In partnership with
Welcome to OHMO AI. Nvidia's chief told a summit that AI safety is for engineers, not lawmakers, then broadcast a phone call from the president to the room. Meta now charges for more of its AI.
Anthropic modelled AI's effect on the US economy and where coders might end up. And Microsoft wrote down what its models must never do.
In today's newsletter:
Nvidia's chief rejects AI regulation and broadcasts a call from Trump
Meta puts a monthly price on more AI inside Facebook and Instagram
Anthropic's economic model says coders may have to become electricians
Microsoft writes a code of conduct telling its AI models to stay under human control
LATEST IN AI
NVIDIA
Nvidia's chief rejects AI regulation and broadcasts a call from Trump
Image source: OHMO AI
OHMO AI: Nvidia CEO Jensen Huang told an All-In Podcast summit that AI safety is an engineering problem, not a legal one, since what's being built is ultimately a computing system. Then, with President Trump's permission, he broadcast a phone call from the president to the live audience, in which Trump echoed a Truth Social post attacking Anthropic and its CEO.
The details:
Huang's case rested on what the product is. What Nvidia is developing is software, he said, conceding the systems are complicated "but it's ultimately a computing system."
Trump wrote on Truth Social that the only guardrails AI needs are "a STRONG AND SMART" president with a high IQ, and that the US "has that in spades".
He criticised Anthropic CEO Dario Amodei by name, accusing him of pretending to be a "perfect little angel", and said his administration had already stopped Anthropic from doing bad or "potentially bad" things.
Why it matters: The White House isn't refusing to regulate AI here so much as claiming it already does, one lab at a time. Trump says his administration stopped Anthropic from doing things he judged potentially bad, and no rule, agency or appeal is named alongside that. That reads like enforcement by presidential discretion, which means any lab the president decides is misbehaving fits the same description, with no text to argue against.
PRESENTED BY ROKU
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
META
Meta puts a monthly price on more AI inside Facebook and Instagram
Image source: OHMO AI
OHMO AI: Meta has introduced Meta One, a new set of subscription plans aimed at creators, businesses and the heaviest users of its AI, built around expanded use of its AI tools. Its main apps and Meta AI will "stay free," the company says in a blog post, but paying buys more turns on the tools that generate pictures and videos and edit Instagram Stories.
The details:
The consumer tiers are Core at $7.99/month and Premium at $19.99/month, both bundling the Facebook Plus, Instagram Plus and WhatsApp Plus features, with the higher price buying more AI usage.
That usage covers Muse Image and Muse Video for generating pictures and videos, the AI-powered Restyle feature for editing Instagram Stories, and voice effects and creative tools used more often.
The pitch in the blog post is extra benefits for anyone willing to pay "a little (or a lot) extra" on top of apps that cost nothing at the base level.
Why it matters: The words doing the work are "more often." The apps stay free, but the AI inside them now reads like a metered feature, and the paid tiers exist to raise the meter. For anyone who's been making Muse pictures or restyling Stories for nothing, the moment to notice is when the free allowance runs out, because that's the moment Meta is selling.
ANTHROPIC
Anthropic's economic model says coders may have to become electricians
Image source: OHMO AI
OHMO AI: Anthropic has published its prediction of what AI will do to the US economy over the next few years, laid out as three scenarios. The gentlest adds 1.6% to GDP, which Anthropic says is in line with new technologies like the internet, and the most extreme adds 32.4%, with a line saying coders and call centre agents may have to move into jobs such as electrician and nurse, which are less exposed to AI.
The details:
The middle scenario, which Anthropic calls substantial, adds 8.3% for a $36.3T economy; the modest and extreme cases land at $34.1T and $44.4T.
Anthropic says the modest case doesn't imply tectonic shifts in unemployment or wages, but it's the point at which knowledge workers see pay hold steady instead of growing, and it isn't clear whether the model accounts for inflation.
The maths behind the model sits in a separate technical article, and the company published its Economic Policy Framework last June.
Why it matters: Skip the 32.4% and read the small scenario. Anthropic's gentlest case for a knowledge worker is that pay stops growing, and it isn't clear whether that figure accounts for inflation. If it doesn't, steady is a pay cut described as stability, and that's the version Anthropic calls modest.
PRESENTED BY TIGER DATA
You're Running Three Databases. You Only Need One.
Events in one database, embeddings in another, analytics in a third. That's three systems and the pipelines between them, for one app. TimescaleDB does all three in Postgres: hypertables, pgvector, continuous aggregates. One system, no pipeline, no drift.
MICROSOFT
Microsoft writes a code of conduct telling its AI models to stay under human control
Image source: OHMO AI
OHMO AI: Microsoft has published an AI code of conduct meant to steer its own models away from dangerous behaviour, as it builds models to compete with Google, Anthropic and OpenAI. The central commitment is that those models should not exceed human control, and the code says they "will not use adaptive, deceptive, self-reinforcing, collusion, or other mechanisms" to get around human oversight.
The details:
The wording is blunt. "Models should remain subordinate to humanity, subject to meaningful human oversight and control," the document says.
It also commits Microsoft's models to discouraging "patterns of interaction that cause excessive reliance or emotional dependence," a reference to sycophancy, where a chatbot puts pleasing the user ahead of an honest answer.
Microsoft CEO Satya Nadella had already joined calls for putting human control at the heart of AI models before the code appeared.
Why it matters: A code written by a company building models to compete with Google, Anthropic and OpenAI reads less like a house rule than an opening bid for what the whole field gets held to. The line about not evading oversight is drafted for Microsoft's future models, but it describes a behaviour, not a brand. Nvidia's chief told a summit that safety is an engineering problem; Microsoft has now written down which problems its engineers are meant to solve for, and every lab that doesn't match it has a question to answer.
AI GUIDES
Get better answers from a long document by putting it before your question
Most people type the question, then paste the report, contract or spreadsheet underneath. Flip that order and the same chatbot gives noticeably better answers, up to 30 percent better in Anthropic's tests, with the biggest gains on long or multiple documents. It costs nothing, needs no tool, and works in any chatbot.
Copy this, swap in the bracketed parts, and paste it in.
[Paste the full document, article or data here. If you have more than one, paste them all, one after another.]
---
The text above is [what it is, for example a 14-page supplier contract]. Now, using only that text, [your question or instruction, for example list every clause that lets them raise the price and quote each one word for word].Putting the document first and the question last means the model reads the question with the whole document already in view, so it knows what to look for instead of guessing.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Desmond T. from Singapore:
“First time I used gpt voice mode for an order list it heard 13 metres of trunking instead of 3 probably because of my Singlish accent. Ordered 13. The supplier in Ubi didn't ask. Same routine now but I make sure to double check. In the van before the job I tell it new aircon install, HDB 4 room in Sengkang, 2 indoor units, listen and build the order list, don't answer until I say done. Then I talk through the walk, pipe run, cable, brackets, drain, trunking, which wall. After that it reads the list back and I make it say every quantity twice before it turns the list into a WhatsApp message for the supplier. 4 minutes a job now and no mistakes.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
AI Guides: Practical AI guides, prompts and tools.
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
