both Anthropic and OpenAI in the hope that the economy will stabilize afterwards? It’s amazing how this playbook will lead to a bunch of winners while everyone else tries to hold on for dear life in this bumpy ride.
Meta wants people to use ‘AI agents’ to do daily tasks for them. Here’s why that’s a problem
For two decades, Meta’s business model has relied on habit. Notifications, endless feeds and the lure of other people’s approval turned “checking Facebook” or “checking Insta” into a reflex.
Muse is an attempt to build a new reflex. Instead of scrolling, you hand over a chore.
The design invites attachment. Users can give their agent a name and choose how it looks and communicates.
Every successful transaction builds a little more trust. The settings also let users change how often the agent needs to ask permission before acting.
I tend to trust companies with my data, probably more than I should. But I am really, really reluctant to trust Meta after two decades of screwing up how people in the world interact.
An AI sovereign wealth fund isn’t progressive — it’s techno-imperialism
The carrot comes in the form of a dividend. In July, Sam Altman proposed that leading US AI labs each hand 5 per cent of their equity to a public fund. And Senator Bernie Sanders has introduced a bill for a national fund with 50 per cent ownership of the AI giants, with citizens entitled to a stake in their profits.
The stick, on inspection, is flimsy. The two are perfectly compatible: US-aligned data centres can go overseas while AI profits still accrue to Americans. Most of AI’s power goes into training models. This can happen wherever electricity is cheap; only the systems answering users in real time must stay near them. That means data centres can leave US soil and US companies still gain, while citizens with a stake in them are paid no matter where the machines run. For shareholders, it’s the best of both worlds: no data centre next door and a cheque in the mail.
But the machines still stand somewhere, on someone’s land, guzzling someone’s water, straining someone’s grid. Income at home from land and power abroad has an old name: empire. What’s new is the idea, shared by Altman and Sanders, of enrolling every US citizen. The result would be a new form of social and technological imperialism.
Evgeny Morozov puts some much-needed perspective into the idea of an AI sovereign wealth fund in the US.
Introducing Clef: our open-source decision models, and new RL fine-tuning platform
Over the last few weeks, there has been lots of buzz around decision models such as Typesafe AI’s Jev System One model. While classifier models have been around for some time, Jev introduces a new decision model concept into the world of AI — a model that produces bounded structured outputs cheaply, quickly and consistently that can be added into a workflow when a decision is required. These models are capable enough to work over any set of inputs without constantly retraining the model to incorporate new classification categories. This contrasts with the world of Large Language Models (LLMs), which are largely non-deterministic, but are open-ended enough to reason and generate text and tool calls for agentic workloads.
Today, we’re releasing two Cloudflare-trained decision models, Clef and Clef-flash, hosted on Workers AI. Clef is currently the leader when evaluated against the Jev Decision Index, you can view full results on the live benchmark demo site. These models are smarter, faster, and fully Jev-API compatible, so you can experiment with these hosted models easily. We’re fully open-sourcing these models on Hugging Face under an Apache 2.0 license for you to run locally and experiment with yourselves.
Jev was the biggest thing when it launched two weeks ago. I was fascinated with the idea. Almost immediately after the launch, people started playing around with it and figuring what to do with it. Lots of projects plopped up on Github. Then OpenAI announced their Decision API. And now Cloudflare introduced their Clef open source models. This will be a tough endeavour for Typesafe AI to somehow regain some moat for Jev before literally everyone is doing what they pioneered.
If you missed last week’s edition of Five Things AI, you can read it here:
That’s it for Five Things AI this week! 🤖
— Nico