whatstheofficerproblem

I will take Buffett criticism from anyone and everyone except this crook.

The problem with AGI is that once you have it you're close to singularity and once that happens world will be in an apocalypse with UBI, revolts and everything. Which is why it's all the more puzzling to me why anyone at all would allocate capital towards AGI, the ethical implications of it aside, it's literally impossible based on the current way we train models. As for who wins, obvious answer would be Verisk, Constellation, Thompson Reuters & Epic systems kind of companies that have siloed prop data that these model sellers won't have access to. No amount of intelligence artificial or otherwise can conjure up 40 years of financial terminal data or the medical records of half of America's hospitals. The data is the alpha. Then come the embedded workflow companies i.e. the ServiceNow, GitLab, Salesforces of the world. They win not because of what they build but because of what their customers have built into them. Decades of integration, customization, compliance configuration, and institutional muscle memory. They're model agnostic as discussed previously & they benefit from falling model costs, and their switching costs are measured in years of enterprise pain. AI makes their products better and stickier without threatening their position. Then of course, AI infra names. The physical and the scarce is something AGI cannot replicate. All roads lead to TSMC, but you could go further down into HBM, NAND, Opticals, Energy, Cloud & neo-clouds etc.

That egg is constantly wrong about everything. All these guys including Jensen have been shilling "AGI has been here since yesterday" for 3 years now.

The counter argument defeats itself. Yes, you're right on it being lowest hanging fruit as it's deterministic. But even in this "easiest" domain, METR studies show experienced developers are 19% slower with AI, AI-generated code produces 1.7x more bugs, code churn has nearly doubled, and 95% of GenAI pilots fail to reach production. If AI can't reliably close the good-to-great gap in the domain with the tightest, most unambiguous feedback loop available, what exactly is the basis for believing it will handle domains where feedback is messier, more subjective, and more delayed? The AGI argument has a dependency chain, current models improve -> we get AGI -> AGI solves everything. Each step here requires clearing a bar higher than the last. My point is that probabilistic models are the wrong mathematical tool for modeling reality. You will never reach AGI using RL. Discussion around AGI feels a lot like cold fusion to me lmao. "sure, the current business model doesn't work, but once we achieve cold fusion, energy will be free." Prediction of AGI as you mention above, is not evidence of it but rather a bet. Who are people making the loudest predictions here? Usual suspects that are spending hundreds of billions into this bottomless pit with the most financial incentive to make those predictions. I mean, Sundar himself dodged the AGI bullet and even called it semantics. Great, so even Google, spending $180 billion in capex, won't explicitly commit to an AGI timeline. You'd have to assume a monotonic curve i.e. capabilities keep improving, therefore they will keep improving until AGI. But the research once again shows something radically different. Positive sentiment toward AI tools dropped from over 70% to roughly 60% in 2025. Trust in AI-generated code accuracy fell from 40% to 29%. AI projects have an 80% failure rate that has remained stubbornly consistent despite better tools and growing expertise. These are not the metrics of a technology on an exponential path to transcendence but looks like something hitting a ceiling and plateauing to me. I allude to this in my write-up above. I'm not arguing AI isn't transformative, but we need to stop conflating technological potential with economic accrual. Fiber was important and transformative, who made money? Cloud was important and transformative, who made money? The question never was if AI will be important, but who will make the money. Even if AGI arrived tomorrow, the economic questions we've been discussing don't disappear. They get worse. If AGI can do everything, then the model is the ultimate commodity as in everyone will have it, nobody can differentiate on it, and value accrues to whoever has the proprietary data, the embedded workflows, and the customer relationships. Which is, again, the software companies, not the model providers. Think about what AGI actually means if you take the claim seriously. It means a system that can do any cognitive task at human level or above. Now follow that to its logical conclusion. If OpenAI builds AGI, what stops AGI from building a competitor to OpenAI? If the system is truly generally intelligent, it can design new architectures, optimize training runs, discover novel algorithms. The moment AGI exists, the intellectual moat a la "Saar we have the smartest engineers saar" evaporates because the AGI is the smartest researcher. And it's running on hardware that anyone with enough capital can rent. The scarcity was always human genius and doesn't AGI eliminate it? The open-source dynamic accelerates this argument to absurdity. DeepSeek replicates frontier capability at a fraction of the cost using clever engineering. That clever engineering was done by humans who are scarce and expensive. In a post-AGI world, the clever engineering is done by the AGI itself, which means the replication cycle collapses from months to days or hours. Meta releases Avocado or Llama or whatever, AGI optimizes it overnight, and now everyone has frontier capability. The model providers' lead time vs peers is their most valuable asset and moat and that would shrink to nothing. I said Google survives because it embeds AI and vertically integrates. But if AGI is generally available, every competitor can build equivalent products. The reason Google's integration works today is that it takes thousands of brilliant engineers years to build these systems. AGI removes that constraint and suddenly a startup can build a search engine, a video platform, a cloud infrastructure all in weeks. Google's moat wasn't really the AI, never was. It was the accumulated complexity that only a massive organization could manage. AGI is the universal solvent for accumulated complexity, at least that's what it inherently means. Today, the stack looks like: hardware -> foundation model -> application -> customer. Value concentrates at the model layer because building models is hard. In a post-AGI world, building models is trivial as the AGI can do it. Building hardware is still constrained by physics as you need fabs, silicon, rare earth materials etc. So value either drops down to hardware and energy (things that are physically scarce and can't be replicated by intelligence alone) or rises up to the customer relationship layer (brand trust, regulatory approval, data access, contractual lock-in). The model layer, which is pure intelligence, becomes the cheapest part of the stack because intelligence is now abundant. If you zoom out and look at all this retardation from a philosophical perspective, intelligence has never been the bottleneck for most economic activity. The bottleneck is trust, coordination, regulation, physical resources, and institutional legitimacy among other things. A brilliant doctor who isn't licensed can't practice medicine. A genius architect whose building isn't up to the building code gets shut down. AGI adds infinite intelligence to a system where intelligence was already becoming abundant and other things were scarce. It's like adding infinite water to a desert where the constraint is actually arable soil. So, AGI bull case is a paradox. The more powerful the intelligence becomes, the less the intelligence itself is worth, because it's the one thing that's no longer scarce. Everything else becomes relatively more valuable and what becomes valuable is not something these guys can provide. That was all a mouthful, but yeah, been going down on this rabbit hole for while.

Karthik Sharma is a billionaire now. His investment in the name is such a case study.

Think RL is a fundamentally flawed way to model general intelligence. Problem here is you're pretty much domain locked to places where you have a tight feedback loop. The reason these models do so well in coding is because the feedback loop is tight and unambiguous i.e. either the code works or it didn't. AI will 100% replace software engineers if judgement and management are zero value adds. The fundamental question that AI will replace software companies diverges into two different conversations. Will AI write most code? Almost certainly yes, and soon. The trajectory is clear we've gone from autocomplete to copilot to full agentic coding in about three years. Sundar himself said he started a project and didn't even know what language the AI chose until after it was running. Grunt work is disappearing. Will AI replace the thinking behind software? That's much harder. What should we build? Why? For whom? How should systems interact? What tradeoffs are acceptable? What happens when requirements are ambiguous, contradictory, or politically loaded? That's the messy and there are no reward signals there. The likely outcome is something like what happened to farming. Agriculture went from 90% of employment to under 2%, but we produce more food than ever. Software output will explode while the number of people writing code by hand collapses. But someone still needs to decide what to grow and where to plant it. The deeper irony is that AI replacing coders is actually bearish for the AI companies. If coding becomes near-free, the cost of building competitor products also becomes near-free. Every moat gets shallower. The same capability that lets OpenAI move fast lets three people in a garage replicate their product in a weekend. I don't mean that in a way someone will come up with GPT 5 overnight, but while the training moat is real, the product moat isn't. DeepSeek for example, while didn't replace OAI's infra, found ways to get comparable results with a fraction of the compute through architectural clevernessfound ways to get comparable results with a fraction of the compute through architectural cleverness. They didn't match the full stack but they matched the output well enough that it didn't matter. This pattern will become recurring imo. You won't need to replicate the infra but emulate the capability at the layer the customer touches. Three things make this possible over time. First is open-source models keep closing the gap. Sundar himself said it i.e. Gemma 4 is based on Gemini 3 architecture, fits on a USB stick, and the gap between it and the frontier is "both huge and not so huge in terms of time." The frontier leads, but the open-source tier follows faster each generation. Second, distillation and fine-tuning are cheap. You can take an open-source base model, fine-tune it on a narrow domain, and get frontier-level performance for that specific use case at a tiny fraction of the cost. Most businesses don't need general intelligence and are content with AI that's great at their workflow. Third, inference costs keep dropping. Even if you can't train the model, you can serve it. And if you're building a product on top of someone else's API, your competitor can build on the same API tomorrow. All of this yet again ties back to the fiber buildout parallels. The people who laid the fiber went bankrupt. The people who built applications on top of cheap fiber, the Google, Netflix, Amazons of the world, will come out on top. Companies like ServiceNow and GitLab have something the model providers don't i.e. embedded workflows and customer lock-in. ServiceNow doesn't win because it has the best AI. It wins because every Fortune 500 company has spent years wiring their IT service management, HR processes, and procurement workflows into it. Ripping that out is a nightmare regardless of how good a competitor's AI is. They can plug in whichever foundation model is cheapest or best at any given moment and pass the value to their customers. They're model-agnostic, which means they benefit from the competition between model providers driving prices down. But there's a real threat to them too, in that if coding becomes near-free, the cost of building a ServiceNow competitor drops dramatically. Today, replicating ServiceNow's platform would take hundreds of engineers years of work. In three years, maybe a small team with AI agents can build 80% of the functionality in months. The workflow lock-in is real, but it's not infinite, even more so if a customer is paying massive enterprise license fees and suddenly a lean alternative appears. The real winners are companies that combine proprietary data with AI in ways nobody else can replicate. Think Bloomberg with financial data, or Epic with healthcare records. The model is commodity. The data isn't. That's where durable value lives. The question then becomes, will the foundational model providers survive long enough for there to be budding competitors for incumbent software companies. I don't see how the model providers escape that trap. Scenario one: they compete on price. Margins get crushed. They've spent billions on training and infrastructure, but the API becomes a commodity. The software companies love this as they get cheaper and cheaper inputs while maintaining their own pricing power through workflow lock-in. The model providers bleed cash, and the economics never work. Close enough, welcome back Telecom bubble. Scenario two: they collude or consolidate to hold pricing and run the OPEC playbook. But unlike oil, you can't control supply here because open-source exists. The moment OpenAI and Anthropic try to maintain fat margins, Meta releases the next Llama, DeepSeek ships something competitive, and the software companies just switch. Once again, Sundar Gemma 4 fits on a USB stick and is closing in on the frontier. The software companies being model-agnostic would make a pricing cartel impossible. Scenario three: open-source wins outright. We go Linux trajectory & Linus Torvald didn't kill Microsoft or Oracle overnight, but he fundamentally restructured the economics of infrastructure software. If open-source models reach "good enough" for 90% of enterprise use cases and the trajectory here suggests they will, then the foundation model providers are selling the remaining 10% of premium capability at massive cost and that's a bad business to be in. The really vicious dynamic is that all three scenarios can play out simultaneously across different market segments. Open-source eats the bottom. Price competition eats the middle. Only the absolute frontier commands premium pricing, but maintaining the frontier requires ever-increasing spend, and the lead time before open-source catches up keeps shrinking. You're running faster and faster to stay in the same place. The one escape hatch seems to be vertical integration. Instead of selling the model, embed it in Search, YouTube, Cloud, Workspace, Waymo, and extract value through those products. The model becomes a cost center that accelerates your actual businesses, not a product you need to monetize directly. Microsoft is doing the same through Office, Azure, and GitHub. OpenAI and Anthropic don't have that luxury. They are the model. Which is why OpenAI is frantically trying to become a consumer platform and Anthropic is pushing enterprise deeply. They need to build the application layer before the model layer commoditizes underneath them. All this considered, software companies are probably in the strongest position of anyone in this chain. They're the arms dealers selling to both sides. The model providers are the ones with the existential math problem. And ironically, the better the technology gets, the worse the economics get for the people building it. Which is, once again, exactly what happened with telecom. So I think the software sell off makes no sense and the broader market is making the mistake of chasing the most exciting story instead of following the economics. Model companies have the breathless headlines, the frontier breakthroughs, the "this changes everything" demos. Software companies have become boring. Nobody writes breathless articles about ServiceNow's latest sprint. So it makes sense that capital flows toward the story. Another thing I think is being conflated here is that who captures the value versus who creates it. The model providers are creating enormous technological value but may capture very little of it economically, a la Telecom. The software companies are positioned to capture value by sitting on top of commoditizing infrastructure and maintaining pricing power through switching costs. The market is pricing model companies for the value they create and software companies for the disruption risk they face, when it should arguably be the reverse. Disruption risk is mispriced here in both directions. Software companies are being penalized for the threat that AI makes them replaceable. But as discussed above, their moat is workflow lock-in, customer data, and integration complexity, NOT CODE! Meanwhile model companies are being priced as if they'll maintain margin dominance in a market where open-source is closing the gap every six months and the product is inherently commoditizable. The market is overweighting disruption risk for the companies with moats and underweighting it for the companies without them. As much as I hate bringing up ServiceNow as a comp here over and over again, they're running at 80%+ gross margins selling subscriptions with multi-year contracts all while model providers are burning cash at historic rates with no clear path to comparable margins because every efficiency gain gets competed away or replicated by open-source. The market is rotating out of proven high-margin recurring revenue into speculative negative-margin businesses where the competitive dynamics get worse as the technology improves. To draw an non-Telecom parallel here, I would like the readers to ponder on this question for me. Circa 2006-2010, if you had to choose between investing in AWS itself or the SaaS companies that would be built on top of AWS, what would you choose? The right answer is the SaaS companies. CRM, WDAY, NOW etc. rode the cloud commoditization wave and printed money because cheaper infrastructure made their businesses better, not worse. AWS also did well, but only because Amazon had the vertical integration escape hatch discussed above. The pure-play cloud providers that tried to compete with AWS on infrastructure alone mostly got crushed. I am of the opinion that the market today is completely skipping asymmetry of outcomes. For a software company, AI going well means their product gets better and stickier. AI going badly means they keep running their existing business. For a model company, AI going well means they might monetize it before commoditization hits. AI going badly, hell, even going well but commoditizing would mean their entire business thesis collapses. The risk-reward is dramatically more favorable for the software names, but the market is pricing it backwards because the model company narrative is more exciting. I have no crystal ball into when this might reverse, but I'd say OpenAI & Anthropic going public would be a start.

Any recommendations for local brews in Bozeman? Going to be visiting Yellowstone end of May.

Not for the time being. Will likely add after I take profit on INBX.

Good to see the spread between COHR and LITE increase, was overdue. This was a consequence of the Goldman note this morning where they said TAM for these guys is going to 30x or something like that. Anyways, not worried that I sold it. Enough factor swings here so will be able to buy-in again, would also be a good time to add if the street keeps revising up.

Trimmed the fund's SNDK position by half today as the stock basically almost doubled from our cost basis. Think the Nasdaq100 flows today are helping, spread between MU & SNDK widening which I think will revert. Punted COHR for a 50% gain. Semis are very volatile and the factor swings are crazy, these names will come down again for no apparent reason or CXMT IPO will catalyze another money rotation zero sum game at which point can add back to these positions at a cheaper multiple. Rotated the proceeds into INBX which has an imminent catalyst on the horizon and will likely be a multi-bagger from here.

Good day to be a Tegridy Value LP it seems. We're up ~8% since inception so far. Thanks to favorable developments at SNDK, CVRX & RVMD. Just need Judge to get LQDA out of the way now and we're golden.

Enough said. Trying this, will post my thoughts here.

Hoodlum, as I've mentioned elsewhere on this forum, I judge liquor on a price:taste:buzz basis. Never really understood the fascination with wines, they're a layman's version of Guinness imo i.e. folks like it for the sake of liking it. No offence to all the wine drinkers, just my thoughts . That said, some of the best wines I've had were those brewed locally in IA or WI. Their sweetness to buzz ratio is unmatched, ofc, given how much I enjoy alcohol, I will never shy away from trying your recommendations. I'd like to stick to my Scotches and IPAs. Bourbons I enjoy occasionally as a few LQDA investors sent me cartons of those, might as well share a glass with @Gregmal when I'm in his state. @MMM20 would end up owing me crate of MD 20/20s but I've never tried those. I enjoy RedBreast the most. None of the scotches awaken a desire to sleep in my insomniac body like the RedBreast does, tastes so sweet and the notes are excellent.

I'm officially 23 now. Getting closer to being called "Unc". That one virus really did ruin my generation's perception of time. Went to the liquor store today, regular beers aren't cutting it anymore for me as I only drink them when having a meal. I have been exclusively drinking IPA for a few years now, and given the history of IPAs, I'm not surprised at all - only natural that I like them as they were quite literally made to cater to my palate! Wanted to get Michelob Amber Bock but couldn't find those, I've already tried all the IPAs on the shelf especially all the Steel Reserves and Voodoos, but this one I haven't tried. Told my friends I was buying a G-Force and they thought I was talking about NVDA chips . It's ok, tastes better than a Guinness, but that's a low bar. I'm saving my golden drops of the Ardbeg for when LQDA wins the suit. Enjoying it, bought three 19.2fl oz cans. Anyways, cheers.

Ongoing UBS ad check call is shaving 100bps from each of the remaining quarters for total ad spend currently, Mainly coming from brand budgets and CTV. Reads incredibly bullish for MNTN, MGNI & NFLX.

Hugging any benchmark makes no sense when the portfolio is constructed the way it is. Crazy momo and factor swings in the index especially with the macro. There will obviously be time periods where the indices will outperform our portfolio at tegridy, it's simply the price you pay for higher conviction / misunderstood names. Just look at the shit show in indices below since Tegridy's inception through Mar 31. This would make me look like a genius, when I am in fact not. Any short term outperformance or underperformance is just noise. Or at least that's the way I see it. S&P500: (5.60)% NDX: (7.28)% IWM: (6.35)% XBI: (0.24)% DJI: (5.62)% Tegridy: 2.66%

I usually disclose weightage, looks like I was in a rush to put this out and forgot. This is as of March 31. We're up ~7% since inception as of Apr 2. Ridiculous how much 2 days can change in the portfolio.

Feel free to post any questions here. Will get to them.

Dear LPs, Please refer to the document below for our 1Q letter. Tegridy Value LP 1Q26 Letter.pdf

The spread on my backwardation play is closing. Looks like more folks getting involved. Short CLU26 and long CLZ26, spread was $6+ last week, now $4.

These guys really think they have FICO level pricing power? The enshittification is glaring.

Don't like it. Orange Spark perhaps the best flavor among Velos, pity they don't sell that in the US.

Happy Friday, pleased to announce my SHIPMENT has arrived. The corrupt Democrats sent me MAGO FLAME instead of ORAGE SPARK, thinking that would make me feel bad. Worry not I am going to SCALP these corrupt Democrats like I scalped SLEEPY Joe and get a FULL REFUND. THANK YOU FOR YOUR ATTENTION TO THIS MATTER!!!! DJT @John Hjorth, I might have to call some favors with you and ask me to send the Swedish stuff, the american ones suck really bad. The ones below are Swedish, the sensates at least. I have another shipment currently seized by the FDA , this is what they're doing instead of approving and testing life saving drugs. Might have to switch to GOAT 20mg+, these 10mg aren't cutting it anymore.

Added to COHR today because price action was plain stupid and sell side desks were banging the table on S&P inclusion. It was indeed added alongside LITE, VRT & SATS.

I suggest going through the MS TMT call for both LITE & COHR. I think the latter's moat really shows in the current backdrop.