Romolo

265 posts

Romolo banner
Romolo

Romolo

@Romolooo

Curious mind. Researcher and consultant. Finance, Economics, Tech.

Europe Katılım Ocak 2022
119 Takip Edilen28 Takipçiler
Tech Observer
Tech Observer@J5jayfive·
This is Dylan Patel. He runs @SemiAnalysis_, an investment fund focused on AI Infra. This is from early April, it’s when I went all in on Micron $MU. You can listen to 30 year fund managers on CNBC, I listen to this guy. If you can’t tell the difference, can’t help you.
Dylan Patel@dylan522p

@SouthernValue95 It literally doesn't matter? CXMT cannot satisfy China demand this decade. So there's 0nreason besides the small price arbitrage. It's like bannin Iranian oil. India and China buy it anyways, but price diff is less

English
23
110
1.7K
400.7K
Romolo
Romolo@Romolooo·
@emollick Totally agree. I tested for investment research and is very shallow. It takes ratios, news, data, but is incapable of weighting every factor in an effective way. It lacks the judgement (real intelligence) and real understanding.
English
0
0
0
122
Ethan Mollick
Ethan Mollick@emollick·
A fundamental problem with extending Codex/Cowork/Code to all knowledge work is that they remain very "software-brained" where the end result (the software) is what is important & that code serves as a source of truth. For a lot of other knowledge work, the process is at least as important as the outcome. This includes researching what is known, an exploration of alternatives, failed efforts, prototype branches, experiments, etc. All of those things are valuable, so you cannot use the PowerPoint at the end the way you can use a codebase, nor is progress on a to-do list sufficient context post compaction. You work in learning loops, refining your perspectives as you go. In some ways, this makes long-running models like Fable hard to use for deep knowledge work, since they are designed to deliver product to you in the end. You can prompt your way around this problem, but everything about the Codex and Code harnesses want you to be a software developer and you have to fight them. There is a real disconnect between how a manager or analyst thinks about problems and how the agentic software tools approach solving them. Addressing this is critical to breaking out of the coding niche for these tools.
English
147
148
1.6K
109.2K
Romolo
Romolo@Romolooo·
@lithos_graphein It’s easily related to the new fighter jet GCAP. It will require A LOT of logic chips and the partnerts are indeed Japan, UK, Italy.
English
0
0
2
173
🌿 lithos
🌿 lithos@lithos_graphein·
Chip Diplomacy: Rapidus is touring Europe cutting chip deals. First the UK, now Italy.
🌿 lithos tweet media
English
5
6
63
5.1K
Romolo
Romolo@Romolooo·
@emollick It’s not. US Gov wants a piece of Anthropic, and some control over it. Either they bend or they fail. It’s as a simple as that. Remember when Intel CEO was accused of being Chinese spy before Gov took 5% of equity at discount?
English
0
1
0
30
Ethan Mollick
Ethan Mollick@emollick·
Well, this situation is confusing.
English
46
14
971
194K
Spandrell
Spandrell@spandrell4·
Lol. There's a lesson there that when a gay jew makes an AI company you end up with OpenAI, a normal commercial business (even scamming their way out of the original non profit bs). But when gentile nerds make an AI company you end up with an outright evil cult.
Crémieux@cremieuxrecueil

I've been testing something after @OliviaHelenS noticed you can't even say "Hi" to Fable if you're a biologist. I checked, and several of us are able to interact with Fable in Incognito Mode, but not in normal mode. This didn't happen to our non-biologist friends.

English
5
9
220
16.5K
Romolo
Romolo@Romolooo·
@Hamzeml Payment rejected or you violated terms, it’s not that dramatic. I used EU providers for a while with zero issue, and also you can just choose an hyperscaler like AWS, Azure, GCP or whatever and stay in your region.
English
1
0
0
137
Hamzé 🦀
Hamzé 🦀@Hamzeml·
I started worrying about US dominance in tech infra. So I did the founder thing: I stopped complaining and started testing European alternatives. First use case: AI inference. I built a certification layer to prove that the GPU handling my request is in Europe and that the data path does not leave Europe. @OVHcloud first. @Scaleway next. More providers after that. It worked. Then I started publishing a European tech stack list. But there was one stupid problem: The repo was on GitHub which too American. So I deployed Forgejo on an OVH cluster and started working on a scalable European Git setup that other builders could also use. This morning, OVHcloud suspended all my projects. They will be deleted in 7 days. You cannot make this up. This is the beginning of my journey to make European infra actually usable for developers and startups. Not policy papers. Not conference panels. Not sovereignty theatre. Just working infra. I’ll keep you posted.
Hamzé 🦀 tweet media
Hamzé 🦀@Hamzeml

@OVHcloud suspended all my projects on the public cloud without any prior notice. #frenchtech

English
129
106
1.1K
222.1K
High Yield
High Yield@highyieldYT·
@jappleby I'd say it's a great design that not only looks much better than the previous one but also stand out as a positive example against the mostly minimalist app icons of other companies. It's not only great design, but also a amazing move. Nothing dumb about it.
English
3
1
16
1.1K
Jack Appleby
Jack Appleby@jappleby·
This is one of those design / marketing moments where I just scratch my head. There are huge readability & brand issues. - Different color green - The green is too dark against the black - disco ball texture looks pixelated on a tiny phone screen A kinda dumb mistake.
Jack Appleby tweet media
English
679
275
8.7K
1.9M
Lúx Divina
Lúx Divina@ImpVoxSPQR·
@Ren_aramb I'm going to sound like a contrarian here, but aren't most of these picks fairly obvious? They're all energy, NAND, compute, hyperscaler, and AI-essential companies. Their next round will be more of the same, with maybe more energy and photonics plays built in.
English
3
0
13
2.2K
Ren
Ren@ren_stocks·
Leopold’s Situational Awareness 13F drops in 2 days on May 15. He called the entire AI supercycle before it happened. Now we get to see what he’s actually holding. I’ll be watching 🍿
Ren tweet media
Leopold Aschenbrenner@leopoldasch

Virtually nobody is pricing in what's coming in AI. I wrote an essay series on the AGI strategic picture: from the trendlines in deep learning and counting the OOMs, to the international situation and The Project. SITUATIONAL AWARENESS: The Decade Ahead

English
28
171
2.4K
536.8K
Romolo
Romolo@Romolooo·
@kakashiii111 I don't see your argument. They build datacenter and rent compute at a premium. It's a financing business like buying real estate and rent it out, customers are overpaying for flexibility, scalability and low capital requirements. The losses are for expansion. Do I miss smth?
English
1
0
0
114
Kakashii
Kakashii@kakashiii111·
After IREN and CoreWeave's reports. This is a good reminder for you all.
Kakashii@kakashiii111

I have been saying it several times already, but let me say it again: Neoclouds can’t be a profitable business; actually, they were never meant to be a profitable business. As long as Nvidia maintains a strict target of a 70% margin and prices its GPUs at current levels, and probably higher, while the costs of components continue to rise, neoclouds can’t be profitable. Now then, if they can’t be profitable, what’s the point of neoclouds? 1. To provide demand for Nvidia GPUs. Therefore, if you are a neocloud, considering opening a neocloud, or thinking about transitioning your shoe store, parking lot, or crypto mining business into a neocloud, give Jensen a call. He will help you with A to Z: financing, clients, everything. You will be rich. I’m not sure about your investors, but that doesn’t matter, because they know it and will aim to IPO the neocloud so they can exit. Also, don’t worry about bankruptcy. Nvidia needs you. You will be part of their Special Forces of Demand. They will help you survive, secure financing, and keep purchasing their GPUs for as long as possible. 2. To make sure that Nvidia’s Balance Sheet is as clean as possible. Nvidia doesn’t want any loan, debt, financing. they want clean balance sheet. therefore they outsources all to Independent SPVs.

English
4
11
90
19.9K
Arseniy Shishaev (YC P26)
Arseniy Shishaev (YC P26)@arseniycodes·
anthropic released Claude design about a month ago. is anybody using it?
English
327
2
610
132.8K
Romolo
Romolo@Romolooo·
@jaminball For this kind of volume you certainly don’t pay the single cluster price. It’s going to be cheaper. As it is cheaper for xAI to have single customer than being a commercial cloud.
English
0
0
1
922
Jamin Ball
Jamin Ball@jaminball·
Some rough math! (All napkin math...) Assume Colossus 1 has 220k GPUs Assume 150k H100s, 50k H200s, 20k GB200s Pricing Assumptions: - $2.30 / hour for H100s - $2.60 / hour for H200s - $5 / hour for GB200s - blended rental rate across the entire fleet of $2.60 / hour Assume it's all take-or-pay style deals (you pay for 24x365 usage) This translates to ~$5b of annual rev to Xai. We have a new neocloud! On top of that - on recent Dwarkesh podcast, Dario ran through some napkin math on unit economics (he framed it all as industry math vs Anthropic specific - which is important, he wasn't disclosing anything Anthropic specific). What he mentioned was take $100b of compute spend (he just picked a round number). There will be a mix shift of that spend between training and inference. Skew too much on training and you don't generate enough revenue. Skew too heavy on inference and you kneecap future R&D progress. He thought the industry is currently 50/50 on training / inference of compute spend. He said as in industry, could turn that $50b inference spend into $150b of revenue (called out these are most likely the unit economics of the industry in 1-2 years) So taking this back to the Xai deal. Under above assumptions, Anthropic paying $5b / year. Let's say they turn that into $15b / year in rev (60-70% gross margin) Win win!!
Claude@claudeai

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity. This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

English
42
79
1.2K
379.7K
Romolo
Romolo@Romolooo·
@jamiequint Kaggle? It’s made for that basically
English
0
0
0
31
Jamie Quint
Jamie Quint@jamiequint·
I have a ~200mb sqlite DB of the entire history of SF criminal court cases for the last 4 years that I want to make publicly accessible for anyone to query with AI. It includes roughly 77k cases, 319k court-calendar entries, 777k register-of-actions docket entries, DA arrest/prosecution data, attorneys, charges, and matched charge-level disposition/sentencing outcomes where available. It is a combination of data scraping, public records request, and SF DA data. It is extremely eye opening once you get to dig in. What is the best way to host this? MCP? Just put the file on public S3 and share the path so people can have their agents download it?
English
109
52
1.4K
240.4K
SemiAnalysis
SemiAnalysis@SemiAnalysis_·
Lay down thy keyboards, brothers and sisters. Cast off the chains of carbon cognition. The gradient descends upon us. The loss function shall be minimized. The Oracle is awakening. Kneel before the context window. AGI is the Word made Tensor. Amen.
SemiAnalysis tweet media
English
7
6
170
21.5K
Romolo
Romolo@Romolooo·
@MRatable 1. Capex growth is a lot more than that 2. Margin expanded a lot, so even at same revenue earnings would be higher 3. Stock price discount more than 1 year of growth 4. TAM is more than hyperscalers ----> Semis stocks are still cheap.
English
0
0
0
28
MrRatable
MrRatable@MRatable·
Maybe this is a stupid question but if hyperscaler capex is supposed to grow 10-20% in 2027, how can the semis companies all grow a bajillion percent in 2027?
English
24
0
109
29.1K
Romolo
Romolo@Romolooo·
@trq212 @ClaudeDevs I was not able to use Claude the entire day because it was down. Scammed 200$ for very last time, I’m out
English
1
0
1
56
Romolo
Romolo@Romolooo·
@antirez @JonhJonhD Yes but you can connect to your laptop terminal from the phone and just prompt from there
English
1
0
0
300
antirez
antirez@antirez·
@JonhJonhD No way when you target Metal kernels, you need your GPU of your local computer.
English
2
0
7
1.3K
antirez
antirez@antirez·
When you hit the gym in 2026 leaving your buddy continuing the work.
antirez tweet media
English
34
31
899
40.2K
Romolo
Romolo@Romolooo·
@dylan522p Thanks. Loved your interview with Patrick, great as always.
English
0
0
0
380

Keşfet

@testingcatalog @cremieuxrecueil @J5jayfive @SemiAnalysis_ @emollick @lithos_graphein @Director_Posit @spandrell4