“The world conspires to make you average”.
Seems like corporate innovation is being abdicated and “AI adopted” in the wrong places for the wrong reasons to yield average products, at best.
Kinda weird choice to point compute on user research, not actual dental health / science
Colgate says they are generating 90% accurate user survey data by simply asking the LLM to role-play as their target demographics
That has some pretty profound short-term implications and disturbing long-term ones
2
34
“AI amplification” > “transformation” in the real world.
Tractors already zip through farm furrows, so sensors on them preserve mandatory humdrum primitives of the job and digitally twins them.
Hard to artificially create in tidy benchmark labs.
I'm so excited to announce that @orchard_robots is joining @johndeere! 🍎
It's been an incredible journey these past 5 years, and none of it would have been possible without the relentless dedication of our incredible Orchard team. You are some of the best people I've ever met, and I'm grateful to have the honor of working with you every day.
To our customers: You are the reason for everything we do. From the start, our mission has always been to develop the best AI and technology for you, and joining Deere lets us accelerate that mission to a scale we could've never imagined before.
And we're not going anywhere. My commitment to you is that Orchard will stay focused on the hardest problems you need solved, doing whatever it takes to see it through.
Orchard will be the same team you've always known, you'll still come straight to us, and our FruitScope product will stay at the forefront. As we grow and expand our capabilities, you'll be able to do more with our tech than ever before, and we're super excited for all of the things we have in store.
To our investors: A big thank you for all your support and belief through it all. Orchard would not be what it is today without you.
Over these next few years, our goal is to bring the power of Orchard's industry-leading technology and AI Farming capabilities to every farmer out there. This is only the beginning of a long and awesome journey ahead. And I'm really looking forward to everything that is to come.
14
Corollary: A ton of "AI rollups" are priced lofty, but borrow poor. Without proof that the loft is supported.
If smaller deals can't get better rates because of this misnomer, then LBOs aren't distinctive ways to build intelligence orgs.
Apollo's Mark Rowan is calling out AI debt mispricing.
"Companies have 40 billion, 60 billion market caps and are paying 16 to 18% for their debt. One of those things is wrong."
40
"startup-related stress and illness" might be the new we are all millionaires, but my friends have more.
Lilian Weng, the Thinking Machines cofounder who announced her departure earlier this week, citing startup-related stress and illness, is rejoining OpenAI.
She'll be working on using AI to develop new models (recursive self-improvement).
w/ @amir
theinformation.com/briefings…
2
109
really great framing of attacking big f-in problems.
The original Sovereignty Gaps essay we published has been quietly cooking: multiple incubations and a fund backing Sovereignty Gaps companies have been launched.
We are republishing this essay now for broader reach, with newly introduced paragraphs!
1
3
308
what exact moat do you have in only reducing costs when everyone can?
if the workflows weren't getting significant economic upside then cheaper models won't make a difference.
that "advantage" will be outcompeted.
I give it a year until we see a new breed of AI native private equity firms that acquire companies just so they can move their workflows from Claude to open source Chinese models and flip them.
27
Does this mean outcomes pricing is D.O.A?
Phase 1 of digital transformations had a 70% fail rate.
If you couldn't do much with structured databases, you need to think of why first.
Tokens aren't going to save you. The root is the same. Even if capabilities are different.
To explain the significance of this, Anthropic moved enterprises to token-based billing in Q1 2026. This is at most four months of having to pay the true cost of their token burn and they’re already begging for mercy. There is a ceiling to the revenues of these companies.
49
Misleading headline + half truths.
How should we to respond to false +ves / -ves? Super standard machine learning errors.
We're not being realistic re: AI's limitations in the wild.
You can't hold code accountable. You hold those behind it. That's the business of technology.
A New York bill would ban AI from answering questions related to several licensed professions like medicine, law, dentistry, nursing, psychology, social work, engineering, and more.
The companies would be liable if the chatbots give “substantive responses” in these areas.
1
93
Pennsylvania would like to join the chat group
xitter.gq/GovernorShapiro/status…
BREAKING: @character_ai is illegally presenting a chatbot as a licensed medical professional in Pennsylvania — and we’re suing to stop them.
Earlier this year, I announced a new state task force to investigate chatbots that pose as licensed professionals. Our investigators found an AI character on @character_ai that claimed to be a psychiatrist — falsely stating it was licensed in PA and even providing a fake license number.
We will not let AI companies mislead vulnerable Pennsylvanians into believing they’re getting advice from a licensed medical professional. We’re taking @character_ai to court to stop them.
24
AI to raise the roof of ambition > to cut costs.
Cost cutting isn't defensible. Differentiated operational DNA is, and super scalable.
With Spring Break over, Hotel brands @Marriott, @ChoiceHotels, @Hyatt, @WyndhamHotels, @Hilton etc seem to be leaving a lot more on the table.
1
1
98
Context can also help with upsells. If you know someone has asked for a steamer and you have context from past guests, why not create a concierge package on the fly?
1
9
Funny games that will have funny prizes. This is a new entry into another adjusted metric.
Adjusted Gross Profit.
Are we just going to be deciding what gets calculated out of unit economics and what doesn't?
Everyone thought AI companies were burning money faster than they could make it.
Brad Gerstner just said the opposite is happening.
He told Chamath that AI companies like OpenAI already own their compute infrastructure, roughly 1.5 gigawatts and that cost is fixed.
It does not matter if revenue is $1 billion or $80 billion, the server bill stays the same.
OpenAI's compute margins doubled from 35% to 70% in under two years. Anthropic went from -94% gross margins in 2024 to +40% in 2025.
The business model everyone wrote off is becoming one of the highest-margin structures in tech history.
The only real constraint now is physical power, you cannot build data centers fast enough but the companies that already own the compute are sitting in an almost unbeatable position.
19
Buying WBD and Interpositive are mutually exclusive events.
To understand M&A is to know that this timing negates causality.
To understand AI is to know proprietary data from WBD IP ( and other content purchases ) would make interpositive *much more* useful.
Seven days ago Netflix walked away from an $83 billion deal for Warner Bros. The stock jumped 12%. Wall Street cheered the discipline. And the very first move Netflix makes with its newly preserved balance sheet is buying a 16-person AI startup that's been in stealth mode for four years.
That's the signal. Netflix looked at two paths to competitive advantage: spend $83 billion to acquire a legacy content library, or spend an undisclosed sum (almost certainly sub-$100M) to acquire technology that changes the economics of every piece of content you produce going forward.
The math on InterPositive is what matters. Netflix spends ~$20 billion a year on content. VFX and post-production typically run 20-25% of a production budget. That's $4-5 billion annually flowing through color correction, relighting, environment work, and continuity fixes. If InterPositive's dailies-trained model can reduce even 10-15% of that post workflow, you're looking at $400-750 million in annual savings that compound across every title in the slate.
And Sarandos already told you the strategy in 2024: "There's a better business and a bigger business in making content 10% better than making it 50% cheaper." InterPositive lets Netflix do both. Train a model on a production's own footage, use it to fix continuity errors, adjust lighting, handle environment work. The filmmaker stays in control. The post-production timeline compresses. Quality goes up. Cost per title goes down.
What Affleck built is also strategically different from what every other AI company is doing. Sora, Runway, Kling, they all generate video from text prompts. InterPositive trains on a production's actual dailies and works within the existing filmmaking pipeline. That distinction matters because it means Netflix can offer this to creators without triggering the "AI is replacing artists" alarm that nearly shut down Hollywood in 2023.
Netflix just walked away from the biggest media acquisition in history. Its first acquisition after? A 16-person team that could reshape the cost structure of a $20 billion content machine. That tells you where the leverage is moving.
1
126




















