LBO_Guy (@The_Analyst_Lab) header
LBO_Guy (@The_Analyst_Lab) profile picture

Pushing the limits of Microsoft Excel.

Not Financial Advice
Joined January 2023
Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)Photo by LBO_Guy (@The_Analyst_Lab)
Here is a brief NotebookLM video summary of Aggregator Wars. The short version: AI agents go straight for the one asset that made the internet giants rich, the direct relationship with the customer. That is why nobody will take their foot off the accelerator. Data wins the customer, recurring cash flow keeps you in the fight, and profits migrate to hardware providers nobody can copy and to whoever owns the customer. Google and Meta are best placed, consensus has Meta upside down, and the AI labs and compute sellers face a funding squeeze into 2027 as capital markets run dry. If you want a deep dive on what an aggregator is, how agentic AI changes legacy aggregation theory, who is best positioned to be the next aggregator, what each company's profit engine means in this capex cycle, what it means for AI labs and compute sellers, and where value migrates to, I recommend reading the full note. It is as condensed as possible, but packed with information. I'd invite anyone to push back on any angle of this framework.
1
2
13
4,988
The timing will make this conference legendary.
big thanks to @GavinSBaker, @AntonioGracias and iConnections for holding @TomorrowXSummit at such a key moment for AI. vibe check never been more valuable, and I’ll bet the food is better than the junk GS and MS feed us at the Palace SF
7
4
1
143
21,297
Alguien por Barcelona este martes-jueves? Alguna recomendación para hacer por la tarde/noche por la ciudad?
3
797
LBO_Guy retweeted
i want to bring to the attention of people outside spain what i think is a pretty sad hijacking of a public institution, one that should serve everyone, to favour the party in government. the CIS (centro de investigaciones sociológicas, or centre for sociological research) is spain's public polling institute. it goes back to 1963, when its predecessor was founded, and it took its current name in 1977. its president is appointed and removed by the government, but by law it must act with "objectivity and neutrality". this institution is a big provider of public goods: besides its flagship product (the monthly political barometer), it runs dozens of surveys a year on things few private firms would pay to ask about, from taxes and inequality to sexual health or love and relationships, and it makes the microdata freely available. it has resources no private pollster has, and researchers and journalists (me included) use its data all the time. in june 2018, a few weeks after pedro sánchez became prime minister, his government appointed josé félix tezanos as president of the CIS. tezanos is a sociology professor and has been a member of the socialist party (PSOE), the party that leads the government, since 1973. when he was appointed he sat on the party's national executive (its top leadership body) as head of research and policy programmes, which put him in charge of the party's election manifestos (he left that post about ten days later, after criticism from the opposition). since then, the CIS's vote estimates have leaned systematically towards the socialists. one can show this with a very simple exercise. for each general election, wikipedia compiles the results of every published opinion poll (for the 2026 election, they are here: en.wikipedia.org/wiki/Opinio…). using those tables, for every CIS vote estimate since 1997 i take the PSOE lead over the centre-right people's party (PP) and subtract the average lead in the private polls whose fieldwork ended within 21 days of the CIS's. i leave out polls commissioned by the parties themselves and the re-estimates that other analysts make with the CIS's own raw data. each firm counts once, even if it published several polls in the window (in that case, i take the average of its polls). from 1997 to 2018 the CIS was in line with everyone else: on average, a difference of about zero (+0.1 points). but since tezanos was appointed, the CIS has given the PSOE a lead 8.4 points larger than the average of private pollsters. in 91% of its estimates it gives the PSOE a bigger lead than every single private pollster in its window. if i showed this chart at an economics seminar, i think most people would find it pretty convincing evidence that something changed at the CIS in 2018. but let's go through some potential objections. suppose private pollsters have drifted to the right since 2018 (it would be odd for them to jump so sharply just after tezanos was appointed, but for the sake of argument let's treat it as a real possibility). in that case, a CIS that stayed put would look more and more pro-PSOE simply because everyone else moved. or the mirror image: everyone, the CIS included, leaned to the right before 2018, and the CIS was the one that fixed it after that. the data suggest this is not what happened. first, the gap holds no matter who you compare the CIS with. against 40dB alone, el país's pollster (el país being a centre-left newspaper), it is 7 points since march 2019, compared with 9.2 with all firms on the same dates. dropping five firms that mostly work for right-leaning outlets (sociométrica for el español, nc report for la razón, data10 for okdiario, target point for el debate and hamalgama métrica for vozpópuli), it is 7.9. second, if private pollsters had drifted right, then their final polls would miss towards the PP more often after 2018. across the 90 general, regional and european elections since 2008 for which the CIS ran a pre-election survey, that is not what we see. private final polls missed the PSOE-PP lead by 0.1 points in the PP's favour on average before 2018 and by 0.4 points in the PSOE's favour after. in both periods they erred towards the PP about half the time. the CIS's average miss, meanwhile, went from 0.8 before tezanos to 3.5 points in the PSOE's favour after. this also rules out the mirror story. before 2018 there was no right-wing bias to fix, since private polls were mostly on target, and after 2018 it is the CIS that ends up further from the results. needless to say, the result does not depend on the time window used to pick the private polls the CIS is compared with. with 7, 15 or 30 days, or comparing with the polls of the same calendar month, the gap since 2018 stays between 8.2 and 8.4 points, and the one before 2018 stays at about zero. so what changed? kiko llaneras (@kikollan) has looked at this in depth. since tezanos arrived, the CIS has overestimated the left in practically every election (general, regional and european). even in the two general elections where "it did well" (april 2019 and 2023), its only merit was that it overestimated the left when everyone else underestimated it (and it still missed, just in the other direction). for instance, in 2023 the PSOE and sumar (another left-wing party) won 152 seats. the polling average gave them 142, and the CIS gave them 171. that fits what i find. a large part of the bias seems to come straight from the sample, which overrepresents PSOE voters. for example, take the CIS's huge pre-election survey for the 29 november election (45,000 interviews). among respondents who report how they voted in 2023, 41.8% say they voted PSOE, when the PSOE actually got 31.7% of the vote. PP voters, meanwhile, make up 23.6% of the sample, compared with an actual 33.1%. in other words, PSOE voters are overrepresented by 10 percentage points, and PP voters underrepresented by almost as much. until 2018 the CIS corrected this kind of imbalance in its estimate by weighting the data according to what respondents say they voted last time (you can check its own methodological notes here: cis.es/documents/d/cis/es320…). in september 2018, less than three months after tezanos arrived, it stopped doing that (compare the note on the vote estimate in july, cis.es/documents/d/cis/es321…, with the one in september, cis.es/documents/d/cis/es322…). i am not a political scientist, and i don't know what tezanos's intentions are, or whether this even helps the PSOE win votes. but missing in the same direction in almost every election for eight years looks like a deliberate choice. i hope this reaches as many people as possible. the CIS does a lot of valuable work and it should belong to all spaniards, whoever they vote for. running a public institution this way should not be acceptable, and watching its credibility erode is genuinely sad, especially because that gives people a cheap excuse to distrust every other statistical body whose work depends on being seen as independent and reliable.
19
258
20
796
57,197
Elon pushing for Jensen out of fear that he won't get enough GPUs for his superclusters will never be funny.
432
8B tokens a week looks like my steady run-rate. One thing I've noticed is that even if I keep moving work from the model into plain code (my Sep 28 setup would probably cost 70% less today), the savings never show up!! Every optimization frees room for new ideas, so I spend the same or more. Probably do 3-4x the work I did on sep 28th but spend exactly the same amount of tokens. All subscriptions maxed, that is the limiting factor for me. If I had an unlimited budget I'd spend much more. We are europoors after all. Anyways, next couple of weeks will probably run higher as there are a few interesting things I wanna do. It really is astounding the amount of work you can do, LOL.
2
6
669
What is Xi Jinping's succession plan? He doesn't have one. What would happen if he died tomorrow? A massive financial crisis. Then Xi's wife or daughter would probably choose a successor. @vshih2
77
392
47
4,953
483,731
Dwarkesh is probably one of the most intelligent, out-of-the-box thinkers I follow, and he’s now more bullish on superintelligence than ever (until recently he thought it was a decade away). I agree with him. I fully recommend watching the explainer. Truly phenomenal. Would love an interview between @patrick_oshag and @dwarkesh_sp. As AI gets obviously better at everything you can check, from maths and coding to AI research itself, the timelines keep collapsing. It’s truly bizarre to look back at what these models could do in 2024, 2025 and 2026, and it keeps compounding as Dwarkesh puts it "from failing four-digit sums in 2023 to OpenAI claiming a Millennium Prize problem in 88 hours this autumn." Basically the core of the idea is the following: AI research is mostly recombining ideas and running experiments against a score, which is exactly what models already crush. If you automate that, then you get thousands of top-level researchers working at 10x speed, 24/7, squeezing years of progress into one. Dwarkesh puts his median at 2028, although it could be longer or sooner obvsly. Daniel Kokotajlo, one of the few people whose predictions have actually held up, now sees it in 2027. Back in 2021, more than a year before ChatGPT, he called chatbots going mainstream, scaling continuing to work, multimodal models, clumsy early agents that slowly got useful, and the US–China chip war. So pretty accurate guy I'd say. Now METR estimates AI is already accelerating Anthropic’s progress by 1.7x, way ahead of what even he expected. If this is right, the read-across is pretty clear. Spend keeps going like crazy, because whoever gets there first wins everything. Nobody will open-source a frontier model when what you have in-house, and how fast it’s improving, is that valuable. You can already see it at Anthropic: it explains why they want to go public and raise $90-100bn at almost any price. They don’t care about a lower valuation as long as they get the money, because they think they’re close and thus want to do the IPO ASAP. Compute, memory, power and fabs remain the bottleneck. Especially memory and fabs. The only real risk I see is political. Something bad, like a cyberattack on a big bank or a big global company, scares people off and forces a slowdown, or populism keeps rising and people push to pause or slow data-centre development, which would be ridiculously stupid. Anyways, recommend watching the video.
1
2
9
902
Un debate apasionante entre los cuatro principales partidos para ver quién presume más de defender la subida generalizada de las actuales pensiones conforme al IPC. @MrtnzAlvrz, el 30 de noviembre te puedes cambiar el nombre, porque habrá que seguir haciendo pedagogía.
123
530
62
3,308
115,895
BREAKING: Jensen Huang appeared “broken” in his final hours as Nvidia CEO, per Bloomberg. Nvidia’s board held an emergency meeting that ran late into the night after the latest affair reports, and by the end Huang had agreed to step down. The board left him no choice but to resign. According to people who spoke with him, his usual take-charge style was gone. He responded with indifference, saying things like “whatever” and “I don’t care,” and has since called himself a “piece of sh--.” He says he no longer cares what happens to him. His main concern is his wife and not “ruining the lives” of the women now linked to him. Inside Nvidia, employees are rallying behind him. More than 1,600 comments on internal forums call for his reinstatement, per Wired.
261
502
99
11,820
1,861,477
Dwarkesh is probably one of the most intelligent, out-of-the-box thinkers I follow, and he’s now more bullish on superintelligence than ever (until recently he thought it was a decade away). I agree with him. I fully recommend watching the explainer. Truly phenomenal. Would love an interview between @patrick_oshag and @dwarkesh_sp. As AI gets obviously better at everything you can check, from maths and coding to AI research itself, the timelines keep collapsing. It’s truly bizarre to look back at what these models could do in 2024, 2025 and 2026, and it keeps compounding as Dwarkesh puts it "from failing four-digit sums in 2023 to OpenAI claiming a Millennium Prize problem in 88 hours this autumn." Basically the core of the idea is the following: AI research is mostly recombining ideas and running experiments against a score, which is exactly what models already crush. If you automate that, then you get thousands of top-level researchers working at 10x speed, 24/7, squeezing years of progress into one. Dwarkesh puts his median at 2028, although it could be longer or sooner obvsly. Daniel Kokotajlo, one of the few people whose predictions have actually held up, now sees it in 2027. Back in 2021, more than a year before ChatGPT, he called chatbots going mainstream, scaling continuing to work, multimodal models, clumsy early agents that slowly got useful, and the US–China chip war. So pretty accurate guy I'd say. Now METR estimates AI is already accelerating Anthropic’s progress by 1.7x, way ahead of what even he expected. If this is right, the read-across is pretty clear. Spend keeps going like crazy, because whoever gets there first wins everything. Nobody will open-source a frontier model when what you have in-house, and how fast it’s improving, is that valuable. You can already see it at Anthropic: it explains why they want to go public and raise $90-100bn at almost any price. They don’t care about a lower valuation as long as they get the money, because they think they’re close and thus want to do the IPO ASAP. Compute, memory, power and fabs remain the bottleneck. Especially memory and fabs. The only real risk I see is political. Something bad, like a cyberattack on a big bank or a big global company, scares people off and forces a slowdown, or populism keeps rising and people push to pause or slow data-centre development, which would be ridiculously stupid. Anyways, recommend watching the video.
1
2
9
902
Pedro Sánchez asegura que no estuvo de vacaciones durante la invasión de Ceuta.
161
1,527
59
12,848
472,998
LBO_Guy retweeted
"Pero tienes un avion" NO. SE LE HA DESCOMPUESTO LA CARA.
96
956
38
16,893
1,361,447
"Es una comparacion en la que no salimos ganando en cuanto a peligrosidad" 💀
Quieres comparaciones? Pues toma comparaciones. Lloro con Jordi Wild.
2
914
Pedro Sánchez "NINGÚN miembro de mi gobierno ha insultado a NADIE" Óscar Puente
74
2,465
24
10,662
85,939
Ouch, Claude is out as well.
3
7
807
Replying to @FundamentEdge
Genuine question: there’s a (big) group of Gen X PMs who got very rich operating this model. They live in the spreadsheet, keep pressing the same checks for trend lines, talk to management twice a quarter to detect some slight change in tone or explanation of a driver, then run back to the model and tweak incremental margins or customer adds or whatever for the next two quarters. I feel like this form of investing is both ascendant and dying at the same time. I get that “it comes down to the numbers,” but increasingly I think getting the big annual P&L wins requires being “qualitatively precise”. By that I mean understanding the rate of change within an industry, anticipating inflections based on first principles (not today’s credit card data), and having a deeply informed view of how a business evolves through technological shifts, competitive shocks, and cycles. These approaches obviously aren’t mutually exclusive. But I rarely find the Gen X model-obsessed set equally obsessed with understanding how the product actually works, which architecture survives change, where value accrues, how a market absorbs incremental competition, etc. Maybe this matters disproportionately in TMT. But the guy who runs back to his model to update next quarter’s GM% based on an offhand management comment at a conference rarely seems to be the same guy who can tell you, with any real precision, what the business is going to look like in three years. And I wonder whether the second skill is actually the better way to get the next two quarters right too (or at least treat them as vol or margins of errors as noise toward a more inevitable outcome). Checks still matter. Listening for change matters. But knowing what the heck you own really matters and the credit card data or next quarters RPO within 100bps doesn’t tell you that.
30
40
9
898
147,930
LBO_Guy retweeted
Meet Gemini 4 Hydrogen. It brings cutting-edge intelligence to everyone, with speed and efficiency for everyday coding and agentic work. Across 4 thinking levels, it stands on its own or teams up with Carbon and Argon — 2.5X faster than 3.8 Flash, outpacing competing models in its class.
368
728
318
7,071
752,465
Hay que aprender a cerrar la bragueta chavales… $NVDA
*NVIDIA'S JENSEN HUANG OUT AS CEO
1
1
3
727
Really smart move, the board cannot get sued as Jensen resigned. Certainly not in anyone’s bingo card!
1
543