The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically.
Instead focus on methods for creating knowledge that scale with computation, like search and learning.
I promoted Lorna Hajdini to Executive Director at JPMorgan because she understood something most bankers never learn.
Ownership.
Not deal ownership. People ownership. The kind of leadership where you don't just manage a pipeline. You manage the person building it. Their trajectory. Their compensation. Their future at the firm. Their references when they try to leave.
I taught her that. Not at NYU Stern. Not at Harvard. Here. In Leveraged Finance. In my corner office on the forty-second floor with the framed Tombstones from every deal that made this division what it is.
Lorna's handshake could restructure a cap table. That's not a compliment. That's a performance review.
When the complaint came across my desk, I read it twice. Not because it was disturbing. Because it was familiar.
Every behavior described. The direct communication. The after-hours mentorship. The expectation that juniors earn their advancement through demonstrated commitment to the team. That's the playbook. My playbook. The one I handed Lorna when she made Executive Director and inherited a book of direct reports who needed to understand the hierarchy.
"I own you."
I've said it to thirty-one analysts over twenty-two years. It means: I control your rating, your bonus, your promotion slate, and whether the next firm you apply to hears "top-decile performer" or dead air. It's in the HR manual under "direct management accountability." We call it alignment of incentives.
The complainant. A Senior VP in Originations who couldn't close. He alleges Lorna tied his advancement to "pleasing" her. I've read the promotion policy. An Executive Director has full discretion over direct-report advancement recommendations. Full discretion. We designed that authority. It incentivizes loyalty. It builds culture. It creates the kind of deep mentorship relationships that retain top talent.
If he interpreted "full discretion" as something other than what every Managing Director on this floor has understood since the division was founded in 1998, that's a communication gap on his end. Not a policy failure.
Harvard Business School profiled Lorna last month. "Leveling Up with Perspective, Practice, and People." She described a striking level of humility. A palpable hunger for knowledge. She talked about growing personally and professionally alongside her team. About being curious about perspectives different from your own. I wrote her recommendation for that program. I said: Lorna understands ownership the way very few people at her level do.
The profile is still live on the Harvard website. Nobody took it down. That's not an oversight. That's an editorial decision by people who evaluate leaders for a living.
The investigation lasted six weeks. I was consulted on a Thursday. They interviewed fourteen employees. Reviewed badge data. Calendar invites. Email metadata. Found no policy violation.
The complainant declined to participate. He was already on wellness leave by then. Unrelated.
Two witnesses are cited in the lawsuit. They were not cited in the investigation. I am told this is because the investigation's scope was determined prior to the filing.
Scope is important. Without scope, every investigation into a Managing Director candidate with eighteen active deal mandates and a direct line to three of our top-ten private equity clients becomes a fishing expedition that puts nine figures of annual revenue in jeopardy. We are not in the business of fishing.
Lorna remains employed. The complainant does not. His systems access was revoked on a Tuesday. I know it was a Tuesday because I approved the ticket. Standard offboarding protocol. The building badge, the Bloomberg terminal, the health insurance portal. All deactivated within the same four-hour window. He found out when his laptop locked at 2 PM and his key card stopped working at the elevator bank.
The threatening phone calls started that week. "Just wait till you're back in New York, Brown boy." Someone knew his personal number. Someone knew he was out of state. Someone knew the racial thing would land. Those are outside the scope of the firm's responsibility. We cannot police what former colleagues discuss on personal devices during personal time. We did advise him to contact local law enforcement. In writing. Via his personal email, since his corporate account had already been deactivated. I am told he received that email.
People keep asking if I'm concerned.
I thought about him once. The complainant. On a Wednesday, I think. I was reviewing Lorna's Q3 revenue attribution and his name appeared on a deal she closed after he left. His origination work. Her closing credit. Standard reassignment. And I thought — briefly — about what it must feel like to watch your work get credited to the person who.·
Anyway. Revenue attribution follows the active relationship manager. Policy is clear.
Am I concerned?
I built Lorna's career. I taught her how ownership works in Leveraged Finance. I watched her apply those lessons with a level of intensity I haven't seen since my own early years on the desk, back when nobody filed complaints because everybody understood the cost of being the person who didn't understand. If the system produced what that lawsuit describes, then I'm the system.
But the investigation found no merit.
So I'm just a mentor.
@juliarturc@curiouswavefn His narrative is he is helping US win AI war by not selling latest chips. If that action is causing Chinese AI to be more efficient he rather not want to mention it. Also anything that suggests frontier models wont need latest AI chips is against his story
Why didn’t Huang pull the Deepseek card? It’s the cleanest example where China made a huge amount of progress with old chips. You can even argue they made the progress *because* not having the most powerful chips pushed them to be more creative.
I went long a basket of BDCs ( $CGBD, $FSK, $MFIC, $MSDL, $NCDL, $OCSL) & shorted $BKLN against it. Negative vibes against private credit has reached a fevered pitch & BDCs are now mispriced vs leveraged loan comps. Thread on my thesis. 🧵 (1/n)
@mackay3688418@kevg1412 Yes they always come back with “we are helping price discovery” and then you see stuff like 20% up and down movements on earnings calls even though the earnings is exactly what market consensus is. After covid this got much worse with retail pretending to be portfolio managers
@kevg1412 Spent a number of years on the Street. The longer I am away from it, the more saddened I am that so many smart people do it. Could be solving real problems or building legit businesses. Instead, they’re gamblers wasting their life away.
Divya Nettimi went from intern to portfolio manager in 7 years at Viking Global (where she led the firm's TMT investment team) before starting Avala Global.
She's kept a very low public profile, but last week sat down with Lone Pine's Co-CIO Kelly Granat to discuss how to navigate market volatility and all things AI--investment opportunities, risks, bubbles, impact on trading strategies, role in hedge fund operations, human vs machine investing, hiring, and more.
Letter #321: Divya Nettimi and Kelly Granat (2026)
Divya Nettimi is the Founder and CIO of Avala Global. Previously, she was a PM at Viking Global, where she led the firm's TMT investment team.
Kelly Granat is the Co-CIO of Lone Pine Capital.
[Full interview in link in bio]
CC and Opus 4.5 was really the moment late in 2025, so I do believe there will be more dispersion by 2H in software for some SaaS cos that got the message, embraced the tools, and hit the gas
@F_Compounders For it. A company fist needs to make employees happy so they can make customers happy. Investors are short term greedy. If sbc is not allowed, then investors should also need to be required to hold long term. Investors and employees are on the same side. zero sum thinking is BS
Geoffrey Hinton says the current path of scaling is hitting a limit
Most high-value data is locked inside companies, and the "free internet" is largely exhausted
The solution is for models to generate their own training data through reasoning
"that's how AlphaGo beat humans"
Fed/ Williams put today.
Fed’s 3rd priority is to maintain stability and yesterday/early AM today was anything but.
Look for a fed funds cut in Dec.
I like AGNC and NLY mortgage REITs.
Regional banks like COLB and NWBI cheap if acquisitions pick up in 2026.
Jeff Bezos: “You don’t understand my audience”
When Charlie Rose asks him if the iPad is a “Kindle killer” in this 2012 interview, Jeff gives an incredible response:
“It’s a very different product, and I can give you an example. If I came to you and said: ‘Charlie, I love your show, but we’ve got to sex it up. We need fast cuts and vicious arguments between your guests.’ You would rightly say to me: ‘Jeff, I love you man, but you don’t understand my audience. We have a cerebral conversation here. It’s what we do, and it’s how we’re differentiated.’”
He continues:
“When people come to me and say: ‘You’ve got to have full-motion video and color on the Kindle.” I say: ‘Why? You think Hemingway is going to pop more in color?’… You don’t understand my audience.”
Jeff explains that his vision for the Kindle is a purpose-build device for reading where no tradeoffs have been made—every single design decision is optimized for reading.
And knowing exactly who his audience is gave him conviction in the Kindle team’s design decisions—even when everyone at the time was saying that the iPad would crush them.
@Citrini7 Why do you think CLO market is tanking? XFLT, olxc, eic. Not all of them are levered, some have good balance of loans and CLO debt and clo equity. They are below covid lows. Something is very off
banks have tossed more trash to non-fin entities than it appears, loans to private credit, VIEs, BDCs etc…
you can get an idea of exposure but anyone who tells you they have a good read on credit quality / contagion is lying to you. probably fine. also could not be.
@jeremyphoward@karpathy Had the same experience. The best models (codex, gpt-5, sonnet 4.5) can't write working PyTorch code to train a model successfully. Works great in other areas (CRUD, web, etc)
Almost 5 hours into the work day at my manufacturing plant:
1) Tried out new warehouse candidate yesterday and he didn't show up today. Next man up. Guy lasted one day and didn't even text to say he wasn't interested. Standard sadly. I texted to confirm no interest and lined up a new candidate tomorrow morning to start
2) Traded texts with other SMB owners on minimum cash and how they think about it. Some just keep enough for payroll the next week with the balance in money markets while clowns like me need $500k to sleep at night
3) Another SMB buddy texted me last night he's feeling good about a potential tuck in
First day that's really felt like Fall. A little brisk today. Feels nice to be honest.