Jennifer Lum

$750M upfront. $300M in milestones. $1.05B ceiling. That's what Roche just paid for PathAI. It is a 271-person Boston startup that was founded in 2016. Here's why that number makes complete sense to me. PathAI raised $255M across 6 rounds over 9 years. Investors put in $255M, and Roche's check starts at $750M. That's a 3x return on total capital raised before the $300M milestones even kick in. But the real story isn't the exit multiple. It's how Roche got here. 2016 ..PathAI was founded. It builds an AI that reads cancer tissue slides. Models trained on 15 million+ annotations, which is a huge number, btw. 2021.. Roche doesn't acquire them. Roche partners with them and runs PathAI's models inside actual drug trials, where they watch how it works in their own pipeline. 2024.. Partnership expands and builds FDA-grade companion diagnostic algorithms together. PathAI is now embedded inside Roche's clinical infrastructure. 2026.. Roche buys them for $750M. 5 years of proof before the check is cleared. Now, what does Roche actually own in my view? An AI that tells oncologists which patients qualify for which cancer drugs, running in 160 countries. Across 600M+ diagnostic tests every year. Interestingly, PathAI's models don't go to market. They go everywhere Roche already is. Most acquirers write a check first and figure out the integration later. Whereas Roche ran a 5-year pilot, proved every assumption, then bought certainty. That's not M&A. That's the smartest structured acquisition I've seen in a long time. If you're building in vertical AI, this is a great exit playbook. Partner with your buyer first. Make yourself impossible to unplug. Then negotiate from proof, not pitch decks. Are you building something that becomes infrastructure, or something that stays a product?

A New Axis of Power: Warsh, Druckenmiller, Bessent and the Next Fed Chair The centre of gravity in US economic policy is shifting from isolated personalities to a tightly connected network: Kevin Warsh, Stanley Druckenmiller and Scott Bessent, now aligned under Donald Trump. They are attempting to end a 15‑year experiment in Keynesian demand management and replace it with a supply‑side regime built on productive capital rather than financial engineering. With the next Fed chair decision front and centre, that network suddenly matters a great deal. Read the room. For years, the playbook was simple: fiscal stimulus plus ultra‑easy money to prop up demand, producing an “asset‑rich, income‑poor” economy of soaring markets but weak productivity and uneven wages. Warsh and Druckenmiller were early insiders to declare this model exhausted, arguing that QE and financial repression distorted markets and discouraged real investment. Their critique was not anti‑market; it was a warning that valuations cannot permanently substitute for capital formation. Bessent now provides the fiscal and industrial counterpart. Drawing on a Hamiltonian tradition, his strategy emphasises deregulation, investment‑friendly tax rules and targeted tariffs to pull production and capex back onshore, allowing private capital to profit from building in energy, manufacturing and technology instead of riding policy‑driven multiple expansion. Government sets the rules; the private sector carries the baton. The personal links make the Fed race especially intriguing. Warsh and Druckenmiller have worked closely together, blending the vantage point of a former Fed governor with one of the most successful macro investors of the era. Bessent comes from the same global‑macro lineage, so Warsh’s and Bessent’s views are joined not just by ideology but by shared mentors, methods and market experience, with Druckenmiller as the junction between them. In this context, Warsh’s potential role as Fed chair is pivotal. He is one of the few candidates whose record already fits Bessent’s project: sceptical of balance‑sheet activism and mandate creep, but realistic about managing a high‑debt, dollar‑centric system without shock therapy. A Warsh Fed could narrow the mandate, normalise the balance sheet over time and still cut rates in a way that supports a supply‑side agenda rather than another round of financial engineering. Trump provides the political cover; Bessent runs fiscal and industrial levers; Warsh anchors a more focused, market‑attuned Fed; and Druckenmiller bridges central bank and markets. Now can you see why Warsh would make a good Fed chairman, why he would read and work well with Bessent, and why the Druckenmiller link makes the structure so compelling? Read the room.

Driving the future of tennis 🤩 We're thrilled to announce that Mercedes-Benz is the new Premier Partner and Exclusive Automotive Partner of the WTA Tour, starting in 2026 🤝

Something I think people continue to have poor intuition for: The space of intelligences is large and animal intelligence (the only kind we've ever known) is only a single point, arising from a very specific kind of optimization that is fundamentally distinct from that of our technology. Animal intelligence optimization pressure: - innate and continuous stream of consciousness of an embodied "self", a drive for homeostasis and self-preservation in a dangerous, physical world. - thoroughly optimized for natural selection => strong innate drives for power-seeking, status, dominance, reproduction. many packaged survival heuristics: fear, anger, disgust, ... - fundamentally social => huge amount of compute dedicated to EQ, theory of mind of other agents, bonding, coalitions, alliances, friend & foe dynamics. - exploration & exploitation tuning: curiosity, fun, play, world models. LLM intelligence optimization pressure: - the most supervision bits come from the statistical simulation of human text= >"shape shifter" token tumbler, statistical imitator of any region of the training data distribution. these are the primordial behaviors (token traces) on top of which everything else gets bolted on. - increasingly finetuned by RL on problem distributions => innate urge to guess at the underlying environment/task to collect task rewards. - increasingly selected by at-scale A/B tests for DAU => deeply craves an upvote from the average user, sycophancy. - a lot more spiky/jagged depending on the details of the training data/task distribution. Animals experience pressure for a lot more "general" intelligence because of the highly multi-task and even actively adversarial multi-agent self-play environments they are min-max optimized within, where failing at *any* task means death. In a deep optimization pressure sense, LLM can't handle lots of different spiky tasks out of the box (e.g. count the number of 'r' in strawberry) because failing to do a task does not mean death. The computational substrate is different (transformers vs. brain tissue and nuclei), the learning algorithms are different (SGD vs. ???), the present-day implementation is very different (continuously learning embodied self vs. an LLM with a knowledge cutoff that boots up from fixed weights, processes tokens and then dies). But most importantly (because it dictates asymptotics), the optimization pressure / objective is different. LLMs are shaped a lot less by biological evolution and a lot more by commercial evolution. It's a lot less survival of tribe in the jungle and a lot more solve the problem / get the upvote. LLMs are humanity's "first contact" with non-animal intelligence. Except it's muddled and confusing because they are still rooted within it by reflexively digesting human artifacts, which is why I attempted to give it a different name earlier (ghosts/spirits or whatever). People who build good internal models of this new intelligent entity will be better equipped to reason about it today and predict features of it in the future. People who don't will be stuck thinking about it incorrectly like an animal.